oversight

Advertising Mail: Mail Mix Matters

Published by the United States Postal Service, Office of Inspector General on 2019-04-15.

Below is a raw (and likely hideous) rendition of the original report. (PDF)

TABLE OF CONTENTS     EXECUTIVE SUMMARY                        OIG SYNOPSIS      RCF REPORT   APPENDIX


                    Office of Inspector General | United States Postal Service
                    RARC Report
                    Advertising Mail:
                    Mail Mix Matters
                    Report Number RARC-WP-19-003 | April 15, 2019
TABLE OF CONTENTS                                   EXECUTIVE SUMMARY                                                    OIG SYNOPSIS                                              RCF REPORT   APPENDIX

                                                                                                                                                                                                BACK to COVER




Table of Contents

        Cover
        Executive Summary....................................................................................................................................... 1
           Results of Analysis...................................................................................................................................... 1
        OIG Synopsis..................................................................................................................................................... 3
           Introduction................................................................................................................................................... 3
                 What Influences How Customers Treat Advertising Mail?................................................. 4
           Findings........................................................................................................................................................... 4
                 Mail Mix Matters..................................................................................................................................... 4
                 Demographics Influence Advertising Mail Effectiveness.................................................... 5
                 Who Sends the Mail Influences Advertising Mail Effectiveness....................................... 7
                 Mailpiece Features Matter................................................................................................................ 7
           Business Implications................................................................................................................................ 9
           Conclusion..................................................................................................................................................... 10
        RCF Report ....................................................................................................................................................... 11
        Appendix: Management’s Comments................................................................................................... 48
        Contact Information...................................................................................................................................... 49




Advertising Mail: Mail Mix Matters                                                                                                                                                                         2
Report Number RARC-WP-19-003
TABLE OF CONTENTS                            EXECUTIVE SUMMARY                                   OIG SYNOPSIS                                    RCF REPORT                                      APPENDIX

                                                                                                                                                                                                   BACK to COVER




Executive Summary
Advertising mail is a significant source of revenue for the Postal Service.1 At
$20 billion in fiscal year (FY) 2017, advertising mail represented 29 percent of
total Postal Service revenue.2 It was approximately 58 percent of total mail.3 Its
importance to the Postal Service’s financial success cannot be overstated.
                                                                                                                    Highlights
                                                                                                                    Mail mix matters — the presence of a non-advertising mailpiece
Previous work by the U.S. Postal Service Office of Inspector General (OIG)
                                                                                                                    enhances the likelihood that a household will read, positively react
has shown that advertising mail has many inherent strengths — qualities like
                                                                                                                    to, and respond to a piece of advertising mail.
the ability for the sender to receive direct feedback on its effectiveness, higher
response rates than other direct marketing channels, memorability, and privacy                                      Increases in First-Class Mail and Periodicals as shares of total mail
— that make it still relevant in the digital age.4 In other words, advertisers use it                               received were both equally important in driving increases in reading,
because it works.                                                                                                   reaction to, and response to advertising mail.

Given how important this type of mail is to both the Postal Service and                                             Our analysis shows that other factors like demographics, existence
marketers, we analyzed what factors enhance the three “R’s” of advertising                                          of a past business relationship, and coupons had a positive effect in
mail effectiveness — reading, reaction and response.5 Chief among the                                               the reading, reaction, and response to advertising mail.
factors studied was how various types of mail (in other words, the “mail mix”) a
household receives interact with one another. Simply put, does the presence of
non-advertising mailpieces enhance the value of advertising mail? We studied
additional factors that drive advertising mail effectiveness. These include                                 Results of Analysis
demographic factors, past business relationship, presence of a coupon, and                                  ■■ Mail mix matters. A higher non-advertising share, including both First-
the shape of the advertising mailpiece. Specifically, we used three separate                                   Class Mail and Periodicals, is associated with an increased likelihood that
regression models to explain the effect of these factors on the probability that                               the household will read, have a positive reaction to, and respond to the
1) a piece of advertising mail was read; 2) the piece generated a positive reaction                            advertising mail they receive.
(for example the recipient found it useful or interesting); and 3) the household is
considering responding to the mailpiece.6                                                                   ■■ We found that mail mix matters even when the other factors influencing the
                                                                                                               likelihood of the three Rs are controlled for in our analysis. In other words,
                                                                                                               a household that is similar to other households in terms of age, education,


 1	 Total advertising mail is composed of Marketing Mail and First-Class advertising mail.
 2	 FY 2018 Household Diary Study data were not available when this report was written. OIG calculation based on the Postal Service’s Cost, Revenue and Analysis Reports (CRA). U.S. Postal Service, Cost and Revenue
    Analysis Report FY 2017, https://about.usps.com/who-we-are/financials/cost-revenue-analysis-reports/fy2017.pdf and U.S. Postal Service, The Household Diary Study: Mail Use & Attitudes in FY 2017, March 2018,
    https://www.prc.gov/docs/105/105134/USPS_HDS_FY17_Final%20Annual%20Report.pdf, p. 2.
 3	 Ibid.
 4	 U.S. Postal Service Office of Inspector General, Advertising Mail: Past and Present, Report No. RARC-WP-16-006, March 28, 2016, https://www.uspsoig.gov/sites/default/files/document-library-files/2016/RARC-
    WP-16-006_0.pdf, pp. 12-13.
 5	 Marketing Mail comprises the bulk of the Postal Service’s advertising mail volume. For example, in 2017, Marketing Mail constituted approximately 90 percent of advertising volume. As such, we use the terms
    advertising mail and Marketing Mail interchangeably. OIG calculation based on U.S. Postal Service FY 2017 CRA and U.S. Postal Service FY 2017 HDS.
 6	 For this analysis, the OIG worked with RCF Economic and Financial Consulting, a firm with deep expertise in postal economics.

Advertising Mail: Mail Mix Matters                                                                                                                                                                               1
Report Number RARC-WP-19-003
TABLE OF CONTENTS                    EXECUTIVE SUMMARY                       OIG SYNOPSIS                          RCF REPORT                             APPENDIX

                                                                                                                                                           BACK to COVER




   technology, or race will be more likely to read advertising mail when there is a   ■■ Households headed by a college graduate are less likely to read, have
   greater share of non-advertising mail found in the mailbox.                           a positive reaction to, and respond to a piece of advertising mail than
                                                                                         households with only a high school degree.
■■ Within First-Class Mail, we found that transactions mail (for example bills and
   statements) was a more important driver of household reading, reaction, and        ■■ Existence of a past business relationship between the mailer and the
   response to advertising mail than was correspondence mail (for example                household has a strong impact on household reading of advertising mail, a
   personal letters).                                                                    stronger impact on household reaction, and an even stronger impact on the
                                                                                         likelihood of response.
■■ Demographic factors play influential roles in the household treatment of
   advertising mail. Consistent with previous OIG studies, we found that older        ■■ The presence of a coupon is found to significantly raise the reading, positive
   people are more receptive to the mail than younger people. However, younger           reaction, and response rates, with the strongest impact on the response rate.
   people are not entirely uninterested in mail and, therefore, are potential new
                                                                                      ■■ The shape of the mailpiece has a significant impact on household reading,
   customers for marketers.
                                                                                         reaction, and response. Flats, for example, are significantly more likely than
                                                                                         letters to be read, create a positive reaction, and generate a response.




Advertising Mail: Mail Mix Matters                                                                                                                                    2
Report Number RARC-WP-19-003
TABLE OF CONTENTS                               EXECUTIVE SUMMARY                                       OIG SYNOPSIS                                       RCF REPORT                                          APPENDIX

                                                                                                                                                                                                                BACK to COVER



OIG Synopsis
Introduction                                                                                                        Figure 1: Total Postal Service Volume in 2017

In a departure from the past, most of the recent volume growth in postal products
has been in the Shipping and Packages business.7 However, this does not mean                                            TODAY, MAIL CONTINUES TO BE A VIABLE AND
that letter and flat mail are relics of the past. Indeed, this type of mail continues to                                PROFITABLE CHANNEL OF COMMUNICATION
be a viable channel of communication throughout the United States. In 2017,                                             THROUGHOUT THE UNITED STATES
95 percent of the Postal Service’s total volume was letter and flat mail (First-Class                                   The Postal Service delivered 149 billion pieces of mail in 2017. Fifty-eight
Mail, Marketing Mail, and Periodicals Mail).8 As shown in Figure 1, advertising                                         percent of total volume is advertising mail, making this mail segment vital to
mail — defined as the total of Marketing Mail and advertising mail in First-Class                                       Postal Service’s current financial survival and future success.
— is still a $20 billion business, accounting for 58 percent of the Postal Service’s
                                                                                                                                                     Shipping and              International
volume in FY 2017.9                                                                                                                                   Packages                      1%
                                                                                                                                                         4%
                                                  Given the significance of advertising mail,                                        Periodicals
                                                                                                                                         4%
                                                  what factors influence the “three R’s” of
	 Letters and flats are                           advertising mail effectiveness — reading,
                                                  reaction, and response? We posed this                                                                                             First-Class
  extremely important                             question to RCF Economic and Financial                                                                                         Correspondence
                                                                                                                                                                                  + Transactions
  elements of the                                 Consulting (RCF), experts in postal                                                                                                   33%
                                                  economics.10 Using regression analysis,
  Postal Service’s mail                           RCF analyzed data from the Postal Service’s
                                                                                                                                                         Total Advertising
                                                                                                                                                                58%
  mix. In 2017, letters                           annual Household Diary Study (HDS) from
                                                  2013 to 2017 to determine how mail mix
  and flats comprised                             and other factors affect the probability that
  95 percent of the                               1) a piece of advertising mail was read by
                                                  someone in the household; 2) the advertising
  Postal Service’s total                          mailpiece generated a positive reaction (for
                                                                                                                        Sources: U.S. Postal Service 2017 Report on Form 10-K and The Household Diary Study Mail Use
                                                                                                                        and Attitudes in FY 2017. FY 2018 HDS data were not available when this report was written.
  mail volume.                                    example, it was found useful or interesting);                         Note: First-Class Mail advertising is excluded from First-Class Mail volume. Total advertising is
                                                                                                                        comprised of Marketing Mail and advertising in First Class Mail.
                                                  and 3) the household is considering
                                                  responding to the mailpiece.11
 7	 During the 11-year period from 2007 to 2017, First-Class Mail volume fell 38 percent, Marketing Mail 24 percent, and Periodicals 40 percent. Shipping and Packages volume increased 181 percent. OIG calculation
     based on FY 2007 and FY 2017 CRA reports.
 8	 OIG calculation based on FY 2017 CRA and FY 2017 HDS. The FY 2018 HDS data were not available when this report was written.
 9	 OIG calculation based on U.S. Postal Service, Form 10-K FY 2017, November 2017, https://about.usps.com/who-we-are/financials/10k-reports/fy2017.pdf, p.19 and FY 2017 HDS. First-Class Mail and Marketing Mail
     total volume exclude parcels. Shipping and Packages total volume include Priority Mail, Priority Mail Express, USPS Retail Ground, Parcel Select Mail, Parcel Return Service Mail, Marketing Mail Parcels, Package
     Service Mail, First-Class Mail Parcels, First-Class Package Service, and Priority Mail Express. It excludes International packages.
 10	 RCF Economic and Financial Consulting, http://www.rcfecon.com/.
 11	 This analysis cannot predict when an advertising mail piece will actually be read. The purpose of this model is to identify certain characteristics that may influence the probability that a household will read, positively
     react, or respond to a piece of advertising mail. In addition, it should be noted that the HDS reports on intended rather than actual responses of the survey participants. For the technical analysis, see RCF’s report.

Advertising Mail: Mail Mix Matters                                                                                                                                                                                             3
Report Number RARC-WP-19-003
TABLE OF CONTENTS                                EXECUTIVE SUMMARY                                        OIG SYNOPSIS                                         RCF REPORT                                           APPENDIX

                                                                                                                                                                                                                     BACK to COVER




What Influences How Customers Treat Advertising Mail?                                                                ■■ Demographic factors are relevant in explaining the likelihood that a piece of
                                                                                                                        advertising mail will be read, generate a positive reaction, and lead to a response.
A number of factors will affect the effectiveness of advertising mail. We looked
at two main issues in our analysis. The first is whether the mix of mail — defined                                   ■■ Existence of a past business relationship with the household enhances the
as the share of a household’s weekly mail that is non-advertising — affects                                             effectiveness of an advertising mailpiece.
household reading, reaction, and response to advertising mail. For the purposes
of this analysis, First-Class correspondence, First-Class transactions, and                                          ■■ The presence of a coupon was found to be the most important, significantly
Periodicals Mail are considered non-advertising mail. Packages are not included                                         raising the reading, positive reaction, and response rates, with the strongest
in the household mail mix because they are sometimes received separately from                                           impact on the response rate.
other mail, at a different time of day, or even via a different carrier. As a result,                                ■■ Shape matters when it comes to an advertising mailpiece’s effectiveness.
their impact on the advertising mail reading, reaction, and response rate is likely                                     Flats — items like large envelopes, newsletters, and magazines — for
to operate in a different manner from that of letters and flats. Therefore, their                                       example, are significantly more likely to be read, create a positive reaction,
effect is beyond the scope of this paper.                                                                               and generate a more likely response than letters.
Next, to isolate the effect of mail mix only, we controlled for other influencing                                    Mail Mix Matters
factors that affect reading, reaction, and response.12 These other factors
are: 1) household demographics; 2) mailer characteristics; and 3) features
                                                                                                                     The presence of non-advertising mailpieces                       	 The presence of
                                                                                                                     enhances the effectiveness of advertising
of the individual advertising mailpiece. We also look at whether the different
                                                                                                                     mail, increasing the probability that
                                                                                                                                                                                        First-Class Mail
components of non-advertising mail (First-Class correspondence mail, First-Class
transactions mail, and Periodicals Mail) have separate impacts on household
                                                                                                                     households will read, favorably react, and                         increases the value
                                                                                                                     respond to an advertising mailpiece.
treatment of advertising mail.13                                                                                                                                                        of advertising mail
                                                                                                                     In addition, the impact of mail mix on
Findings                                                                                                                                                                                in a measurable way,
                                                                                                                     advertising mail varies from component
Several factors were found to strengthen                            	 The existence of                               to component of non-advertising mail as                            as highlighted by the
the value of advertising mail. We found the                                                                          shown in Table 1.14 We found that increases
following:                                                            non-advertising                                                                                                   results in our study.
                                                                                                                     in both First-Class Mail and Periodicals Mail
■■ Variation in the mail mix enhances the                             mail increases the                             increase the effectiveness of advertising
                                                                                                                     mail.15 Surprisingly, within First-Class Mail,
   likelihood that the household will read,                           effectiveness of
                                                                                                                     transactions mail (for example bills and statements) is a more important driver of
   positively react to, and respond to an
   advertising mailpiece.
                                                                      advertising mail.                              household reading and reaction to advertising mail than is correspondence mail
                                                                                                                     (for example personal letters and business correspondence).16

