Newspaper Reports and Consumer Choice - CiteSeerX
-
Upload
khangminh22 -
Category
Documents
-
view
6 -
download
0
Transcript of Newspaper Reports and Consumer Choice - CiteSeerX
1
Newspaper Reports and Consumer Choice: Evidence from the Do Not Call Registry
Khim-Yong Goh,* Kai-Lung Hui, ** and I.P.L. Png*
February 2010
Abstract
* National University of Singapore; ** Hong Kong University of Science and Technology. Corresponding author: Ivan Png, NUS Business School, National University of Singapore, 15 Kent Ridge Drive, Singapore 119245, Singapore. We thank Hal Varian for pointing us to the federal Do Not Call registry, and Junhong Chu for valuable comments.
Despite annual U.S. expenditures on public relations exceeding $19.42 billion, businesses lack practical guidance about the effectiveness of publicity on sales. Here, we assembled a rich and novel dataset of newspaper reports and consumer sign-ups with the U.S. “do not call” (DNC) registry to gauge the impact of news reports on consumer behavior. We found robust evidence that news reports did increase DNC registration. Specifically, a one percent increase in the number of news reports would raise DNC registrations by 0.02 percent. This estimate provides an upper bound on the sales-response function for media publicity.
We also found that the politics of newspapers matters. Only reports in neutral newspapers (unaffiliated with either Democratic or Republican party) affected DNC registrations. Reports in Democratic and Republican newspapers had no effect.
2
1. Introduction
As providers of information, the media provide the key infrastructure for governance and a
well-functioning polity: “the central purpose of journalism is to provide citizens with accurate
and reliable information they need to function in a free society” (Pew Research Center 2009).
Historically, the media were so important to politics that they were recognized as the “fourth
estate,” the first three estates being the clergy, nobility, and commoners (Wikipedia 2010).
Owing to their broad and possibly targeted reach, the media have become an essential vehicle
for advertising and public relations.
The ostensible function of the media is to report the news and inform their audience.
However, it is widely believed that news reports go beyond conveying information to
influencing behavior, that is, that the media do not just report the news, but also, “make” or
“define” the news (Alsem et al. 2008). The media might influence behavior through pure
exposure by selecting what to report (Gentzkow and Shapiro 2004; George and Waldfogel
2006) as well as slant and spin in what is reported (Mullainathan and Shleifer 2005;
Gentzkow and Shapiro 2006; Gentzkow et al. 2006; Xiang and Sarvary 2007).
Most empirical studies of the impact of the media on individual behavior have
focused on electoral politics, an area in which the bulk of developments, news,
opinions/views, and policies are conveyed to individuals through the media (Strömberg 2004;
Gentzkow et al. 2006). The literature has documented that the media has a strong and robust
influence on voting behavior. For example, by substituting for other media such as
newspapers and radio, television decreased voter turnout (Gentzkow 2006); the entry of Fox
News in cable markets increased voter turnout and, more importantly, Republican support in
Presidential and Senate elections (DellaVigna and Kaplan 2007); Spanish-language local
television news raised Hispanic voter turnout (Oberholzer-Gee and Waldfogel 2009); in
Washington, D.C., exposure to newspapers raised voter turnout in a gubernatorial election but
3
media slant had no effect (Gerber et al. 2009); the closure of the Cincinatti Post reduced
voter turnout and campaign spending and resulted in incumbents becoming more likely to
win re-election (Schulholfer-Wohl and Garrido 2009).1
There have been relatively few studies of the impact of news reports on consumer
behavior outside of the political arena. While news reports are a primary channel of voter
information about politics, news reports are only one (arguably secondary) channel by which
information about non-political goods and services is communicated to consumers. The
major channels of communication are advertising and in-store promotions, supplemented by
“word of mouth” (Kotler and Keller 2008).
From the viewpoint of industry, the great attraction of media publicity as a way of
communicating information about goods and services is that it is, apparently, “free,” whereas
advertising and sales promotion are costly (Smith 2008). However, in 2008, U.S. public
relations expenditures on wages for managers and specialists alone totaled $19.42 billion, as
compared with wage expenditures on advertising managers of just $3.42 billion.2 Hence,
industry indeed spent considerable resources to cultivate the “free” publicity.