 12	 By doing this, we know that when we observe the impact of mail mix on, say, the reading rate of advertising mail, we know we are not mixing in the impact of some other factor, like education or age, that might be
     correlated with the mail mix.
 13	 Table 7 in the RCF Report describes the variables used in the models.
 14	 A positive sign indicates the factor has a statistically positive response. For example, we found an increase in transactions mail has a positive impact on the probability the household will read, react, and respond to
     advertising mail. A negative sign shows the opposite — the presence of the factor has a statistically negative impact. For example, if the advertising piece is from a financial sender (as will be discussed later), it is less
     likely to be read, reacted to, or responded to. No impact means that the result was not statistically significant. For the detailed mail mix results of each regression model, see Tables 8-10 and 29 in the RCF Report.
 15	 See Table 28 in the RCF Report for the results on the separate impact of First-Class Mail and Periodicals on advertising mail.
 16	 See Table 29 in the RCF Report for the results on the separate impact of First-Class correspondence and transactions mail.

Advertising Mail: Mail Mix Matters                                                                                                                                                                                                  4
Report Number RARC-WP-19-003
TABLE OF CONTENTS                               EXECUTIVE SUMMARY                                        OIG SYNOPSIS                                       RCF REPORT                                           APPENDIX

                                                                                                                                                                                                                  BACK to COVER


Table 1: A Healthy Mail Mix Positively Contributes Toward Advertising Mail                                           Table 2: Demographics Matter
Effectiveness

                                                                                                                         Demographic factors help in explaining the likelihood that a piece of
    We found that a favorable mail mix enhances the likelihood that the household                                        advertising mail will be read, generate a positive reaction, and response.
    will read, positively react to, and respond to an advertising mailpiece.
                                                                                                                                                                            Reading            Reaction          Response
                                                     Reading             Reaction            Response
                                                                                                                                              25-34                               -                  -                 -
                        Correspondence                    +             No impact            No impact
      First-Class                                                                                                                             35-44                        No impact          No impact          No impact
      Mail
                        Transactions                      +                   +                    +                                          45-54                        No impact                +            No impact
                                                                                                                           Age
                                                                                                                           Base: 18-24        55-64                              +                  +                  +
      Periodicals                                         +             No impact            No impact

                                                                                                                                              65-74                              +                  +                  +
    Source: RCF analysis of U.S. Postal Service Household Diary Study, Mail Use & Attitudes.

                                                                                                                                              75+                                +                  +                  +

These findings demonstrate the importance of defending all segments of mail.                                               Education          Some College or
                                                                                                                                                                                  -                  -                 -
From the perspective of the Postal Service, these results suggest that efforts to                                          Base: High         Tech School
increase the volume of advertising mail will work better if they are accompanied                                           School or          At Least College
by efforts to increase (or slow the decrease in) the volume of First-Class and                                             less                                                   -                  -                 -
                                                                                                                                              Graduate
Periodicals Mail. This combined effort will help maintain a more favorable mail mix
and strengthen the value of advertising mail, thereby further encouraging its use.                                                            HH receives bills
                                                                                                                           Technology
As such, the Postal Service should continue to defend First-Class Mail volume.                                             Use
                                                                                                                                              and statements                      -                  -                 -
But we acknowledge that this is challenging, as previous OIG work has shown.17                                                                online

                                                                                                                                              Hispanic                           +                  +                  +
Demographics Influence Advertising Mail Effectiveness
We analyzed how demographic factors influence advertising mail effectiveness.                                              Race Base:         African American                   +                  +                  +
These results are presented in Table 2.18                                                                                  Caucasian          Asian                              +                  +            No impact

                                                                                                                                              Other (non-white)                  +                  +                  +

                                                                                                                         Source: RCF Analysis of U.S. Postal Service Household Diary Study, Mail Use & Attitudes.
                                                                                                                         Note: Where noted the results are the relative impact as compared to a base group.




 17	 U.S. Postal Service Office of Inspector General, A New Reality: Correspondence Mail in the Digital Age, Report No. RARC-WP-18-004, March 5, 2018, https://www.uspsoig.gov/sites/default/files/document-library-
     files/2018/RARC-WP-18-004.pdf and U.S. Postal Service Office of Inspector General, Transactional Mail: Implications for the Postal Service, Report No. RARC-WP-18-007, April 16, 2018, https://www.uspsoig.gov/
     sites/default/files/document-library-files/2018/RARC-WP-18-007.pdf.
 18	 The positive (+) and negative (-) signs in Table 2 are relative measures in that they are relative to the omitted category. For example, for Age, the impact is relative to the 18-24 category; for Education, the impact is
     relative to a high school degree or less; for Race, the impact is relative to white households. For the detailed results on how demographic factors influence advertising mail effectiveness, see RCF Report.

Advertising Mail: Mail Mix Matters                                                                                                                                                                                                  5
Report Number RARC-WP-19-003
TABLE OF CONTENTS                              EXECUTIVE SUMMARY                                      OIG SYNOPSIS                                       RCF REPORT                                         APPENDIX

                                                                                                                                                                                                             BACK to COVER




Age of Household Head                                                                                             Ethnicity and Race
Households headed by someone who is older are more likely to read advertising                                     The ethnicity of the household appears to influence the effectiveness of
mail, with the likelihood of a piece of advertising mail being read increasing as                                 advertising mail. As can be seen in Table 2, the non-Caucasian households are
the age of the household increases, as shown in Table 2.19 Age also affects                                       more likely to read and react to advertising mail than Caucasian households, and
household reaction and response with younger households less likely and older                                     all but Asian-households are more likely to respond.
households more likely to have a positive reaction and respond to advertising
                                                                                                                  Of all the ethnic groups in the HDS, Hispanic households are more likely to
mail.20 This finding is consistent with previous OIG findings that older people tend
                                                                                                                  read, positively react, and respond to advertising mail than are non-Hispanic
to rely on mail more than younger people.21
                                                                                                                  households.24 It is meaningful that Hispanic and non-white households are more
Education of Household Head                                                                                       receptive to advertising mail because these households receive less advertising
                                                                                                                  mail than white households, as shown in Figure 2.25 Our analysis indicates that
In contrast to age, education has a negative impact on the effectiveness of
                                                                                                                  non-white households may be an untapped market for direct mail marketers.
advertising mail. Households headed by a college graduate, for example, are
significantly less likely to read, respond or react to a piece of advertising mail than                                       Figure
                                                                                                                  Figure 2: Pieces      2: Pieces
                                                                                                                                   of Advertising       of Advertising
                                                                                                                                                  Mail Received per Week
households with only a high school degree or less (see Table 2).22 For marketers,
these negative impacts need to be balanced against the advantages of targeting
                                                                                                                                           Mail Received per Week
                                                                                                                                13.8
higher education households that are also likely to have higher incomes and
                                                                                                                                                         11.7
purchase more goods and services than less educated households.                                                                                                                   10.5
                                                                                                                                                                                                            9.4
Technology Use – Online Bill Presentment

Households that receive online bills or statements are less likely to read,
positively react, or respond to advertising mail as shown in Table 2.23 All of these
results are consistent with the idea that these households are less connected to
their mail than households that rely exclusively on the mail for the receipt of their
bills and statements.
                                                                                                                       Caucasian/white Asian-American                          Hispanic           African-American

                                                                                                                      Sources: RCF Analysis of HDS Data.
                                                                                                                       Source: RCF Analysis of HDS Data



 19	 See Tables 8-10, 14 in the RCF Report for the detailed technical results.
 20	 See Tables 8-10, 14 in the RCF Report for the detailed technical results. All these coefficients are measured relative to the youngest age group (18 – 24) and are consistent with view that older people are generally
     more receptive to the mail than younger people.
 21	 See for example, U.S. Postal Service Office of Inspector General, A New Reality: Correspondence Mail in the Digital Age and U.S. Postal Service Office of Inspector General, Transactional Mail: Implications for the
     Postal Service.
 22	 See Tables 8-10, 16-17 in the RCF Report for the detailed technical results.
 23	 See Tables 8-10, 20-21 in the the RCF Report for the results.
 24	 See Tables 8-10, 18-19 in the RCF Report for the detailed technical results.
 25	 Although non-white and Hispanic households may have a different mail mix than white non-Hispanic households, the impact of any difference in mail mixes is already accounted for within the regression equation.

Advertising Mail: Mail Mix Matters                                                                                                                                                                                             6
Report Number RARC-WP-19-003
TABLE OF CONTENTS                                  EXECUTIVE SUMMARY                                         OIG SYNOPSIS                                        RCF REPORT                           APPENDIX

                                                                                                                                                                                                      BACK to COVER




Who Sends the Mail Influences Advertising Mail Effectiveness                                                             Yet, it is important to recognize that companies must send mail to households
                                                                                                                         with which they have not had yet a past relationship to generate new customers.
We also looked at what impact sender characteristics — things like past business
                                                                                                                         While this “prospecting” mail is far less likely to be read, it can be an important
relationship, nonprofit mailings, and financial sender mailings — had on the
                                                                                                                         first step to creating new customers and future business relationships.
effectiveness of advertising mail. The findings are summarized in Table 3.26

Table 3: Mailing Type Influences Reading, Reaction, and Response of
                                                                                                                         Nonprofit Postage
Advertising Mail                                                                                                         Households are also significantly more likely to read, react positively, and respond
                                                                                                                         to mail sent using nonprofit postage as seen in Table 3.29 These pieces are used
       Mailing type influences how customers perceive advertising mail. For example, our                                 by nonprofit and other social agencies that qualify for the reduced postage rate.
       analysis shows that when there is a past business relationship, households are far                                The strongest effect is on reaction suggesting that even when households do not
       more likely to read, react positively, and respond to an advertising mail piece.                                  respond to nonprofit mailings (most of which are requests for donations) they still
                                                                                                                         view these mailings positively.30
                                                   Reading              Reaction             Response
                                                                                                                         Financial Industry Sender
        Past Business Relationship                       +                   +                     +                     In contrast to mail from nonprofit senders, households are significantly less likely
                                                                                                                         to read, react positively, and respond to mail sent by the financial sector as shown
        Nonprofit                                        +                   +                     +                     in Table 3.31 Many of these mailings are solicitations from credit card companies.
                                                                                                                         Perhaps, one reason for this result is that the decision to get an additional credit
        Financial Sender                                 -                    -                    -
                                                                                                                         card is a bigger decision than one to use a coupon.
       Source: RCF analysis of U.S. Postal Service Household Diary Study, Mail Use & Attitudes.
                                                                                                                         Mailpiece Features Matter
                                                                                                                         Mailpiece features, such as shape, whether it is addressed to recipient (i.e.
Past Business Relationship                                                                                               targeted mailpiece), presence of a coupon, and presence of a return envelope
                                                                                                                         were also analyzed to see whether they had any impact on the effectiveness of
Existence of a past business relationship has a strong impact on the effectiveness                                       advertising mail. The findings are summarized in Table 4.32
of advertising mail as shown in Table 3.27 In fact, advertising mail sent by a
business that has a past business relationship with the recipient is more than                                           Our analysis finds that the shape of the advertising mailpiece does impact its
twice as likely to be read and generate a positive reaction and is six times as                                          effectiveness. We looked at seven different mailpiece shapes: letter, flat, catalog,
likely to get a response than mail sent by businesses with no past relationship.28                                       detached label card (DAL), postcard, flyer, and newsletter.33 The results are
                                                                                                                         shown in both Table 4 and in Figure 3 below.
 26	   See Tables 8-10 in the RCF Report for the results.
 27	   See Tables 8-10, and 22 in the RCF Report for the results.
 28	   See Table 23 in the RCF Report for the results.
 29	   See Tables 8-10 and 22-23 in the RCF Report.
 30	   Ibid.
 31	   Ibid.
 32	   See Tables 8-10 and 24-25 in the RCF Report.
 33	   Within the regression the omitted shape category is “letter,” so the coefficients reflect differences in the reading, reaction, and response rates of non-letter pieces relative to letters.

Advertising Mail: Mail Mix Matters                                                                                                                                                                               7
Report Number RARC-WP-19-003
TABLE OF CONTENTS                             EXECUTIVE SUMMARY                                    OIG SYNOPSIS                                    RCF REPORT                                        APPENDIX

                                                                                                                                                                                                      BACK to COVER



Table 4: Several Mailpiece Features Impact Consumers’ Perception of                                           Figure 3: Shape of the Advertising Mail Piece Impacts Its Effectiveness
Advertising Mail


    Other factors impact advertising mail effectiveness. Specifically, a flat-
    shaped piece is more likely to be read, create a positive reaction, and
    generate a response than a letter. Similarly, the presence of a coupon
    positively influences the three Rs of advertising mail.

                                                       Reading         Reaction        Response

      Specifically Addressed
                                                            +                +               +
      to HH member

      Has Coupon                                            +                +               +

      Return Envelope
                                  No Postage                 -               +               +
                                                                                                                  Sources: RCF Analysis of HDS Data.
      Base: Response ‘No’
                                  Postage Paid               -               -               +

                                  Flat                      +                +               +                We found that a flat-shaped mailpiece — an item like a large envelope,
                                                                                                              newsletter, and magazine — is significantly more likely to be read, create a
                                  Catalog                    -               +                -
                                                                                                              positive reaction, and generate a likely response than letters.34 It follows then, that
                                  Detached                                                                    a household that gets more flats will have a higher reading rate (all else being
                                                             -               -                -
                                  Label Card                                                                  equal) because household members are more likely to read those flats.
      Shape Base: Letters
                                  Postcard                  +                +               +                Catalogs have an interesting relationship with households. The results suggest
                                                                                                              that they are no more likely to be read or responded to, but they create a
                                  Flyer                     +                +               +
                                                                                                              significantly strong positive reaction as Figure 3 shows.35 One element of catalogs
                                  Newspaper/                                                                  that needs to be noted here is that they are not likely to be read immediately;
                                  Newsletter/               +                +                -               instead, they are often set aside for later reading. As measured in our study,
                                  Magazine                                                                    “read” measures the mailpiece being read as soon as it is received. However,
                                                                                                              “set aside for later” is one of the responses households can give to the reading
    Source: RCF analysis of U.S. Postal Service Household Diary Study, Mail Use & Attitudes.
    Note: Where noted the results are the relative impact as compared to a base group.
                                                                                                              question and catalogs have a high “set aside” rate.36 Thus, our analysis suggests
                                                                                                              that people enjoy receiving catalogs even if they do not immediately read them.

 34	 The Postal Service uses the word “flats” to refer to large envelopes, newsletters, and magazines. Flats must: have one dimension that is greater than 6-1/8 inches high OR 11-½ inches long OR ¼ inch thick.
     Additionally, they may be no more than 12 inches high x 15 inches long x ¾ inch thick. U.S. Postal Service, “Sizes for Large Envelopes and Flats,” https://pe.usps.com/businessmail101?ViewName=Flats. See Table 24
     in the RCF Report for the results.
 35	 See Table 24 in the RCF Report for the results.
 36	 RCF analysis of HDS data.

Advertising Mail: Mail Mix Matters                                                                                                                                                                                  8
Report Number RARC-WP-19-003
TABLE OF CONTENTS                               EXECUTIVE SUMMARY                                     OIG SYNOPSIS                                     RCF REPORT                                        APPENDIX

                                                                                                                                                                                                          BACK to COVER




Similar to flats, postcards are more effective advertising mailpieces than                                       An important implication
letters.37 The same holds true for flyers, though the impact is not as strong as for                             suggested by these results is
postcards.38 Newsletters are more likely to be read, and like catalogs are far more                              that the continued declines in     	 The decline in First-Class
likely to generate a positive reaction.39                                                                        First-Class Mail and Periodicals      Mail may have a negative
                                                                                                                 Mail volumes could have a
The impact of four other mailpiece                                                                                                                     impact on the effectiveness of
                                                                                                                 negative spill-over effect on
features are examined: (1) whether 	 Our study shows that                                                        advertising mail. As these            advertising mail. This negative
the mailpiece was specifically
addressed to a household
                                        coupons have a notable                                                   non-advertising portions of the
                                                                                                                 mail decline, households would        impact has the potential to
member; (2) whether it contained        positive impact by                                                       likely respond by reducing the
a coupon; (3) whether it came with                                                                                                                     cause mailers to look for
a return envelope without paid
                                        increasing the likelihood                                                reading of advertising mail and
                                                                                                                 lowering their overall perception     alternatives to advertising.
postage; and (4) whether it came        that a piece of advertising                                              of this mail. This in turn could
with a return envelope with paid
postage.40 Of these, the presence
                                        mail will be read, create                                                reduce the effectiveness of
                                                                                                                 advertising mail as an advertising medium and reduce the volume of direct mail
of a coupon is found to be the most     a positive reaction, and                                                 sent. On the other hand, policies that slow the decline in First-Class Mail and
important, significantly raising the
reading, positive reaction, and
                                        generate a response.                                                     Periodicals Mail would have a secondary positive impact of maintaining a more
                                                                                                                 favorable mail mix and stabilizing (or even increasing) the effectiveness and use
response rates, with the strongest                                                                               of advertising mail.
impact on the response rate.41 A previous OIG study found that Millennials are
enthusiastic coupon clippers and they strongly appreciate receiving coupons in                                   This paper finds that certain mail characteristics are important in driving
the mail.42                                                                                                      reading, reaction, and response to advertising mail. The Postal Service could
                                                                                                                 play a significant role by working with advertisers to ensure direct mail has
Business Implications                                                                                            characteristics that increase its effectiveness. For example, the Postal Service
                                                                                                                 could consider offering temporary incentives to mailers that use a flat mail piece,
The effectiveness of advertising mail increases if the Postal Service can increase,
                                                                                                                 such as a newsletter, or postcard. The Postal Service could furthermore work with
or even maintain, the share of non-advertising mail — both First-Class Mail and
                                                                                                                 advertisers to incentivize the use of coupons, especially when the advertising
Periodicals Mail. Within First-Class Mail, transactional mail has a more positive
                                                                                                                 mailpiece is targeting Millennials. Additionally, as our study has shown, minority
effect on advertising mail than correspondence. This important finding suggests
                                                                                                                 households may be an untapped market. The Postal Service could collaborate
that the Postal Service has even more reason to fight for all segments of the mail:
                                                                                                                 with advertisers to study how to better target these households to ensure they
First-Class Mail, Marketing Mail, and Periodicals Mail.
                                                                                                                 receive content that is relevant to them and therefore increase advertising mail
                                                                                                                 effectiveness.

 37	   See Table 24 in the RCF Report for the results.
 38	   Ibid.
 39	   Ibid.
 40	   Table 7 in the the RCF Report discusses variables used in the regression models.
 41	   See Tables 8-10 and 26-27 in the RCF Report.
 42	   U.S. Postal Service Office of Inspector General, Millennials and the Mail, Report No. RARC-WP-18-011, July 30, 2018, https://www.uspsoig.gov/sites/default/files/document-library-files/2018/RARC-WP-18-011.pdf, p. 7.

Advertising Mail: Mail Mix Matters                                                                                                                                                                                      9
Report Number RARC-WP-19-003
TABLE OF CONTENTS                    EXECUTIVE SUMMARY                          OIG SYNOPSIS   RCF REPORT   APPENDIX

                                                                                                            BACK to COVER




Conclusion
With continued declines in First-Class correspondence and transactions mail
and Periodicals Mail, the Postal Service not only risks a loss of revenue from
these types of communications, but also risks a loss of revenue if advertising
mail becomes less effective. As such, what this analysis shows is that the
Postal Service’s ongoing efforts to maintain other mail, especially First-Class
Transactions mail, benefits the Postal Service in two ways. The first is through
the revenue from the mail itself. The second is through the benefit of increasing
the effectiveness of advertising mail. In other words, the benefit of what is in the
mailbox is greater than the value of its independent parts.




Advertising Mail: Mail Mix Matters                                                                                     10
Report Number RARC-WP-19-003
TABLE OF CONTENTS                    EXECUTIVE SUMMARY         OIG SYNOPSIS   RCF REPORT   APPENDIX

                                                                                           BACK to COVER



RCF Report



                                       Prepared by
             RCF Economic & Financial Consulting, Inc.
                         333 N Michigan Avenue, Suite 2000
                              Chicago, Illinois 60601
                                www.rcfecon.com



                                           for

                    The Office of Inspector General (OIG) of
                       The United States Postal Service




Advertising Mail: Mail Mix Matters                                                                    11
Report Number RARC-WP-19-003
TABLE OF CONTENTS                             EXECUTIVE SUMMARY                                     OIG SYNOPSIS            RCF REPORT   APPENDIX

                                                                                                                                         BACK to COVER




The Impact of Mail Mix on Household Reading, Reaction and
Response to Marketing Mail
I.	Introduction

A 2016 study of Swiss households “Mail Composition and Recipients Reaction to Direct
Mail”43 found that there was a relationship between the mix of mail received by a household
and the household’s reaction to advertising mail. The authors studied 544 recipients during
a single week in March 2016 and found that a higher share of non-advertising mail was
associated with more positive reactions to advertising mail. The purpose of this study
is to test whether that relationship also holds true in the U.S. We use Household Diary
Study (HDS) data from 2013 to 2017 to analyze the impact of the mix of mail received
by households on 1) the probability that a given piece of Marketing Mail will be read by
someone in the household; 2) the probability that the household will have a positive reaction
to the mail piece; and 3) the probability that the household indicates that they are likely to
respond to the advertising.

The HDS is an annual survey conducted by the U.S. Postal Service, administered by
NuStats of Austin, Texas. RCF has years of experience analyzing this data which has been
collected annually since 1987. Each year, approximately 5,000 households complete a
weekly diary of mail received, with about 100 households completing the diary each week
of the year. From 2013 to 2017, the HDS recorded nearly 350,000 pieces of Marketing Mail
received by over 25,000 households, creating a sample size for this study far larger than the
Swiss study.

Households report information about each piece of mail received during the week thereby
providing information on their weekly mail mix. Households also report detailed information
about each mail piece including such things as the content and shape of the mail piece and
the industry of the sender. For Marketing Mail, households report whether the piece was
read, whether the piece generated a positive reaction, and whether they are likely to respond
to the advertising. In a companion survey – known as the Recruitment Survey – households
provide demographic information such as age and education as well as information about
their use of various technological alternatives to the mail such as online bill presentment.


 43	 “Mail Composition and Recipients’ Reaction to Direct Mail,” T. Geissmann*, C. Jaag, U. Trinkner and M. Maegli. 2017.
     Mail Composition and Recipients’ Reaction to Direct Mail. Topics in Regulatory Economics and Policy: The Changing
     Postal and Delivery Sector. Cham: Springer, 271-282

Advertising Mail: Mail Mix Matters                                                                                                                  12
Report Number RARC-WP-19-003
TABLE OF CONTENTS                    EXECUTIVE SUMMARY                         OIG SYNOPSIS         RCF REPORT   APPENDIX

                                                                                                                 BACK to COVER




Therefore, in addition to testing the relationship between mail mix and household reading,
reaction, and response to Marketing Mail, the data also allow for analysis of the other factors
which influence the relationship between households and advertising mail. The results of this
study provide valuable information to the Postal Service and to marketers regarding ways to
increase the value of direct mail. The results also provide a basis for thinking about the inter-
relation between the different types of mail in the mailbox and more broadly, about the role of
the Postal Service in the daily life of Americans.

Our report is organized as follows: Following this introduction, Section II discusses the HDS
study and data used in this report and presents key summary statistics. In Section III, the
logistic model used to estimate the impact of different variables is explained. Section IV
discusses the results of the analysis.

II.	    Overview of Household Diary Study Data

The Household Diary Study is an annual survey of approximately 5,200 households
conducted by the Postal Service and administered by NuStats of Austin, Texas. Each week
about 100 households record information about every piece of mail they receive that week,
including information on the sender, the physical characteristics of the mail piece and, for
advertising mail, how they interacted with the mail piece. In an initial recruitment survey,
households also provide demographic information about themselves including the age,
ethnicity, race, and educational attainment of the household head and their use of various
technological alternatives to the mail such as online presentment of bills and statements.

The HDS’s full account of the mail households receive in a week allows the household’s
weekly mail mix to be calculated, including the share of mail that is non-advertising, and,
more specifically, the share that is correspondence, transactions or periodicals. Table 1 reports
the average number of pieces of various kinds of mail received in a week by households
included in this study. The table also shows the shares of non-package mail accounted
for by the different types of mail received by households. As can be seen from the table,
about 36% of pieces recorded in the HDS were non-advertising and 64% were advertising.
Packages are recorded in the HDS but excluded from most of the analysis because they
are often not received at the same time as other mail. A later section presents results for
packages and their share of the mail mix is presented separately there as well.




Advertising Mail: Mail Mix Matters                                                                                          13
Report Number RARC-WP-19-003
TABLE OF CONTENTS                     EXECUTIVE SUMMARY                      OIG SYNOPSIS        RCF REPORT   APPENDIX

                                                                                                              BACK to COVER




           Table 1: HDS Volumes of Mail Received by Households, 2013-2017


                                                Pieces received per
                                               household per week,
                                                among household
                                                used in the reading      Share of all
       Type of mail                                   model           non-package mail
       Non-advertising                                       8.03              35.8%
             Periodicals                                     1.27               5.6%
             Correspondence                                  2.82              12.6%
             Transactions and other                          3.94              17.6%
       Advertising                                          14.41              64.2%
             First Class ads                                 0.72               3.2%
             Marketing Commercial                           11.03              49.2%
             Marketing Nonprofit                             2.66              11.8%
       All mail                                             22.44             100.0%

Households that completed the HDS from 2013 to 2017 reported receiving nearly 350,000
pieces of Marketing Mail and recorded information on all the variables used in this study for
more than 290,000 of those pieces.

Households completing the HDS record whether household members read the Marketing
Mail they received, whether they thought it was useful or interesting, and whether anyone in
the household was considering responding to it. Table 2 below reports the number of pieces
of Marketing Mail for which each answer choice was selected for reading, reaction and
response. Answer choices in bold were counted in this study as indicating that a mail piece
was read, reacted to positively or likely to be responded to, respectively. It should be noted
that the average “yes” response rate of 12.3 percent is much higher than actual response
rates to direct mail, which are typically 3 percent or less. However, the HDS question merely
asks if someone in the household is considering responding. Nonetheless, considering
responding is likely a necessary first step to actually responding.