Given the substantial expenditures, it is imperative to assess the role of the media on
consumer behavior. However, despite the apparent attractiveness of media publicity and the
considerable expenditures, industry lacks guidance about the effectiveness of media publicity
or indeed how to measure publicity – whether in terms of number of media reports or the
content of the reports (Smith 2008). Further, while we know that media slant and bias affect
1 A separate but related literature studies how individuals demand slant in newspapers. Gentzkow and Shapiro (forthcoming) compiled an index of political slant based on phrases frequently used by Democrat vis-à-vis Republican members of Congress. They found that readers systematically preferred newspapers with slant, and newspapers responded to the readers’ preferences by tailoring their news reporting. This finding is generally consistent with a setting in which Bayesian consumers make quality inference of media based on the match of the media reports to their prior expectations (Gentzkow and Shapiro 2006), and with settings in which the media confer positive “preference externalities” (George and Waldfogel 2003, 2006). 2 Authors’ calculation based on data from the U.S. Bureau of Labor Statistics (see, also, Falconi 2006). U.S. advertising expenditure in 2008 was $141.7 billion (TNS Media Intelligence 2009).
4
voting behavior, we do not know whether they affect consumer behavior in a non-political
context.
Accordingly, our research seeks to address the following questions: To what extent
do media reports affect consumer choice? How important is the content of the report as
contrasted with the report itself? Do slant and bias matter in media reports of (non-political)
goods and services?
We address these research questions in a novel context – the federal Do Not Call
(DNC) registry. The U.S. federal government established the DNC registry as a free service
to help consumers avoid telemarketing. Consumers can sign up for the DNC registry either
by telephone or through the Internet. Since its inception in June 2003, the service has been
frequently reported in many newspapers. The DNC registry was supported by both major
political parties,3 and so media slant and bias should not matter in this context. Therefore, it
provides a neat setting in which to study the impact of news reports on consumer choice.
We assembled a rich and novel data-set from multiple sources. Unlike prior studies
of media impact which focused on overall slant of the media (and so technically looked at the
media itself rather than the reporting), we focus on actual reporting. Our data-set includes all
reports of the DNC registry in 117 newspapers that were circulated in more than 2,800 U.S.
counties. We matched the registration rates and news reports with an array of newspaper and
county demographic characteristics to compile a cross-section of DNC registrations,
newspaper reports, and county characteristics. We then related DNC registrations to various
measures of newspaper reports to identify their impact in a regression model.
The key issue in the econometric estimation was to distinguish the impact of news on
DNC registration from the reverse effect, that newspapers increased reporting of the DNC
registry because of an increase in registrations. In addition to including a battery of
3 See, for example, the Congressional votes reported in the Washington Post (2010).
5
demographic controls, we applied three different identification strategies. The first focused
on news reports published before the opening of the DNC registry. Clearly, these news
reports could not have been driven by consumer registrations. The next strategy applied a
difference-in-differences estimation strategy according to the extent to which the state added
telephone numbers from a state-level do not call registry to the federal registry. The
responsiveness of consumers to the federal registry would be lower in which a larger
proportion of telephone numbers were registered by the state. The third strategy applied
instrumental variables, using staff size and newsprint consumption as instruments for news
reports. Using the three approaches, we found robust evidence that newspaper reports did
increase DNC registration. The estimates implied that a one percent increase in the number
of news reports raised the number of DNC registrations by 0.02 to 0.06 percent.
The three approaches above allowed us to test the overall impact of newspaper reports
on consumer choice, and measure the “sales-response function” in our context. In addition,
we compiled various characteristics of the newspaper reports to explore the extent to which
the content of the newspaper reports affected DNC registration. We found little evidence that
the content of news reports affected DNC registration more than the report itself.
Finally, motivated by the studies of the impact of media reports on electoral politics,
we investigated the impact of media slant or bias on consumer behavior in the DNC context.
To do so, we compared the impact on DNC registration of reports in Democratic, Republican,
vis-à-vis unaffiliated newspapers, and found that only reports in unaffiliated newspapers
affected DNC registration.