Advertising Mail: Mail Mix Matters                                                                                       14
Report Number RARC-WP-19-003
TABLE OF CONTENTS                      EXECUTIVE SUMMARY                  OIG SYNOPSIS       RCF REPORT   APPENDIX

                                                                                                          BACK to COVER




    Table 2: Reading, Reaction, and Response Rates to Marketing Mail, 2013-2017

                                                      Observations for
                                                     which reading was
 Reading: was the mail piece…                            recorded        Observations used
 Read by a member of the household                          124,308             112,389
 Read by more than one member of the household               21,993              19,556
 Looked at but not read                                      64,341              58,350
 Discarded without being read                                96,821              87,437
 Set aside for reading later                                 18,409              16,390
 Total                                                      325,872             294,122
 Reading rate                                                 44.9%               44.9%

                                                     Observations for
                                                     which reactions
 Reaction: would this mail piece be described as…     were recorded      Observations used
 Useful information we like to receive                      134,646             121,912
 Interesting or enjoyable, but not useful                    50,756              45,866
 Neither interesting, enjoyable, nor useful                 137,311             123,788
 Objectionable or offensive                                   2587                2244
 Total                                                      325,300             293,810
 Positive reaction rate                                       57.0%               57.1%


                                                     Observations for
                                                     which intent to
                                                      respond or not
 Response: is anyone in your household considering     respond was
 responding?                                             recorded        Observations used
 Yes                                                         40,007              36,078
 No                                                         228,060             206,654
 Maybe                                                       55,382              49,956
 Total                                                      323,449             292,688
 Response rate                                                12.4%               12.3%

Advertising Mail: Mail Mix Matters                                                                                   15
Report Number RARC-WP-19-003
TABLE OF CONTENTS                    EXECUTIVE SUMMARY                     OIG SYNOPSIS        RCF REPORT   APPENDIX

                                                                                                            BACK to COVER




The total number of observations is smaller for reaction than for reading because
households answered the question on reaction less frequently than the question on reading.
The number of observations for response is smaller still for the same reason. The number
of observations used in the model is always smaller than the number for which reading,
reaction, or response was recorded because observations were dropped when other
important information about them was left out of the HDS.

Age, income, education, ethnicity, race and use of technological substitutes to the mail
could potentially determine the household members’ propensity to read, react positively
to, and respond to Marketing Mail independently of mail mix. Households completing the
HDS provide information about themselves on all these characteristics, though many
decline to provide income information. In its many years analyzing HDS data, RCF has
found the education level of the household head—a question more often answered—to
be a reasonable proxy for income. Receipt of online bills or statements was selected as a
measure of household members’ use of technological substitutes to the mail.

From 2013 to 2017, 26,304 households completed the HDS: 22,580 of which reported
receiving at least one piece of Marketing Mail during their HDS week for which they provided
a full set of information. The number and share of these households having the various
demographic characteristics included in this study is reported in Table 3 below. In some
cases, the questions in the HDS are more detailed than the categories used in this study. In
those cases, the categories used in the study are presented in the bold, shaded rows of the
tables. In all cases, the number of households recorded is the number used in the reading
model.




Advertising Mail: Mail Mix Matters                                                                                     16
Report Number RARC-WP-19-003
TABLE OF CONTENTS                    EXECUTIVE SUMMARY                   OIG SYNOPSIS   RCF REPORT   APPENDIX

                                                                                                     BACK to COVER




                        Table 3: Household Demographics, 2013-2017

                                                         Households
                                                          in reading     Share of
                       Age of household head                model       households
       18-24                                                   331          1.5%
               18-21                                            58          0.3%
               22-24                                           273          1.2%
       25-34                                                 2,639         11.7%
       35-44                                                 3,115         13.8%
       45-54                                                 4,013         17.8%
       55-64                                                 5,323         23.6%
       65-74                                                 4,562         20.2%
               65-69                                         2,661         11.8%
               70-74                                         1,901          8.4%
       75+                                                   2,597         11.5%
                          All households                    22,580        100.0%



                                                        Households in    Share of
             Educational attainment of household head   reading model   households
       High school or less                                   4,530         20.1%
             8th grade or less                                 132          0.6%
             Some high school                                  438          1.9%
             High school graduate                            3,960         17.5%
       Some college or technical school                      5,800         25.7%
             Some college                                    4,479         19.8%
             Technical school graduate                       1,321          5.9%
       At least college graduate                            12,250         54.3%
             College graduate                                6,821         30.2%
             Post graduate work                              5,429         24.0%
                          All households                    22,580        100.0%


Advertising Mail: Mail Mix Matters                                                                              17
Report Number RARC-WP-19-003
TABLE OF CONTENTS                     EXECUTIVE SUMMARY                       OIG SYNOPSIS    RCF REPORT   APPENDIX

                                                                                                           BACK to COVER




                                                      Households
 Did the household report receiving any bills or       in reading      Share of
 statements online in the past month?                    model        households
 No                                                        7,397          32.8%
 Yes                                                      15,183          67.2%
 Total                                                    22,580         100.0%

                                                      Households
 Is the household head of Spanish/Hispanic/Latino      in reading      Share of
 origin?                                                 model        households
 No                                                       21,649          95.9%
 Yes                                                         931           4.1%
 Total                                                    22,580         100.0%

                                                       Households
 Which of the following does the household head         in reading       Share of
 consider themselves to be?                               model         households
 White/Caucasian                                           20,203           89.5%
 Black/African American                                     1317             5.8%
 Asian                                                        544            2.4%
 Other                                                        516            2.3%
       American Indian and Alaska Native                      106            0.5%
       Native Hawaiian and other Pacific Islander              55            0.2%
       Other, specify                                         355            1.6%
 Total                                                     22,580         100.0%

For each piece of Marketing Mail it receives, a household records whether someone in
the household knows or has done business with the sender, as well as the industry of
the sender. Additionally, households report Marketing Mail sent with a nonprofit discount
separately from other Marketing Mail. Variables were included in the model based on
this information, and Table 4 below reports the number and share of observations used
in the reading model exhibiting each characteristic. Although slightly different numbers of



Advertising Mail: Mail Mix Matters                                                                                    18
Report Number RARC-WP-19-003
TABLE OF CONTENTS                     EXECUTIVE SUMMARY                      OIG SYNOPSIS    RCF REPORT   APPENDIX

                                                                                                          BACK to COVER




observations are used in the reaction model and the response model, the distribution of
those observations is virtually identical to what is shown below for the reading model.

            Table 4: Characteristics of Senders of Marketing Mail, 2013-2017


                                                           Observations in      Share of
 Sender characteristic                                     reading model      observations
 No past business relationship                                  102,697            34.9%
 Unknown if there is a past business relationship                29,779            10.1%
 Past business relationship                                     161,646            55.0%
 Nonprofit sender                                                57,924            19.7%
 Financial sender                                                63,575            21.6%
 Total observations                                             294,122           100.0%

For each piece of Marketing Mail, the household also records information on the shape of
the mail piece, whether a return envelope or card was included and whether the postage
was paid, whether the piece was specifically addressed to household members (as opposed
to “current occupant” for example) and whether the piece contained a coupon. Variables
were included in the model based on this information, and Table 5 below reports the number
and share of observations used in the reading model exhibiting each characteristic.




Advertising Mail: Mail Mix Matters                                                                                   19
Report Number RARC-WP-19-003
TABLE OF CONTENTS                    EXECUTIVE SUMMARY                       OIG SYNOPSIS     RCF REPORT   APPENDIX

                                                                                                           BACK to COVER




                   Table 5: Features of Marketing Mail Pieces, 2013-2017


                                                           Observations in       Share of
 Mail piece characteristic                                 reading model       observations
 In a letter size envelope                                      123,258             41.9%
 In an envelope larger than letter size (not catalog)             9,550              3.2%
 Catalog                                                         38,507             13.1%
 Detached label card                                              2085               0.7%
 Postcard                                                         6,763              2.3%
 Addressed flyer/circular/folded piece                          102,106             34.7%
 Newspaper/magazine/newsletter                                   11,853              4.0%
 Pre-stamped or postage paid return envelope or card
                                                                211,491             71.9%
 included
 Return envelope or card that needs a stamp included             36,099             12.3%
 No return envelope or card included                             46,532             15.8%
 Addressed to specific members of the household                 241,079             82.0%
 Contains a coupon                                               69,141             23.5%
 Total observations                                             294,122            100.0%

These HDS data will be used to estimate how household reading, reaction, and response
to Marketing Mail are affected by the mix of mail received by the households, as well as
household demographics, mailer characteristics, and mail piece features.

III.	     Econometric Approach

        1.	 The Logit Regression Model

The effect of mail mix and other variables on the three household treatments of advertising
mail was estimated using the logit model which is a commonly used econometric method
for binary dependent variables. As shown in Table 1 above, the values of the dependent
variable in all three models are binary meaning they only take on values equal to either0
(e.g., the household did not read the mail piece) or 1 (the household did read the mail
piece). Unlike commonly used linear regression models, the logit model forces the predicted
values (or y-hat values) of the dependent variable to fit between0 and 1 by estimating the


Advertising Mail: Mail Mix Matters                                                                                    20
Report Number RARC-WP-19-003
TABLE OF CONTENTS                                                EXECUTIVE SUMMARY                                                       OIG SYNOPSIS             RCF REPORT   APPENDIX

                                                                                                                                                                               BACK to COVER




variable parameters through the common s-shaped logistic curve which is presented below
in Equation 1.
                                                                                                                     1
                                   𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃(𝑦𝑦𝑦𝑦 = 1|𝑥𝑥𝑥𝑥) = 𝑓𝑓𝑓𝑓(𝛽𝛽𝛽𝛽𝛽𝛽𝛽𝛽) =                                                                Equation 1
                                                                                                      1 + 𝑒𝑒𝑒𝑒 −(𝛼𝛼𝛼𝛼+𝛽𝛽𝛽𝛽𝛽𝛽𝛽𝛽)               Equation 1

In order to present the results as a linear combination of parameters and variables, the logit
identity, also known as the log-odds, is used. First, the odds calculation is made by dividing
the probability that the dependent variable is equal to one by the probability that it is not
equal to one (Equation 2). Second, the log-odds calculation is made by taking the natural log
of both sides shown in Equation 3. It is notable that logit is the inverse function of Equation 1
meaning the presentation in terms of log-odds sets the variables as a function of . While
this is useful for presenting results, parameters are estimated through Equation 1
                                                        𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃(𝑦𝑦𝑦𝑦 = 1)
                                 𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂 =                                  = 𝑒𝑒𝑒𝑒 (𝛼𝛼𝛼𝛼+𝛽𝛽𝛽𝛽𝛽𝛽𝛽𝛽) = 𝑒𝑒𝑒𝑒 𝛼𝛼𝛼𝛼 𝑒𝑒𝑒𝑒 𝛽𝛽𝛽𝛽𝛽𝛽𝛽𝛽         Equation 2
                                                      1 − 𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃(𝑦𝑦𝑦𝑦 = 1)
                                                                                      𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃(𝑦𝑦𝑦𝑦 = 1)
                   𝐿𝐿𝐿𝐿𝑃𝑃𝑃𝑃𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿 (𝐿𝐿𝐿𝐿𝑃𝑃𝑃𝑃𝐿𝐿𝐿𝐿 𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂) = 𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙 �                               � = 𝛼𝛼𝛼𝛼 + 𝛽𝛽𝛽𝛽𝛽𝛽𝛽𝛽        Equation 3
                                                                                    1 − 𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃(𝑦𝑦𝑦𝑦 = 1)
Because the model is nonlinear, commonly used linear methods to estimate the equation
parameters such as ordinary least squares (OLS) are not applicable. This is because OLS
assumes the dependent variable is a linear function when estimating the parameters by
minimizing the vertical distance between the observed dependent variable and the predicted
value of a linear function.44

Maximum likelihood estimation (MLE) is commonly used for estimating parameters of
nonlinear models. MLE is based on the principle that out of all possible parameter values
for the function, the value that makes the likelihood of the observed data largest should
be chosen.45 In other words, MLE involves calculating the joint probability of obtaining the
sample of data and what parameter values maximize the likelihood of obtaining this sample.

For a binary dependent variable, the likelihood function takes the form of the probability
distribution function shown in Equation 4. For equal to 1, the function is equal to the
probability . For equal to0, the function is simply equal to 1 minus P.

                                                           𝑓𝑓𝑓𝑓(𝑦𝑦𝑦𝑦|𝑃𝑃𝑃𝑃) = 𝑃𝑃𝑃𝑃𝑦𝑦𝑦𝑦 (1 − 𝑃𝑃𝑃𝑃)1−𝑦𝑦𝑦𝑦 ; 𝑦𝑦𝑦𝑦 = 0,1                           Equation 4

 44	 Wooldridge, Jeffery. 2006. Introductory Econometrics: A Modern Approach, third edition. South-Western College
     Publishing. Cincinnati, OH.
 45	 Ibid.

Advertising Mail: Mail Mix Matters                                                                                                                                                        21
Report Number RARC-WP-19-003
TABLE OF CONTENTS                                        EXECUTIVE SUMMARY                                                      OIG SYNOPSIS            RCF REPORT   APPENDIX

                                                                                                                                                                     BACK to COVER




The likelihood function is the joint probability distribution of all observations (Equation 5).
As shown in Equation 5, by plugging in Equation 1 for and taking the log of both sides,
parameter values ( ) can be obtained.46 Following the MLE principle, the parameter values
that are chosen are the ones which maximize the likelihood function in Equation 5.

                                               1                                                    1
             ℒ(𝛽𝛽𝛽𝛽) = � 𝑦𝑦𝑦𝑦𝑖𝑖𝑖𝑖 ln �         −(𝛼𝛼𝛼𝛼+𝛽𝛽𝛽𝛽𝛽𝛽𝛽𝛽)
                                                                � + (1 − 𝑦𝑦𝑦𝑦𝑖𝑖𝑖𝑖 )ln �1 −          −(𝛼𝛼𝛼𝛼+𝛽𝛽𝛽𝛽𝛽𝛽𝛽𝛽)
                                                                                                                     � ; 𝑦𝑦𝑦𝑦𝑖𝑖𝑖𝑖 = 0,1   Equation 55
                                                                                                                                          Equation
                                      1 + 𝑒𝑒𝑒𝑒                                             1 + 𝑒𝑒𝑒𝑒


       2.	 Model Output – Coefficients, Odds-Ratios and Predicted
           Probabilities

The logit regression produces coefficients that are measured in log-odds, that is, the amount
by which a unit change in an explanatory variable changes the log of the odds of a positive
event (e.g. reading the mail piece) occurring. As shown in Equation 3, the relationship
between the explanatory variables and the log-odds of reading, etc., is linear, with a positive
value indicating that the variable increases the likelihood log odds of reading, etc., and a
negative value indicating that the variable reduces the likelihood.

Taking the anti-log of the log-odds coefficients produces odds-ratios, which can be more
intuitively understood though must be interpreted carefully (see next paragraph). These
represent the multiplicative change in the odds of reading, etc., as a result of a unit change
in the explanatory variable (see Equation 2). Odds-ratios are always positive, with an
odds-ratio greater than 1 meaning that an increase in the value of the explanatory variable
increases the likelihood of the event occurring, while an odds ratio less than 1 means that
an increase in the value of the variable decreases the likelihood of the event. For example,
an odds-ratio of 2 in a model of reading means that a one unit increase in the explanatory
variable doubles the odds of reading. An odds-ratio of0.5 would cut the odds of reading in
half for a unit increase in the value of a variable. Since odds-ratios are multiplicative, a value
close to 1 implies that the factor neither increases nor decreases the likelihood of the event
occurring. The statistical significance of the odds-ratio coefficient is measured relative to a
value of 1.

When interpreting odds-ratios it is important to keep in mind the difference between
probabilities and odds. If a mail piece has a 50% probability of being read, its odds of being
read are 1-to-1. A doubling of the odds due to an increase in the value of a variable with an
odds-ratio of 2 would change those odds to 2-to-1, which is probability of 67%, not 100%.
 46	 Ibid.

Advertising Mail: Mail Mix Matters                                                                                                                                              22
Report Number RARC-WP-19-003
TABLE OF CONTENTS                     EXECUTIVE SUMMARY                         OIG SYNOPSIS        RCF REPORT   APPENDIX

                                                                                                                 BACK to COVER


While the modeled effect of a variable on the odds of reading is constant—that is, an odds
ratio of 2 means the odds of reading will always increase by 100% per unit increase in that
variable—the effect on probability of reading changes depending on the initial probability of
reading. As we have seen, a unit increase in a variable with an odds-ratio of 2 will increase
the probability of reading from 50% to 67%, for a change of about 17 percentage points.
A further unit increase, however—from odds of 2 to 1 to odds of 4 to 1—would bring the
probability or reading to 80%, a change of only 13 percentage points. Further increases
would have even smaller effects on probability as it approaches 100%. Decreases in this
variable would also have smaller and smaller effects on probability as it approaches0%. This
produces the well-known s-shape of the logit model.

The fact that the change in probability depends on the initial probability creates a challenge
for assessing the effects of a variable on probability. For each variable of interest, this study
addresses this issue by presenting predicted probabilities at different values of that variable
when all other variables are held constant at their observed values.

For example, the predicted probability that a piece of mail is read for a household age 18
– 24 years of age is computed by plugging into Equation 1 the estimated coefficients with
the following X values:0 for the age of household head dummy (because age 18 – 24 is the
omitted age category), and for all other X variables, their estimated coefficients ( ) multiplied
by the observed value for each observation. This results in a predicted probability for
each observation assuming the head of the household is 18 – 24 years of age. Finally, the
predicted probabilities for all the observations are then averaged together.

This calculation generates a predicted reading rate for 18 – 24 year-olds of 41.1%. The
calculation of the reading rate for other groups follows the same process except for each
group the estimated age dummy coefficients are also included in the calculation. The
predicted reading rates for each age group are shown in Table 6.

                Table 6: Predicted Reading Rates for Different Age Groups
                                     Coefficient (Log-Odds)     Odds-Ratio       Predicted Value
         Age of HH Head                     Reading              Reading            Reading
  18-24 (omitted/base category)                -                    -                41.1%
              25-34                         -0.127                0.881              38.4%
              35-44                         -0.012                0.988              40.8%
              45-54                          0.074                1.077              42.6%
              55-64                          0.161                1.174              44.5%
              65-74                          0.303                1.353              47.5%
               75+                           0.451                1.570              50.7%
Advertising Mail: Mail Mix Matters                                                                                          23
Report Number RARC-WP-19-003
TABLE OF CONTENTS                              EXECUTIVE SUMMARY                                      OIG SYNOPSIS              RCF REPORT   APPENDIX

                                                                                                                                             BACK to COVER




           3. Assessing Goodness of Fit
As previously discussed, the model parameters are estimated using MLE and, therefore,
traditional goodness-of-fit statistics such as the R2 from OLS cannot be used.

A receiver operating characteristic curve (ROC curve) is a graphical plot that is commonly
used for determining the model fit for logistic regressions. ROC measures how well a model
discriminates between observations that are0 and 1 based on two measures of classification
known as ‘Sensitivity’ and ‘Specificity’.47 Sensitivity measures the percentage of observations
that are 1 and have been classified by the model as ‘1’. Specificity measures the percentage
of observations that are0 and have been classified by the model as ‘0’. While observed
data for the dependent variable is either0 or 1, the predicted probabilities will lie somewhere
between0 and 1. Therefore, in order for the model to discriminate between0 and 1, a cut-off
value needs to be assigned. A common default cut-off value is 50% meaning observations
are classified as ‘1’ if the predicted outcome (y-hat) is greater than or equal to 50%.
Observations are classified as ‘0’ if the predicted outcome (y-hat) is less than 50%. However,
it is arguable that using a cut-off value of 50% for classifying observations as0 or 1 is only
appropriate for models in which 50% of the observations are equal to 1.48

The ROC curve overcomes this issue by examining the probability of detecting a true
positive (Sensitivity) against a false positive (one minus Specificity) for an entire range of
possible cut-off values between0 and 100%.49 ROC then plots Sensitivity (true positive)
against one minus Specificity (false positive) relative to a 45-degree angle. The area
under the ROC curve (AUC) captures the entire space in which the model is detecting true
positives against false positives. AUC can be interpreted as the probability that a randomly
chosen observation with value of 1 is classified with a higher predicted probability than
a randomly selected observation with value of0. An AUC of0.5 would correspond to the
45-degree angle and be considered a poor model that classifies the data no better than
random. Based on an industry paper published by Deloitte, an AUC of0.7 or higher is
considered to be an acceptable model (Deloitte, 2016).50



 47	 Hosmer, D. W., Jr., S. A. Lemeshow, and R. X. Sturdivant. 2013. Applied Logistic Regression. 3rd ed. Hoboken, NJ: Wiley.
 48	 Wooldridge, Jeffery. 2006. Introductory Econometrics: A Modern Approach, third edition. South-Western College
     Publishing. Cincinnati, OH.
 49	 Hosmer, D. W., Jr., S. A. Lemeshow, and R. X. Sturdivant. 2013. Applied Logistic Regression. 3rd ed. Hoboken, NJ: Wiley.
 50	 Skantzos, Nikos; Castelein, Nicolas. 2016. Credit scoring - Case study in data analytics. Deloitte. Available online at:
     https://www2.deloitte.com/content/dam/Deloitte/global/Documents/Financial-Services/gx-be-aers-fsi-credit-scoring.pdf.

Advertising Mail: Mail Mix Matters                                                                                                                      24
Report Number RARC-WP-19-003
TABLE OF CONTENTS                             EXECUTIVE SUMMARY                                       OIG SYNOPSIS       RCF REPORT   APPENDIX

                                                                                                                                      BACK to COVER




There are some measures that attempt to present goodness-of-fit similar to an R2 from an
OLS regression such as the McFadden R2 shown in Equation 6 below. The McFadden R2
compares the log-likelihood calculation from the model                   with only an intercept to the
final log-likelihood with all parameters estimated              from Equation 5.51 Log-likelihood
is strictly negative; therefore, the final log-likelihood of the fitted model will be smaller in
absolute value than the log-likelihood of the unfitted model. By subtracting the ratio of the
two from 1, the value will be greater than0 and a larger R2 can be interpreted as a better
‘fit’. As       grows smaller in absolute value relative to           , the McFadden R2 increases.
Similar to traditional R2, the McFadden R2 is bounded between0 and 1. If the variables in the
model have no explanatory power, then the ratio of the two log-likelihoods will be 1 and the
McFadden R2 will be zero which follows a similar intuition as traditional R2 from OLS.
                                                                                    ℒ(𝛽𝛽𝛽𝛽)
                                             𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂 𝑅𝑅𝑅𝑅 2 = 1 − ℒ(0) > 0         Equation 6


       4.	 Variables Included in the Model

The effect of the mail mix and other variables on the household’s treatment of advertising
mail was modeled using three separate logistic regression models. The three models were
developed to explain the probability that 1) a piece of advertising mail was read; 2) the piece
generated a positive reaction (i.e., was found useful or interesting); and 3) the household is
considering responding to the mail piece. The basic structure of the model, shown for the
reading, is as follows:

                     Prob (Reading) = f (mail mix, household demographics,
                           sender characteristics, mail piece features)

Each observation describes a piece of Marketing Mail a household received and what
the household did with that piece of mail. The primary mail mix variable used in this
report is the share of non-package mail a household receives that is not advertising mail.
Additional analysis decomposed the non-advertising share of mail into separate shares for
First-Class and Periodicals Mail, and also a further decomposition of First-Class Mail into
correspondence mail and transactions mail. To control for other factors that might influence
or determine how a household treats advertising mail, several household demographic
variables, sender characteristics, and mail piece features were also included. The same

 51	 Wooldridge, Jeffery. 2006. Introductory Econometrics: A Modern Approach, third edition. South-Western College
     Publishing. Cincinnati, OH.

Advertising Mail: Mail Mix Matters                                                                                                               25
Report Number RARC-WP-19-003
TABLE OF CONTENTS                       EXECUTIVE SUMMARY                       OIG SYNOPSIS       RCF REPORT   APPENDIX

                                                                                                                BACK to COVER




variables were used in the equations for reading, reaction, and response. Table 7 describes
the variables and Tables 8 through 10 present the regression output for the reading,
reaction, and response models. Table 11 presents regression diagnostics for each of the
three logistic regression models. These models use the non-advertising share of mail as the
mail mix variable. Models using alternative mail mix measures are presented in Section IV of
this report.

                Table 7: Variables Used in the Logistic Regression Models

    Variables                       Description               Type              Values

 Dependent Variables

                    Was the mail piece read by one or       Binary
 Reading                                                                 Yes = 1, No =0
                    more members of the household?          Dummy
                    Does the respondent describe the mail   Binary
 Reaction                                                                Yes = 1, No =0
                    piece as useful or interesting?         Dummy
                    Are household members considering       Binary
 Response                                                                Yes = 1, Maybe or No =0
                    responding?                             Dummy

 Explanatory Variables – Mail Mix

                    Share of non-packages the household
 Non-advertising    received in a week that were not        Continuous   0% to 100%
                    advertising
                    Share of non-packages the household
 Correspondence     received in a week that were            Continuous   0% to 100%
                    correspondence
                    Share of non-packages the household
 Transactions       received in a week that were            Continuous   0% to 100%
                    transactions
                    Share of non-packages the household
 Periodicals        received in a week that were            Continuous   0% to 100%
                    periodicals




Advertising Mail: Mail Mix Matters                                                                                         26
Report Number RARC-WP-19-003
TABLE OF CONTENTS                     EXECUTIVE SUMMARY                          OIG SYNOPSIS         RCF REPORT   APPENDIX

                                                                                                                   BACK to COVER




     Variables                   Description                  Type                Values

  Explanatory Variables – Household Demographics

                                                                         Base category =0
                                                                           - age 18-24
                                                                         Dummy Variables = 1,0
                                                                         otherwise
                                                           Categorical     - age 25-34
  Age               Age of household head
                                                           Dummy           - age 35-44
                                                                           - age 45-54
                                                                           - age 55-64
                                                                           - age 65-74
                                                                           - age 75+
                                                                         Base category =0
                                                                           - High school or less
                                                                         Dummy Variables = 1,0
                    Educational attainment of household    Categorical
  Education                                                              otherwise:
                    head                                   Dummy
                                                                           - Some college /
                                                                           technical school
                                                                           - College degree or more
                    Does the household receive bills and   Binary
  Presentment                                                              Yes = 1, No =0
                    statements online?                     Dummy
                                                           Binary
  Hispanic          Is the head of household Hispanic?                     Yes = 1, No =0
                                                           Dummy
                                                                         Base category =0
                                                                           - Caucasian
                                                                         Dummy Variables = 1,0
                    What is the race of the head of        Categorical   otherwise:
  Race
                    household?                             Dummy
                                                                           - African American
                                                                           - Asian
                                                                           - Other




Advertising Mail: Mail Mix Matters                                                                                            27
Report Number RARC-WP-19-003
TABLE OF CONTENTS                      EXECUTIVE SUMMARY                          OIG SYNOPSIS      RCF REPORT   APPENDIX

                                                                                                                 BACK to COVER




    Variables                     Description                  Type                Values