2. Related literature
One of the celebrated 4Ps of marketing strategy is promotion, which comprises advertising,
publicity, and sales promotion. Marketing scholars have extensively studied advertising and
6
sales promotion (see, e.g., Fornell et al. 1985; Narasimhan 1988; Sasieni 1989; Blattberg et al.
1995; van Heerde et al. 2004; Chintagunta et al. 2006; Johnson and Myatt 2006; Sigué and
Chintagunta 2009). However, there has been relatively less attention to publicity despite its
obvious importance in marketing and consumer behavior.
The limited academic research on publicity has focused on its impact on consumer
attitudes, or the impacts of specialized publicity, specifically, consumer “word of mouth,” or
the impact of publicity in the political arena. For example, through a series of laboratory
experiments, Ahluwalia et al. (2000) found that the effect of publicity on consumer attitudes
toward a product is moderated by the extent of the consumers’ commitment. They
manipulated publicity using synthetic positive and negative newspaper articles, and found
that consumers who were more committed to the product were more responsive to positive
publicity and less responsive to negative publicity.
Similarly, postings on the Yahoo Movies Web site affected box-office sales, especially
in the weeks immediately after the movie opening (Liu 2006), and consumer reviews on the
Amazon and Barnes and Noble websites affected sales of the corresponding books (Chevalier
and Mayzlin 2006). However, the “word of mouth” studies provide little guidance as to the
impact of general publicity, as distinct from targeted reviews or synthetic product reviews in
a laboratory, on consumer choices.
In the political arena, Strömberg (2004) found that media coverage may inform
consumers about elections and may in turn affect government policies. Mullainathan and
Shleifer (2004), Gentzkow et al. (2006), Gentzkow and Shapiro (2006), and Xiang and
Sarvary (2007) studied the underlying structure of the media market – why the media have
incentives to produce slanted and biased news, how competition might affect media diversity
and news quality, and whether consumers demand slanted news.4
4 For a detailed review of this literature, see Gentzkow and Shapiro (2008).
7
Using micro-level data on Fox News penetration, DellaVigna and Kaplan (2007)
found that the introduction of Fox News increased Republican votes by 0.4 to 0.7 percentage
points and convinced 3 to 28 percent of its viewers to vote Republican. Similarly, in a field
experiment, Gerber et al. (2009) found that free subscriptions to either the pro-Democratic
Washington Post or the pro-Republican Washington Times increased voter turnout and
support for the Democratic candidate, suggesting that newspaper affected voter behavior
through exposure rather than slant.
At the end of 2007, a joint operating agreement between the Cincinnati Post and the
Cincinatti Enquirer expired, and the Post was closed. Schulhofer and Garrido (2009) found
that, in communities where the Post had previously provided relatively more coverage, the
closure of the Post led to fewer candidates competing, incumbents becoming more likely to
win, and lower voter turnout and campaign spending in municipal elections.
The studies of voting behavior, however, were generally founded on the overall slant
of the media. That is, they assumed that news reports would follow the political stance of the
medium, and their unit of analysis was effectively the medium rather than any particular
event or incident. While these assumptions might be appropriate in electoral politics, they
offer limited guidance on the effectiveness of the media as a publicity channel for an item
which is not obviously political.
Focusing specifically on economics news, Alsem et al. (2008) related coverage in two
Dutch newspapers to consumer and producer confidence. They applied the vector
autoregression (VAR) method and found that newspaper reports had little influence on both
consumer and producer confidence on the economy. They selected, however, only one day
of newspapers in each month, asked graduate students to rate the overall impression
presented by the newspapers on the overall economy, and used the ratings to measure
8
monthly media intensity. Hence, multiple sources of measurement errors could have reduced
the statistical significance of the newspaper influence.
Kalaitzandonakes et al. (2004) conducted an empirical analysis using an approach that
is perhaps closest to our study. In the context of processed foods, they found that consumer
purchase of taco shells was significantly related to the frequency of news reports of the recall
of products contaminated with genetically-engineered corn. Similar to our study, they
constructed the media coverage variable by transcribing articles from various U.S.
newspapers. Their study, however, did not account for possible endogeneity in newspaper
reports, specifically, that the extent of media coverage might have been caused by the drop in
consumer purchases and increased public interest.5
Generally, the main challenge in studying the impact of media publicity on consumer
behavior lies in compiling the coverage and reports relevant to a particular event or incident,
and in robustly identifying the effect given the possible endogeneity of media coverage. In
our study, we address these challenges by comprehensively transcribing all newspaper reports
of the DNC registry from June 1, 2003 to June 27, 2004, and addressing possible endogeneity
through multiple demographic controls and three different identification strategies.