 Explanatory Variables – Sender Characteristics

                     Was the mail piece from an                            Yes = 1
 Past Business
                     organization someone in the            Scale          Unknown =0
 Relationship
                     household has done business with?                     No = -1
                     Was the sender in the financial        Binary
 Financial Sender                                                          Yes = 1, No =0
                     industry?                              Dummy
                                                            Binary
 Nonprofit           Is the mail piece Nonprofit?                          Yes = 1, No =0
                                                            Dummy

 Explanatory Variables – Mail Piece Features

 Specifically        Was the mail piece specifically        Binary
                                                                           Yes = 1, No =0
 Addressed           addressed to household members?        Dummy
                                                                          Base category =0
                                                                            - Letter
                                                                          Dummy Variables = 1,0
                                                                          otherwise:
                                                                            - Flat
                     What was the shape/type of mail        Categorical
 Shape                                                                      - Catalog
                     piece                                  Dummy
                                                                            - Detached label card
                                                                            - Postcard
                                                                            - Flyer
                                                                             Newspaper/Magazine/
                                                                            Newsletter
                                                            Binary
 Coupon              Did the mail piece contain a coupon?                    Yes = 1, No =0
                                                            Dummy
                                                                          Base category =0
                                                                            - No
                     Was a return envelope or card          Categorical   Dummy Variables = 1,0
 Return Envelope
                     included?                              Dummy         otherwise:
                                                                            - Yes, stamp needed
                                                                            - Yes, postage paid




Advertising Mail: Mail Mix Matters                                                                                          28
Report Number RARC-WP-19-003
TABLE OF CONTENTS                         EXECUTIVE SUMMARY                         OIG SYNOPSIS      RCF REPORT   APPENDIX

                                                                                                                   BACK to COVER




                       Table 8: Logit Regression Results for Reading Rate

                                                Estimated Coefficient
                                                                                 Odds Ratio
 VARIABLES                                            (Log-Odds)
                                              Coefficient     Std. Error   Coefficient   Std. Error
 Mail Mix
       Non-Advertising Share                   0.498***         (0.029)     1.645***      (0.048)
  Household Demographics
  Age of household head (base age ‘18-24’)
       25-34                                   -0.127**         (0.053)     0.881**       (0.047)
       35-44                                   -0.012           (0.052)     0.988         (0.052)
       45-54                                   0.074            (0.052)     1.077         (0.056)
       55-64                                   0.161***         (0.052)     1.174***      (0.061)
       65-74                                   0.303***         (0.052)     1.353***      (0.070)
       75+                                     0.451***         (0.052)     1.570***      (0.082)
  Educational attainment of household head
    (base education level ‘High school or
      less’)
       Some college or technical school        -0.136***        (0.013)     0.873***      (0.011)
       At least college graduate               -0.371***        (0.011)     0.690***      (0.008)
  Technology Use
      Household receives bills & statements
                                               -0.218***        (0.009)     0.804***      (0.007)
      online
  Race/ethnicity of head of household
    (base head of household ‘Caucasian’)
      Hispanic head of household               0.325***         (0.023)     1.385***      (0.033)
      African American head of household       0.552***         (0.020)     1.737***      (0.035)
      Asian head of household                  0.483***         (0.028)     1.621***      (0.045)
      Other (nonwhite) head of household       0.212***         (0.032)     1.236***      (0.040)
 Sender Characteristics
       Past Business Relationship              0.705***         (0.005)     2.025***      (0.009)
       Nonprofit                               0.183***         (0.012)     1.201***      (0.015)
       Financial Sender                        -0.387***        (0.012)     0.679***      (0.008)




Advertising Mail: Mail Mix Matters                                                                                            29
Report Number RARC-WP-19-003
TABLE OF CONTENTS                        EXECUTIVE SUMMARY                            OIG SYNOPSIS      RCF REPORT   APPENDIX

                                                                                                                     BACK to COVER




                                               Estimated Coefficient
                                                                                   Odds Ratio
 VARIABLES                                           (Log-Odds)
                                             Coefficient     Std. Error     Coefficient    Std. Error
 Mail Piece Features
      Specifically addressed to HH members    0.332***         (0.012)        1.394***      (0.017)
      Contains coupon                         0.468***         (0.011)        1.597***      (0.018)
  Shape of Mail Piece
    (base shape ‘letters’)
      Flat                                    0.212***         (0.023)        1.237***      (0.028)
      Catalog                                 -0.143***        (0.014)        0.867***      (0.013)
      Detached label card                     -0.487***        (0.050)        0.615***      (0.031)
      Postcard                                0.633***         (0.029)        1.883***      (0.054)
      Flyer                                   0.097***         (0.012)        1.102***      (0.013)
      Newspaper/newsletter/magazine           0.238***         (0.021)        1.269***      (0.027)
  Return Envelope
    (base response ‘No’)
      Return envelope without postage         -0.111***        (0.015)        0.895***      (0.013)
      Postage paid return envelope            -0.344***        (0.013)        0.709***      (0.009)
  Constant                                    -0.682***        (0.055)        0.505***      (0.028)
 *** p<0.01, ** p<0.05, * p<0.1

                       Table 9: Logit Regression Results for Reaction Rate

                                               Estimated Coefficient
                                                                                   Odds Ratio
 VARIABLES                                            (Log-Odds)
                                              Coefficient      Std. Error   Coefficient    Std. Error
 Mail Mix
       Non-Advertising Share                   0.214***         (0.031)       1.239***      (0.039)
  Household Demographics
  Age of household head (base age ‘18-24’)
       25-34                                   -0.158***        (0.055)       0.854***      (0.047)
       35-44                                   0.023            (0.054)       1.023         (0.055)
       45-54                                   0.123**          (0.053)       1.131**       (0.060)
       55-64                                   0.188***         (0.053)       1.207***      (0.064)
       65-74                                   0.339***         (0.053)       1.403***      (0.075)
       75+                                     0.525***         (0.054)       1.691***      (0.092)


Advertising Mail: Mail Mix Matters                                                                                              30
Report Number RARC-WP-19-003
TABLE OF CONTENTS                         EXECUTIVE SUMMARY                         OIG SYNOPSIS         RCF REPORT   APPENDIX

                                                                                                                      BACK to COVER


                                                Estimated Coefficient
                                                                                    Odds Ratio
 VARIABLES                                             (Log-Odds)
                                               Coefficient      Std. Error   Coefficient    Std. Error
  Educational attainment of household head
    (base education level ‘High school or
      less’)
       Some college or technical school         -0.062***        (0.014)       0.940***      (0.013)
       At least college graduate                -0.209***        (0.012)       0.812***      (0.010)
  Technology Use
      Household receives bills & statements
                                                -0.209***        (0.010)       0.812***      (0.008)
      online
  Race/ethnicity of head of household
    (base head of household ‘Caucasian’)
      Hispanic head of household                0.251***         (0.025)       1.285***      (0.033)
      African American head of household        0.633***         (0.022)       1.884***      (0.042)
      Asian head of household                   0.173***         (0.029)       1.189***      (0.035)
      Other (nonwhite) head of household        0.173***         (0.035)       1.189***      (0.041)
 Sender Characteristics
       Past Business Relationship               0.875***         (0.005)       2.400***      (0.011)
       Nonprofit                                0.452***         (0.013)       1.572***      (0.021)
       Financial Sender                         -0.588***        (0.013)       0.556***      (0.007)
 Mail Piece Features
       Specifically addressed to HH members     0.261***         (0.012)       1.298***      (0.016)
       Contains coupon                          0.672***         (0.012)       1.958***      (0.024)
  Shape of Mail Piece
    (base shape ‘letters’)
       Flat                                     0.478***         (0.024)       1.612***      (0.039)
       Catalog                                  1.231***         (0.017)       3.424***      (0.057)
       Detached label card                      -0.508***        (0.049)       0.602***      (0.029)
       Postcard                                 0.529***         (0.030)       1.697***      (0.051)
       Flyer                                    0.417***         (0.012)       1.517***      (0.018)
       Newspaper/newsletter/magazine            1.270***         (0.026)       3.560***      (0.091)
  Return Envelope
    (base response ‘No’)
       Return envelope without postage          0.099***         (0.016)       1.104***      (0.018)
       Postage paid return envelope             -0.134***        (0.014)       0.874***      (0.012)
  Constant                                      -0.570***        (0.057)       0.566***      (0.032)
 *** p<0.01, ** p<0.05, * p<0.1


Advertising Mail: Mail Mix Matters                                                                                               31
Report Number RARC-WP-19-003
TABLE OF CONTENTS                         EXECUTIVE SUMMARY                          OIG SYNOPSIS     RCF REPORT   APPENDIX

                                                                                                                   BACK to COVER




                     Table 10: Logit Regression Results for Response Rate

                                               Estimated Coefficient
                                                                                Odds Ratio
 VARIABLES                                            Log-Odds
                                              Coefficient    Std. Error   Coefficient    Std. Error
 Mail Mix
       Non-Advertising Share                   0.253***        (0.044)      1.288***       (0.056)
  Household Demographics
  Age of household head (base age ‘18-24’)
       25-34                                   -0.244***       (0.085)      0.783***       (0.067)
       35-44                                   -0.013          (0.083)      0.987          (0.082)
       45-54                                   0.122           (0.082)      1.130          (0.093)
       55-64                                   0.157*          (0.082)      1.170*         (0.096)
       65-74                                   0.176**         (0.082)      1.192**        (0.098)
       75+                                     0.269***        (0.083)      1.309***       (0.109)
  Educational attainment of household head
    (base education level ‘High school or
      less’)
       Some college or technical school        -0.045**        (0.019)      0.956**        (0.018)
       At least college graduate               -0.220***       (0.016)      0.802***       (0.013)
  Technology Use
      Household receives bills & statements
                                               -0.173***       (0.013)      0.841***       (0.011)
      online
  Race/ethnicity of head of household
    (base head of household ‘Caucasian’)
      Hispanic head of household               0.099***        (0.035)      1.104***       (0.038)
      African American head of household       0.471***        (0.027)      1.602***       (0.043)
      Asian head of household                  -0.019          (0.047)      0.981          (0.046)
      Other (nonwhite) head of household       0.117**         (0.048)      1.124**        (0.054)
 Sender Characteristics
       Past Business Relationship              1.053***        (0.011)      2.867***       (0.030)
       Nonprofit                               0.237***        (0.019)      1.268***       (0.024)
       Financial Sender                        -0.691***       (0.023)      0.501***       (0.012)




Advertising Mail: Mail Mix Matters                                                                                            32
Report Number RARC-WP-19-003
TABLE OF CONTENTS                       EXECUTIVE SUMMARY                          OIG SYNOPSIS         RCF REPORT   APPENDIX

                                                                                                                     BACK to COVER




                                               Estimated Coefficient
                                                                                 Odds Ratio
 VARIABLES                                            Log-Odds
                                              Coefficient    Std. Error    Coefficient     Std. Error
 Mail Piece Features
      Specifically addressed to HH members      0.081***       (0.018)       1.084***       (0.020)
      Contains coupon                           1.087***       (0.015)       2.966***       (0.046)
  Shape of Mail Piece
    (base shape ‘letters’)
      Flat                                      0.406***       (0.032)       1.500***       (0.048)
      Catalog                                   -0.103***      (0.022)       0.902***       (0.020)
      Detached label card                       -0.462***      (0.098)       0.630***       (0.062)
      Postcard                                  0.511***       (0.037)       1.668***       (0.062)
      Flyer                                     0.131***       (0.018)       1.140***       (0.020)
      Newspaper/newsletter/magazine             -0.234***      (0.035)       0.792***       (0.028)
  Return Envelope
    (base response ‘No’)
      Return envelope without postage           0.446***       (0.021)       1.562***       (0.033)
      Postage paid return envelope              0.074***       (0.022)       1.076***       (0.024)
  Constant                                      -2.972***      (0.087)       0.051***       (0.004)
 *** p<0.01, ** p<0.05, * p<0.1

                             Table 11: Regression Equation Diagnostics

                                             Reading            Reaction                Response
 Number of Observations                      294,122            293,810                  292,688
 Area Under ROC Curve (AUC)                   0.7191             0.7838                   0.7732
 McFadden R-squared                            0.109              0.189                   0.144

The ROC curves for each model measure an AUC greater than0.7 which indicates that
each model has an acceptable predictive power. The R-squares for these regressions are
not high which is typical of cross-sectional data. There are a wide range of idiosyncratic
characteristics of households that are not reported in the HDS which impact household
treatment of advertising mail. Nevertheless, the large sample size and statistical significance
of many of the variables demonstrates that the model reliably estimates how individual
factors impact household reading, reaction, and response to advertising mail.


Advertising Mail: Mail Mix Matters                                                                                              33
Report Number RARC-WP-19-003
TABLE OF CONTENTS                    EXECUTIVE SUMMARY                         OIG SYNOPSIS         RCF REPORT   APPENDIX

                                                                                                                 BACK to COVER




IV.	    Analysis of Econometric Results

Logistic models are estimated to determine which factors affect the probability that a
household will 1) read a specific piece of Marketing Mail; 2) have a positive reaction to the
mail piece, defined as finding it interesting or useful; and 3) indicate that someone in the
household is likely to respond to the advertising.

Two issues are investigated. The first is whether the mix of mail, defined initially as the share
of a household’s weekly mail that is non-advertising, affects household reading, reaction,
and response to Marketing Mail. Advertising mail includes Marketing Mail and First-Class
advertising mail but our study only looks at household treatment of Marketing Mail. Non-
advertising mail includes other First-Class Mail (e.g., correspondence and transactions mail)
and Periodicals Mail. Packages are not included in the household mail mix because they are
often received separately from other mail. For the purposes of this discussion we refer to a
greater share of non-advertising mail as a more favorable mail mix. The hypotheses are that
a more favorable mail mix increases the likelihood that households will read their Marketing
Mail, react favorably to this mail, and be more likely to respond to the advertising.