3. Data
The U.S. Federal Trade Commission (FTC) opened the federal DNC registry on June 27,
2003. The registry applies to both interstate and intrastate telemarketing calls. Telemarketers
are required to remove numbers on the DNC registry from their call lists every 31 days.
5 Outside of the realm of consumer choice, newspaper reports have been shown to affect stock markets. “Abreast of the Market” is a daily column in the Wall Street Journal that reviews the U.S. stock market, and particularly, the stocks comprising the Dow Jones Industrial Average (DJIA). Tetlock (2007) found that, between 1984 and 1999, the sentiment of the column affected the DJIA index as well as trading volumes. Pessimistic columns led to downward pressure on the index, while unusually high or low pessimism in the column increased market trading volume.
9
Violators are subject to fines of up to $11,000 per offence. Registration for the DNC service
is free to consumers and can be completed by either telephone call or the Internet.
The FTC provided us with DNC registrations from June 27, 2003 (the beginning of
the registry) to January 6, 2006. We used the registration data up to the end of June 2004
(one year after the registry opened) as Americans tend to move during the summer (Hansen
1998), and so some of the DNC entries that were recorded after June 2004 could be “re-
registrations”.6 The FTC records showed registrations by redacted telephone number for
each area code and exchange, called NPA-NXX – e.g., (617) 363-xxxx – by date of
registration. While the U.S. comprised 3,128 counties in our data-set, it comprised 94,342
exchanges. On average, each county comprised about 47 exchanges, but there was
substantial variation in the number of exchanges per county.
To proceed, we used the North American Local Exchange NPA-NXX Database from
Quentin Sager Consulting to identify the counties served by each telephone exchange. For
telephone exchanges spanning multiple counties, we allocated the DNC registrations to the
respective counties based on the relative number of households in the counties as reported by
the U.S. Census 2000. We matched the DNC registrations to counties because we could not
identify the individuals or households who made the registrations, and so could only use their
geographical location – county – to characterize their demographic profiles. We obtained
average demographic data of county residents from the U.S. Census 2000, and matched these
data to the DNC registration data at the county level. 7
6 Further, since early 2005, several waves of phish emails targeted consumers to register their mobile phone numbers on non-existent DNC registries (and so could have deflected mobile phone numbers from the legitimate DNC registry), or falsely claimed that mobile phone numbers registered on the federal DNC registry would be subject to even more telemarketing calls (Federal Trade Commission 2005b, 2006). We were not able to identify the exact time period and geographical locations in which these malicious emails were circulated, but they could have affected DNC registration rates after 2004. 7 On reviewing the registration data, we identified an obvious discrepancy in Williamsburg, VA (FIPS 51830), which had just 3,619 households but a massive 17,127 registrations, or an average of 4.73 per household, which was 20.9 standard deviations higher than the mean registration rate of
10
Using the proprietary news database, Factiva,8 we searched for newspaper reports
including the words “do not call” together with either “FTC” or “Federal Trade Commission”
within the same report between June 1, 2003 and June 27, 2004.9, 10 Our search yielded
1,262 reports in 117 newspapers. For each report, we recorded the name of the publication,
the date, and various characteristics of the report – whether it mentioned number of people
registering, e.g., “The FTC estimates some 60 million phone numbers will be registered out
of 166 million residential phone numbers in America” (New York Post, June 27, 2003),
whether the report included a toll-free number or URL, and the lengths of the headline and
main body.
Figure 1 depicts the number of reports, by day, during the period of study. There
were two significant peaks in reporting. The first coincided with the opening of the federal
DNC registry on June 27, 2003. The second was associated with decisions on September 23
and 25, 2003, by U.S. District Judges in Colorado that forced the FTC to suspend the DNC
registry, but quickly followed by Congressional legislation to ratify the FTC’s authority and
allowing the DNC registry to resume on October 1, 2003 (Federal Communications Commission
2004). Both peaks in reporting are consistent with news reports of the DNC registry being
the result of particular events and exogenous to DNC registration.
<<Insert Figure 1 here>>
0.396 per household. The county with the next highest registration was Pitkin, CO (FIPS 8097) with just 1.62 per household. Accordingly, we omitted Williamsburg, VA from our analyses. 8 Factiva is a comprehensive newspaper database provided by Dow Jones. For more details of this database, see http://factiva.com/. 9 We required either “FTC” or “Federal Trade Commission” as well as “do not call” to screen out reports unrelated to the federal DNC registry. 10 Newspaper reports were collected from June 1, 2003, about a month before the opening of the DNC registry on June 27, 2003. We collected newspaper reports before the commencement of DNC registration so that we could capture the extensive media attention and buzz preceding the launch of the DNC registry. The ending time periods of registration data and newspaper reports were specified to be the same, i.e., the end of June 2004.
11
The impact of a newspaper report in a particular county would depend on the
circulation of the newspaper in that county. From the Audit Bureau of Circulation (ABC),
we obtained the circulation of each newspaper by county and day of week. We then
calculated the intensity of newspapers reports of DNC in county i as
( )i j ijj
I R c , (1)
where Rj is the number of reports of DNC in newspaper j, and cij is the circulation of
newspaper j in county i. Effectively, iI measures the total number of copies of DNC
newspaper reports circulated in county i during the studied period.11
We further compiled other newspaper characteristics from various sources to
supplement the news report data. From the Editor & Publisher International Year Book,
2004, we obtained annual newsprint consumption. From Bacon’s Newspaper Directory,
2005, we procured the total number of employees at each newspaper. From Web sites that
document newspaper politics,12 we extracted newspaper endorsements, if any, of candidates
in the 2004 U.S. presidential election.
Finally, we obtained the percent support, in each county, for the Republican candidate,
George W. Bush, and the Democratic candidate, John Kerry, in the 2004 U.S. presidential
election from Atlas of U.S. Presidential Elections.13
Prior to the opening of the federal DNC registry in June 2003, 27 states had already
established state-level “do not call” registries at various times ranging from November 1996
(Alaska) to April 2003 (California) (Varian et al. 2004). Subsequently, some states added the
telephone numbers on their state registry to the federal registry (Federal Trade Commission
2005a). In one set of analyses, we applied a difference-in-differences strategy, exploiting
11 The measure, Ii, is similar to the concept of “gross rating points”. 12 See, for example, http://www.dkosopedia.com/wiki/2004_Media_Endorsements, http://www.gwu.edu/ ~action/2004/cands/natendorse5.html [accessed January 25, 2010]. 13 See http://www.uselectionatlas.org/ [accessed January 25, 2010].
12
differences in the response of DNC registration to news reports according to the extent to
which telephone numbers on the state registry were added to the federal registry.
We compiled the percentage of state-registered numbers added to the federal registry
from an FTC (2003) report. However, the FTC report was incomplete in that it did not cover
all states that added their state registries to the federal DNC registry. For instance, it
excluded Pennsylvania, which subsequently added its state-registered telephone numbers to
the federal registry. Accordingly, we limited the difference-in-differences estimates to those
states without a prior state registry or those with a state registry covered by the FTC report.
Tables 1 and 2 present summary statistics and correlations of the data. Our data-set,
including characteristics of the newspapers and newspaper reports, demographics, and 2004
presidential election information, was specified at the county level.14
<<Insert Tables 1 and 2 here>>
4. Model
The basic model for analysis is a cross-sectional county-level regression of the form:
ik
kikii XIQ lnlnln , (2)
where iQ is the number of DNC registrations in county i, iI , as defined in (1) above, is the
number of DNC newspaper reports weighted by circulation in county i, kiX represents
demographic characteristics of the county, and i captures any county-level random errors.