The second issue investigated is what other factors, besides mail mix, affect household
reading, reaction, and response. These other factors are categorized as: 1) household
demographics; 2) mailer characteristics; and 3) features of the individual Marketing Mail
pieces. We also look at whether the different components of non-advertising mail (First-
Class correspondence mail, First-Class transactions mail, and Periodicals Mail) have
separate impacts on household treatment of Marketing Mail. Lastly, we look at the impact of
receiving packages. The impact of each of these variables on household reading, reaction,
and response to Marketing Mail are discussed in turn.

        A.	 Mail Mix

The logistic regression results show that a favorable mail mix has a significantly positive
impact on the probability of reading Marketing Mail. Mail mix also affects household reaction
and response to Marketing Mail with a higher non-advertising share of mail being associated
with a generally more positive view of the advertising mail that is received and a higher
probability of responding to any individual mail piece. The impacts of mail mix on the reaction
and response rates are not as large as the impact on the reading rate but are statistically
significant at the 99 percent confidence level.

Advertising Mail: Mail Mix Matters                                                                                          34
Report Number RARC-WP-19-003
TABLE OF CONTENTS                       EXECUTIVE SUMMARY                           OIG SYNOPSIS         RCF REPORT   APPENDIX

                                                                                                                      BACK to COVER




                                Table 12: Mail Mix Odds-Ratios

                                                 Reading            Reaction           Response
          Non-Advertising Share                   1.645***           1.239***           1.288***
                                                  (0.048)            (0.039)            (0.056)
                                          Standard error in parentheses *** p<0.01, ** p<0.05, * p<0.1

In Table 13, the predicted reading, reaction, and response rates at different mail mix
percentages are presented. These predicted rates are calculated using the regression
coefficients, evaluating all the other variables in the regression at their observed values and
calculating the impact of discreet 10 percent changes in the non-advertising share of mail
received by the household.

  Table 13: Predicted Reading, Reaction, and Response Rates at Different Mail Mixes

  Percent of Household Mail          Predicted Value       Predicted Value        Predicted Value
  that is non-advertising               Reading               Reaction               Response
          0%                             41.5%                  55.8%                  11.6%
          10%                            42.5%                  56.2%                  11.8%
          20%                            43.6%                  56.6%                  12.0%
          30%                            44.7%                  57.0%                  12.3%
          40%                            45.7%                  57.4%                  12.5%
          50%                            46.8%                  57.8%                  12.8%
          60%                            47.8%                  58.2%                  13.0%
          70%                            48.9%                  58.6%                  13.3%
          80%                            50.0%                  59.0%                  13.5%
          90%                            51.0%                  59.4%                  13.8%

As shown in Table 13, increases in the non-advertising share of mail are associated with
increases in household reading, positive reaction, and response to advertising mail. For
example, moving from a non-advertising share of 40 percent to a share of 50 percent
increases the reading rate from 45.7 percent to 46.8 percent, the positive reaction rate from
57.4 percent to 57.8 percent, and the potential response rate from 12.5 percent to 12.8
percent. Although these increases are relatively small, they are statistically significant. They
are also likely to be important in terms of the overall value of advertising mail as even small
increases in response rates can be important to direct mail marketers.
Advertising Mail: Mail Mix Matters                                                                                               35
Report Number RARC-WP-19-003
TABLE OF CONTENTS                      EXECUTIVE SUMMARY                           OIG SYNOPSIS         RCF REPORT   APPENDIX

                                                                                                                     BACK to COVER




           B.	 Household Demographics
	          Age of Household Head

Reading of Marketing Mail is strongly correlated with the age of the household head, with
the odds-ratio coefficients increasing as age increases, and with the coefficients on the
oldest age groups being statistically significant. Age also affects household reaction, with
older households being more likely to have a positive reaction to Marketing Mail. Finally, the
odds-ratio coefficients on the likelihood of response are also highly correlated with age and
statistically significant for the older households. All these coefficients are measured relative
to the youngest age group (18 – 24) and are consistent with the view that older people are
generally more receptive to the mail than younger people.

                                     Table 14: Age Odds-Ratios

    Age of household head (base age = 18 -24)      Reading           Reaction          Response
            25-34                                   0.881**            0.854***          0.783***
                                                    (0.047)            (0.047)           (0.067)
            35-44                                   0.988              1.023             0.987
                                                    (0.052)            (0.055)           (0.082)
            45-54                                   1.077              1.131**           1.130
                                                    (0.056)            (0.060)           (0.093)
            55-64                                   1.174***           1.207***          1.170*
                                                    (0.061)            (0.064)           (0.096)
            65-74                                   1.353***           1.403***          1.192**
                                                    (0.070)            (0.075)           (0.098)
            75+                                     1.570***           1.691***          1.309***
                                                    (0.082)            (0.092)           (0.109)
                                         Standard error in parentheses *** p<0.01, ** p<0.05, * p<0.1

Table 15 shows that households headed by someone under the age of 35 read about 40
percent of their Marketing Mail. Beyond that age, household reading increases uniformly
with the age of the household head. Increases in household age beyond the 25-34 year age
group also uniformly increase the likelihood that a household will have a positive reaction
and to indicate that they are likely to respond. For example, households headed by someone
aged 75 or over are about 1.5 times more likely to respond to advertising mail than are
households headed by someone aged 25 to 34 (13.8 percent vs 9.1 percent).

Advertising Mail: Mail Mix Matters                                                                                              36
Report Number RARC-WP-19-003
TABLE OF CONTENTS                        EXECUTIVE SUMMARY                             OIG SYNOPSIS        RCF REPORT   APPENDIX

                                                                                                                        BACK to COVER




                 Table 15: Predicted Reading, Reaction, and Response Rates
                                 and Age of Household Head

    Age of Household         Predicted Value           Predicted Value              Predicted Value
          Head                  Reading                   Reaction                     Response
         18-24                   41.1%                        53.1%                     11.1%
         25-34                   38.4%                        50.1%                     9.1%
         35-44                   40.8%                        53.6%                     11.0%
         45-54                   42.6%                        55.5%                     12.3%
         55-64                   44.5%                        56.7%                     12.6%
         65-74                   47.5%                        59.5%                     12.8%
           75+                   50.7%                        62.9%                     13.8%


	          Education of Household Head

Education has a negative impact on reading, reaction, and response. Households headed
by a college graduate, for example, are significantly less likely to read a piece of Marketing
Mail, less likely to have a positive reaction to it, and less likely to respond to it. For
marketers, these negative impacts need to be balanced against the advantages of targeting
higher education households that are also likely to have higher incomes than less educated
households.

                                  Table 16: Education Odds-Ratios

    Educational attainment of head of
    household (base education level ‘High
    school or less’)                                Reading           Reaction             Response
    Some college or technical school                0.873***             0.940***           0.956**
                                                    (0.011)              (0.013)            (0.018)
    College degree or more                          0.690***             0.812***           0.802***
                                                    (0.008)              (0.010)            (0.013)
                                            Standard error in parentheses *** p<0.01, ** p<0.05, * p<0.1




Advertising Mail: Mail Mix Matters                                                                                                 37
Report Number RARC-WP-19-003
TABLE OF CONTENTS                            EXECUTIVE SUMMARY                                   OIG SYNOPSIS           RCF REPORT   APPENDIX

                                                                                                                                     BACK to COVER




      Table 17: Predicted Reading, Reaction, and Response Rates and Education of
                                    Household Head

                                               Predicted Value          Predicted Value          Predicted Value
 Education of Household Head
                                                  Reading                  Reaction                 Response
 High School Degree or less                          50.4%                    59.7%                    13.7%
 Some College or Technical School                    47.5%                    58.6%                    13.3%
 College Degree or more                              42.4%                    55.9%                    11.6%


           Ethnicity and Race

The logistic regression results show that Hispanic households are more likely to read
Marketing Mail, have a positive reaction to the mail, and respond to the advertising, than
are non-Hispanic households. In all cases, the difference is statistically significant. Similarly,
African-American, Asian-American, and other non-white households are more likely to read
Marketing Mail and have a positive reaction than are Caucasian/white households. African-
American households are also more likely to respond to Marketing Mail advertising than
other households. That Hispanic and non-white households are more receptive to Marketing
Mail is meaningful because these households actually receive less Marketing Mail.52
According to the HDS, non-Hispanic white households received an average of 13.8 pieces
of Marketing Mail per week compared with 10.5 pieces per week for Hispanic households,
9.4 pieces per week for African-American households, and 11.7 pieces per week for Asian-
American households. Our analysis indicates that these households may be an untapped
market for direct mail marketers.




 52	 Although non-white and Hispanic households may have a different mail mix than white non-Hispanic households, the
     impact of any difference in mail mixes is already accounted for within the regression equation.

Advertising Mail: Mail Mix Matters                                                                                                              38
Report Number RARC-WP-19-003
TABLE OF CONTENTS                      EXECUTIVE SUMMARY                          OIG SYNOPSIS         RCF REPORT   APPENDIX

                                                                                                                    BACK to COVER




                            Table 18: Race/Ethnicity Odds-Ratios

 Race/ethnicity of head of household
 (base head of household “Caucasian’)               Reading          Reaction          Response
 Hispanic head of household                          1.385***         1.285***          1.104***
                                                     (0.033)          (0.033)           (0.038)
 African American head of household                  1.737***         1.884***          1.602***
                                                     (0.035)          (0.042)           (0.043)
 Asian head of household                             1.621***         1.189***          0.981
                                                     (0.045)          (0.035)           (0.046)
 Other (nonwhite) head of household                  1.236***         1.189***          1.124**
                                                     (0.040)          (0.041)           (0.054)
                                        Standard error in parentheses *** p<0.01, ** p<0.05, * p<0.1


  Table 19: Predicted Reading, Reaction, and Response Rates by Ethnicity and Race

                              Predicted Value          Predicted Value           Predicted Value
 Ethnicity/Race
                                 Reading                  Reaction                  Response
 Not Hispanic                        44.6%                 56.9%                     12.3%
 Hispanic                            51.6%                 61.5%                     13.3%
 Caucasian/White                     44.1%                 56.5%                     12.1%
 African-American                    55.8%                 67.8%                     17.2%
 Asian-American                      54.4%                 59.7%                     11.9%
 Other Nonwhite                      48.6%                 59.7%                     13.2%


        Technology Use – Online Bill Presentment

Households that receive online bills and statements are less likely to read their Marketing
Mail, are less likely to have a positive reaction to it, and are less likely to respond to the
advertising. All of these results are consistent with the idea that these households are less
connected to their mail than households that rely exclusively on the mail for the receipt of
their bills and statements.




Advertising Mail: Mail Mix Matters                                                                                             39
Report Number RARC-WP-19-003
TABLE OF CONTENTS                          EXECUTIVE SUMMARY                             OIG SYNOPSIS       RCF REPORT   APPENDIX

                                                                                                                         BACK to COVER




                                 Table 20: Technology Use Odds-Ratios

             Household Technology Use                    Reading            Reaction          Response
    Receives bills or statements online                   0.804***          0.812***           0.841***
                                                          (0.007)           (0.008)            (0.011)
                                             Standard error in parentheses *** p<0.01, ** p<0.05, * p<0.1


     Table 21: Predicted Reading, Reaction, and Response Rates and Technology Use

    Receives Bills or            Predicted Value          Predicted Value              Predicted Value
    Statements Online               Reading                  Reaction                     Response
    No                               48.0%                     59.8%                       13.5%
    Yes                              43.4%                     55.9%                       11.8%


           C.	 Mailer Characteristics

	          Past Business Relationship

HDS households are asked whether the Marketing Mail received was sent by a business
with which the household has a past business relationship (i.e. whether at least one
household member is an existing customer). Existence of a past business relationship
has a strong impact on household reading of Marketing Mail, an even stronger impact on
household reaction, and a still stronger impact on the likelihood of response.

                         Table 22: Odds-Ratios for Mailer Characteristics

       Mailer Characteristics                            Reading            Reaction          Response
    Past business relationship                            2.025***          2.400***           2.867***
                                                          (0.009)           (0.011)            (0.030)
    Nonprofit                                             1.201***          1.572***           1.268***
                                                          (0.015)           (0.021)            (0.024)
    Financial sender                                      0.679***          0.556***           0.501***
                                                          (0.008)           (0.007)            (0.012)

                                             Standard error in parentheses *** p<0.01, ** p<0.05, * p<0.1



Advertising Mail: Mail Mix Matters                                                                                                  40
Report Number RARC-WP-19-003
TABLE OF CONTENTS                        EXECUTIVE SUMMARY                     OIG SYNOPSIS        RCF REPORT   APPENDIX

                                                                                                                BACK to COVER




	            Nonprofit Postage

Households are also significantly more likely to read, react positively, and respond to mail
sent using nonprofit postage. These pieces are used by nonprofit and other social agencies
that qualify for the reduced postage rate. The strongest effect is on reaction suggesting that
even when households do not respond to nonprofit mailings (most of which are requests for
donations) they still view these mailings positively.

	            Financial Industry Sender

In contrast to mail from nonprofit senders, households are significantly less likely to read,
react positively, and respond to mail sent by the financial sector. Many of these mailings are
solicitations from credit card companies which tend to be particularly unpopular.

Table 23 presents the predicted reading, reaction, and response rates for the different mailer
characteristics. Marketing Mail sent by a business that has a past business relationship with
the recipient is more than twice as likely to be read and generate a positive reaction, and six
times as likely to get a response than mail sent by businesses with no past relationship. Yet
it is important to recognize that companies must send mail to households with which they
do not have a past business relationship in order to generate new customers. While this
“prospecting” mail is far less likely to be read it can be an important first step to creating a
new customer and a future business relationship.