We used a cross-sectional specification because most of our variables in the data-set,
14 We selected the county as the unit of analysis for two reasons. First, we could associate each registration with a county but not any individual newspaper or newspaper report. Second, a registration may be influenced by multiple newspaper reports (e.g., people read one newspaper at home and another at the workplace (Gentzkow and Shapiro 2008)), and so measurement errors would arise if we associated each DNC registration to a single newspaper report.
13
particularly the newspaper characteristics that we used as instrumental variables for DNC
news reports, did not vary across time. We specified all continuous variables in logarithms.15
In one identification strategy, we applied difference-in-differences among counties
according to the extent to which telephone numbers registered with a state-level DNC
registry were added to the federal registry. Referring to (2), suppose that
0 1 iS , (3)
where iS is the percentage of registrations added from the state registry (number of DNC
registrations added from state registry divided by total number of DNC registrations).
Substituting in (2), we have
ik
kikiiii XISIQ lnlnlnln 10 . (4)
Our hypothesis was that 1 0 , since the impact of newspaper reports should be smaller to
the extent that the state added state-registered telephone numbers to the federal registry.
5. Results
Telemarketing and consumer privacy are regulated by federal and state governments but not
county governments. Accordingly, to control for any possible regulations at the state level,
we included state fixed effects in all subsequent estimations and estimated standard errors
using the Huber-White adjustment, clustered by state. With the inclusion of state fixed
effects and robust clustered standard errors, our analysis focused on explaining differences in
registrations from state-wide averages.
15 Where necessary, we added one to the variable before applying the logarithms to avoid logarithms of zeroes. The double-log specification allowed us to directly interpret the estimated coefficients as elasticities. Moreover, empirical analyses of economic variables often fit better with the variables specified in logarithm (Wooldridge 2006: 197-200). Also, we included the number of households (in logarithm) as an explanatory variable, and so the dependent variable in (2) was equivalent to the county-level registration rate.
14
We first estimated an ordinary least squares (OLS) regression of DNC registrations on
various demographic factors that have been previously identified as affecting DNC
registrations. These covariates were the number of households, household size and income,
commuting time, unemployment rate and retail density (Varian et al. 2004; Goh et al. 2009).
As reported in Table 1, column (i), the results were generally consistent with previous
findings.16
<<Insert Table 3 here>>
We next estimated (2), as a baseline model of the impact of the number of news
reports (weighted by circulation), iI , on DNC registrations over the entire period of study.
As reported in Table 3, column (ii), the estimated coefficient of the number of news reports,
0.021 (±0.007), was positive and statistically significant. As the dependent variable (DNC
registrations) and the number of news reports were specified in logarithms, the coefficient
was the estimated elasticity. Accordingly, a one percent increase in newspaper reports of the
DNC registry, weighted by circulation in a county, was associated with a 0.021 percent
increase in DNC registrations in the county. To set the estimated elasticity in context, it was
about one-fifth of the average short-term advertising elasticity, 0.109, as reported in a meta-
analysis (Sethuraman and Tellis 1991). Our result suggests that the estimated impact of
newspaper reports is substantially smaller than that of advertising.
An obvious concern with this baseline analysis is that, although we included various
demographic controls, the results might be explained by reverse causation or some omitted
variables. While the news reports could have increased DNC registrations, it is also possible
16 The number of registrations should increase with number of households in the county. The likelihood that a person would pick up a telemarketing call would be lower in a larger household, and so large households may have a lower demand for DNC. High income or employed people may incur a higher time cost in receiving telemarketing calls, and so income may positively and unemployment may negatively correlate with DNC registrations. Finally, retail density may reflect the demand for shopping, and so in regions that have relatively more retail shops, the consumer demand for DNC may also be lower because they are more receptive of telemarketing.
15
that the newspapers were spurred by the overwhelming response to the opening of the DNC
registry to report the event. Also, some unobserved factors or incidents, such as adjustments
of telemarketing practices following the opening of the federal DNC registry, could affect
both registration and news reports contemporaneously. To rule out reverse causation and
omitted variables, we applied three different identification strategies.
5.1 Newspaper reports: Day 1
First, we focused on news reports published between June 1-26, 2003, and sign-ups with the
DNC registry on the opening day, June 27, 2003. Clearly, news reports published before
June 27 could not have been influenced by registrations on June 27. The sample size was
much smaller than in the estimate over the full period of study. The reason is that, as Figure
1 shows, news reporting was very subdued until the launch of the federal DNC registry.
Table 3, column (iii), reports the estimates. The coefficient of news reports, 0.061
(±0.024), was positive and statistically significant. Importantly, the elasticity, 0.061 (±0.024),
was substantially larger than the elasticity, 0.021 (±0.007), as estimated using news reports
and DNC registrations over the entire period of study. This difference is expected for two
reasons. First, the pre-launch news reports would have built up a pent-up demand for the
DNC registry when it opened. Second, the people who registered early would be those with
the strongest preference for the DNC registry, leaving those with weaker preference (and less
sensitive to news reports) to register later.
5.2 Newspaper reports: Difference-in-differences
Next, we applied a difference-in-differences strategy, exploiting differences between counties
according to the extent to which telephone numbers on the state-level DNC registry were
added to the federal registry. As a preliminary, we compared newspaper coverage in counties
16
with and without state-level registries. As reported in Table 4, among counties with state-
level registries, the mean number of news reports (weighted by circulation) was 135.18
(181.83), while, among counties without state-level registries, the mean was 231.46
(419.17). Apparently, newspapers did tend to provide more coverage of the federal DNC
registry where a corresponding state-level registry did not exist.
<<Insert Table 4 here>>
Table 3, column (iv), reports the estimate of model (4).17 The coefficients of the
various demographic controls were similar to those in the regression of model (1) (Table 3,
column (ii)). The coefficient of news reports, 0.023 (±0.011), was positive and significant,
and slightly larger than the estimate in model (1).
Importantly, the coefficient of the interaction between the number of news reports and
the percentage of numbers added by the state, -0.044 (±0.018), was positive and significant.
This was consistent with the hypothesis that 1 0 , i.e., to the extent that consumers had
their telephone numbers registered through the state registry, they would be less responsive to
news reports of the federal DNC registry. Based on the average state addition of 42.7%, the
elasticity of DNC registration with respect to news reports in a county with the average
registration through a state registry was 0.023 – 0.044 x 0.427 = 0.004, or about one-fifth of
the elasticity in a county with no state registry.
5.3 Newspaper reports: Instrumental variables
Our third identification strategy was to estimate the structural impact of newspaper reports on
DNC registrations using instrumental variable (IV) regressions. The ideal IVs in this context
would be factors that shifted newspapers’ reporting of DNC, but not DNC registration itself.
17 We continued to use the Huber-White robust sandwich estimator for the standard errors, clustered by state, to account for possible within-state correlations, as recommended by Bertrand et al. (2004) for difference-in-differences tests.
17
One plausible instrument for the number of news reports of DNC registration was the
staff size of the newspaper. A newspaper with a larger staff would be likely to employ more
journalists and reporters, and so produce more content, including reports of DNC registration.
However, it is unlikely that the setting up of the federal DNC registry, which is not a major,
sustained issue, would materially affect the staff size of a newspaper.
Another plausible instrument was the annual newsprint consumption of the newspaper,
adjusted by the newspaper’s circulation.18 A newspaper that consumed more newsprint,
relative to circulation, would have room for more news reports, including reports of DNC
registration. However, it is unlikely that the establishment of the federal DNC registry would
affect annual newsprint consumption.19
We performed the IV estimation using two-stage least squares (2SLS). Table 3,
column (v), reports the results with staff size as the instrument. The sample size was reduced
as we could not find information about the staff size of all newspapers. The coefficient of the
(logarithm of the) number of news reports, 0.023 (±0.007), was positive and significant.
Table 3, column (vi), reports the results with newsprint consumption as the instrument. The
coefficient of the number of news reports, 0.017 (±0.008), was positive and significant.
The estimated elasticities by 2SLS – 0.023 (±0.007) and 0.017 (±0.008) – were close
to the basic OLS estimate of 0.021 (±0.007). However, we should caution that the statistical
evidence of endogeneity was mixed. In a Hausman test, the residuals from the first-stage
regressions with staff size as the instrument were not significant (p = 0.901). Assuming that
news reports were endogenous, the staff size seemed to be a reasonably good instrument.