 Table 23: Predicted Reading, Reaction, and Response Rates by Mailer Characteristics

                                          Predicted Value   Predicted Value   Predicted Value
         Past business relationship?
                                             Reading           Reaction          Response
    No                                          26.0%           35.0%              2.9%
    Yes                                         57.3%           71.9%              18.4%

    Was the sender a nonprofit?
    No                                          44.1%           55.4%              11.8%
    Yes                                         48.0%           63.7%              14.2%

    Was the sender in the financial industry?
    No                                          46.5%           59.6%              13.2%
    Yes                                         38.3%           48.1%              7.5%

Advertising Mail: Mail Mix Matters                                                                                         41
Report Number RARC-WP-19-003
TABLE OF CONTENTS                    EXECUTIVE SUMMARY                          OIG SYNOPSIS        RCF REPORT   APPENDIX

                                                                                                                 BACK to COVER




Table 23 also shows the reading, reaction, and response rates for mail sent by a nonprofit
company and mail sent by a company in the financial industry. One interesting result is that
households are almost as likely to indicate that they are considering responding to a non-
financial for-profit business (13.2%) as they are to a nonprofit company (14.2%). Another
point to keep in mind is that the positive impact of having a past business relationship can
override the negative impact of being from the financial industry.

        D.	 Mail Piece Characteristics

         1.	 Mail Piece Shape

The logistic regression analysis provides information on the relationship between the shape
of the Marketing Mail piece and household reading, reaction, and response to the mailing.
Seven different mail piece shapes are considered: letter, flat, catalog, detached label card,
postcard, flyer, and newsletter. Within the regression the omitted shape category is “letter” so
the coefficients reflect differences in the reading, reaction, and response rates of non-letter
pieces relative to letters.

Flats are significantly more likely to be read, create a positive reaction, and generate a likely
response than letters. Catalogs have an interesting relationship with households. They are
no more likely to be read or responded to, but they create a strong positive reaction. One
feature of catalogs is that they are not likely to be read immediately, instead often being set
aside for later reading. “Set aside for later” is one of the responses households can give to
the reading question and catalogs have a high “set aside” rate. Thus, the analysis suggests
that people enjoy receiving catalogs even if they do not immediately read them.

Postcards are more likely to be read, generate a positive reaction, and a likely response than
letters. The same holds true for flyers though the impact is not as strong as for postcards.
Newsletters are more likely to be read, and like catalogs are far more likely to generate a
positive reaction.




Advertising Mail: Mail Mix Matters                                                                                          42
Report Number RARC-WP-19-003
TABLE OF CONTENTS                      EXECUTIVE SUMMARY                           OIG SYNOPSIS         RCF REPORT   APPENDIX

                                                                                                                     BACK to COVER




                         Table 24: Odds-Ratios for Mail Piece Shape

    Shape (letters are the base)                     Reading           Reaction         Response
 Flat                                                1.237***          1.612***          1.500***
                                                     (0.028)           (0.039)           (0.048)
 Catalog                                             0.867***          3.424***          0.902***
                                                     (0.013)           (0.057)           (0.020)
 Detached label card                                 0.615***          0.602***          0.630***
                                                     (0.031)           (0.029)           (0.062)
 Postcard                                            1.883***          1.697***          1.668***
                                                     (0.054)           (0.051)           (0.062)
 Flyer                                               1.102***          1.517***          1.140***
                                                     (0.013)           (0.018)           (0.020)
 Newspaper/newsletter/magazine                       1.269***          3.560***          0.792***
                                                     (0.027)           (0.091)           (0.028)
                                         Standard error in parentheses *** p<0.01, ** p<0.05, * p<0.1

Table 25 presents the predicted reading, reaction, and response rate by mail piece shape.


         Table 25: Predicted Reading, Reaction, and Response Rates by Shape

                                   Predicted Value       Predicted Value          Predicted Value
 Mail Piece Shape
                                      Reading               Reaction                 Response
 Letter                                44.0%                   49.9%                  11.8%
 Flat                                  48.5%                   59.3%                  16.1%
 Catalog                               40.9%                   72.8%                  10.9%
 Detached label card                   34.0%                   39.8%                  8.1%
 Postcard                              57.4%                   60.3%                  17.4%
 Flyer                                 46.0%                   58.1%                  13.1%
 Newsletter                            49.1%                   73.4%                  9.8%

The model predicts that about 44 percent of Marketing letters are read by someone in the
household, with higher reader rates found for flats, postcards, flyers, and newsletters. Nearly
three quarters of catalogs generate a positive reaction. Households indicate that they are
more likely to respond to postcards than any other type of Marketing Mail.



Advertising Mail: Mail Mix Matters                                                                                              43
Report Number RARC-WP-19-003
TABLE OF CONTENTS                      EXECUTIVE SUMMARY                             OIG SYNOPSIS         RCF REPORT   APPENDIX

                                                                                                                       BACK to COVER




	           Other Mail Piece Characteristics

The impact of four other mail piece characteristics are examined: 1) whether the mail piece
was specifically addressed to a household member; 2) whether it contained a coupon; 3)
whether it came with a return envelope without paid postage; and 4) whether it came with
a return envelope with paid postage. Of these, the presence of a coupon is found to be the
most important, significantly raising the reading, positive reaction, and response rates, with
the strongest impact on the response rate. Interestingly, pieces with return envelopes are
less likely to be read but more likely to generate a response.

                   Table 26: Odds-Ratios for Other Mail Piece Characteristics

    Other Mail Piece Characteristics                   Reading           Reaction        Response
    Specifically addressed to HH members                1.394***          1.298***        1.084***
                                                        (0.017)           (0.016)         (0.020)
    Contains coupon                                     1.597***          1.958***        2.966***
                                                        (0.018)           (0.024)         (0.046)
    Return envelope without postage                     0.895***          1.104***        1.562***
                                                        (0.013)           (0.018)         (0.033)
    Postage paid return envelope                        0.709***          0.874***        1.076***
                                                        (0.009)           (0.012)         (0.024)
                                           Standard error in parentheses *** p<0.01, ** p<0.05, * p<0.1

                  Table 27: Predicted Reading, Reaction, and Response Rates
                               by Other Mail Piece Characteristics

                                           Predicted Value       Predicted Value     Predicted Value
    Mail Piece Characteristics
                                              Reading               Reaction            Response
    Not addressed to household                 39.2%                 53.2%               11.7%
    Addressed to household                     46.2%                 58.1%               12.5%
    Does not contain coupon                    42.4%                 54.0%                9.2%
    Contains coupon                            52.5%                 66.5%               21.3%
    No return envelope                         46.3%                 57.3%               11.6%
    Return envelope without postage            43.9%                 59.1%               16.3%
    Postage paid return envelope               39.0%                 54.8%               12.3%

Advertising Mail: Mail Mix Matters                                                                                                44
Report Number RARC-WP-19-003
TABLE OF CONTENTS                    EXECUTIVE SUMMARY                          OIG SYNOPSIS         RCF REPORT   APPENDIX

                                                                                                                  BACK to COVER




        E.	 Other Measures of Mail Mix

The analysis has shown that the greater the non-advertising portion of a household’s mail,
the greater is the probability that the household will read and react positively to any particular
piece of Marketing Mail. The next section includes further investigations of the impact of
mail mix by looking at specific components of non-advertising mail. In all cases, different
definitions of the mail mix do not meaningfully affect the coefficients on the other variables
(household demographics, mailer characteristics, and features of the mail piece) and
therefore only the mail mix coefficients are presented and discussed.

        Separate Impacts of First-Class Mail and Periodicals

Table 28 compares the odds-ratio coefficients from the single mail mix model (non-
advertising share) and a mail mix with separate shares for First-Class Mail and Periodicals
Mail. The results indicate that the presence of First-Class and Periodicals Mail have
approximately equal importance in raising household reading of Marketing Mail. An increase
in the First-Class Mail share has a statistically positive impact on household reaction and
likely response to advertising mail. The Periodicals share also has a positive impact on
reaction and response (the odds-ratios are greater than 1.0) but in neither case is the impact
as large as for First-Class Mail or statistically significant. Keep in mind that most of the non-
advertising mail received by households is First-Class Mail which is why the First-Class Mail
share odds-ratios are close to the odds-ratios of the non-advertising mail share.

        Table 28: Odds Ratios for Separate First Class and Periodicals Mail Mix

 Mail mix variable                           Reading            Reaction           Response
 Non-Advertising Share                       1.645***            1.239***           1.288***
                                             (0.048)             (0.039)            (0.056)
       First-Class Share                     1.634***            1.264***           1.308***
                                             (0.051)             (0.042)            (0.060)
       Periodicals Share                     1.717***            1.096              1.172
                                             (0.116)             (0.079)            (0.118)

                                      Standard error in parentheses *** p<0.01, ** p<0.05, * p<0.1



Advertising Mail: Mail Mix Matters                                                                                           45
Report Number RARC-WP-19-003
TABLE OF CONTENTS                    EXECUTIVE SUMMARY                          OIG SYNOPSIS         RCF REPORT   APPENDIX

                                                                                                                  BACK to COVER


         Separate Impacts of First-Class Correspondence and Transactions Mail

Given the importance of First-Class Mail in the household mail mix, the total First-Class Mail
share is decomposed into separate shares for First-Class correspondence mail and First-
Class transactions mail. Table 29 presents the results from this mail mix model and shows
that it is primarily the First-Class transactions share that affects household reading, reaction,
and response to Marketing Mail. Transactions mail strongly affects the probability of reading
having a positive reaction to, or considering responding to Marketing Mail. First-Class
correspondence mail (which includes correspondence from both individuals and businesses)
has a positive effect on household reading of Marketing Mail but its impact on reaction and
response is not significant.

     Table 29: Odds-Ratios for Separate Correspondence and Transaction Mail Mix

  Mail mix variable                          Reading            Reaction           Response
                                              1.645***           1.239***           1.288***
  Non-Advertising Share                       (0.048)            (0.039)            (0.056)
                                              1.118**            0.975              0.911
  First-Class Correspondence Share            (0.050)            (0.046)            (0.061)
                                              2.259***           1.579***           1.776***
  First-Class Transactions Share              (0.093)            (0.069)            (0.108)
                                              1.718***           1.097              1.171
  Periodicals Share
                                              (0.116)            (0.079)            (0.117)
                                      Standard error in parentheses *** p<0.01, ** p<0.05, * p<0.1

A final issue investigated in this study is the impact of packages received by households.
Packages were not included in the calculations of mail mix presented so far because in
most cases, packages are not received at the same time and place (e.g., the mailbox) as
other mail. Therefore, the interaction between package volumes and household treatment of
Marketing Mail is less direct than it is with other forms of non-advertising mail. Nevertheless,
given the growing importance of package deliveries to the Postal Service and households,
it is worth looking at whether the package share of mail affects household treatment of
advertising mail. To do this, the mail shares were recalculated including packages in the total
number of mail pieces received. These shares are shown in Table 30. Packages represent
only 3.7 percent of mail received by HDS households from 2013 to 2017, though this share
increased during this time. Household package volumes are highly skewed with most
households receiving zero or one package during their diary week while a few receive a high
volume of packages.
Advertising Mail: Mail Mix Matters                                                                                           46
Report Number RARC-WP-19-003
TABLE OF CONTENTS                      EXECUTIVE SUMMARY                              OIG SYNOPSIS       RCF REPORT   APPENDIX

                                                                                                                      BACK to COVER




     Table 30: HDS Volumes of Mail Received by Households including Packages

                                                   Pieces received per
                                                  household per week,
                                                 among household used             Share of all mail &
  Type of mail                                    in the reading model                packages
  Non advertising                                                  8.88                        38.1%
       Periodicals                                                 1.27                         5.4%
       Correspondence                                              2.82                        12.1%
       Transactions and other                                      3.94                        16.9%
       Packages                                                    0.85                         3.7%
  Advertising                                                     14.41                        61.9%
       First Class ads                                             0.72                         3.1%
       Marketing Commercial                                       11.03                        47.4%
       Marketing Nonprofit                                         2.66                        11.4%
  All mail & packages                                             23.29                      100.0%

Table 31 presents the odds ratios for different types of non-advertising mail. The key
takeaway is that the package share of mail has a significantly negative impact on household
reading, reaction, and response to advertising mail. However, for reasons discussed above,
this analysis warrants further investigation.

            Table 31: Odds-Ratios for Mail Mix Variables including Packages

 Mail mix variable                                  Reading           Reaction             Response
 Correspondence Share of mail and packages             1.156***           1.000               0.925
                                                       (0.053)            (0.049)             (0.064)
 Transactions Share of mail and packages               2.270***           1.578***            1.778***
                                                       (0.097)            (0.071)             (0.111)
 Periodicals Share of mail and packages                1.799***           1.140*              1.188*
                                                       (0.125)            (0.085)             (0.123)
 Packages Share of mail and packages                   0.416***           0.482***            0.528***
                                                       (0.033)            (0.039)             (0.064)
                                          Standard error in parentheses *** p<0.01, ** p<0.05, * p<0.1
Advertising Mail: Mail Mix Matters                                                                                               47
Report Number RARC-WP-19-003
TABLE OF CONTENTS                    EXECUTIVE SUMMARY   OIG SYNOPSIS   RCF REPORT   APPENDIX

                                                                                     BACK to COVER



Appendix:
Management’s
Comments




Advertising Mail: Mail Mix Matters                                                              48
Report Number RARC-WP-19-003
TABLE OF CONTENTS       EXECUTIVE SUMMARY                     OIG SYNOPSIS                       RCF REPORT              APPENDIX

                                                                                                                         BACK to COVER




                                                Contact us via our Hotline and FOIA forms.
                                                      Follow us on social networks.
                                                              Stay informed.

                                                         1735 North Lynn Street
                                                        Arlington, VA 22209-2020
                                                              (703) 248-2100

                                               For media inquiries, contact Agapi Doulaveris
                                                        Telephone: 703-248-2286




                    We conducted work for this white paper in accordance with the Council of the Inspectors General on
                         Integrity and Efficiency’s Quality Standards for Inspection and Evaluation (January 2012).