The Wald statistic was 1444, which well exceeded the rule-of-thumb of 10 (Stock and Yogo
18 The quantity of newsprint consumed already reflects total circulation, and so we weighted newsprint by normalized circulation in the respective counties. 19 The use of staff size and newsprint consumption as IVs is analogous to using supply-side cost shifters as IVs to estimate structural demand parameters. Supply-side cost shifters are appropriate instruments in empirical demand estimations (Berry 1994; Besanko et al. 1998).
18
2005). With newsprint consumption as the instrument, the Hausman test of the residuals
from the first-stage regressions was not significant (p = 0.248), while the Wald statistic was
492.09.
Upon considering the various specifications and diagnostics, we preferred the basic
model, (2), estimated by OLS over the whole period, as reported in Table 3, column (ii). This
specification was parsimonious, was easy to interpret, and provided an estimate of the
elasticity of DNC registration with respect to news reports, 0.021 (±0.07), which was
relatively conservative. The preferred specification was buttressed by estimates using three
identification strategies.
5.4 Content
Having established the impact of newspaper reports on DNC registration, a further issue was
the impact of the specific content of the news reports. To explore this question, we extracted
various characteristics of the newspaper reports – whether the report included a toll-free
number or URL, or mentioned the number of people registering, and the length of the
headline and main text.
It is important to note that content was conditional on the occurrence of a news report.
Only if a newspaper published a report on the DNC registry could it include the toll-free
number, URL, etc. Hence, as is evident from Table 2, the number of news reports and
content variables were highly collinear (all correlations exceeded 0.90). Accordingly, given
the limited sample size, it was infeasible to include both the number of news reports and the
measure of content in a single regression.
<<Insert Table 5 here>>
Hence, we separately estimated the impacts of the various measures of content on
DNC registration. As reported in Table 5, the results were very similar for the various
19
measures. The estimated elasticities with respect to news reports, inclusion of toll-free
number or URL, and mention of the number of registrations were very close, ranging from
0.024 ( 0.007) to 0.027( 0.009), as compared with 0.021 (± 0.007) for the number of news
reports. The estimated elasticities with respect to the length of the headline and main text
were not much different.
To gain deeper insight into the impact of the news contents, in the next set of analyses,
we partitioned the newspaper reports according to whether they did or did not include a
specific type of content – toll-free number, URL, or number of people registering. Using the
preferred specification but limiting the measure of news reports to those that did or did not
include the particular content, we then estimated the impact of news reports on DNC
registrations. We were not able to enter more than one measure of content simultaneously
because of high multicollinearity.
<<Insert Table 6 here>>
As reported in Table 6, in general, the results were similar to those from the previous
analyses. Apparently, there was little difference in the impact of news reports that did or did
not mention the toll-free number, URL, or the number of people registering.
Taken together, the results in Tables 5 and 6 suggest that the various dimensions of
the content of news reports had roughly similar impact on DNC registration as the fact of the
report itself. There being no difference in the impact of reports that did or did not mention
the number of people registering suggests that the impact of news reports lay in exposure
rather than influence.
However, we must caution that our news report and content variables were highly
collinear, and so the above inferences were drawn based on separate estimation of their
impacts. Integrating multiple measures of news content into a single nested estimation, with
the measures of content being conditional on the report, would be a good direction for future
20
research. This would provide more precise results on the issue of whether newspapers affect
behavior through exposure or influence.
5.5 Politics
Given that newspaper reports had a robust influence on DNC registration, and in light of
previous studies of the impact of newspapers on electoral politics, our final question was
whether the politics of the newspaper had any impact on consumer registration. To address
this question, we classified newspapers according to whether they endorsed the Republican,
Democratic, or neither of the major-party candidates in the 2004 presidential election. We
were able to identify the endorsements among 98 newspapers (or 84% of newspapers covered
in our analysis), of which 35 endorsed the Republican candidate, 44 the Democratic
candidate, and 5 neither major-party candidate.
Then, we re-constructed the newspaper report variables as
j
pij
pj
pi cRI )( , (5)
where p = Democratic, Republican, or Neutral. The model that we estimated was then
ik
kikNiN
RR
DiDi XIIIQ
i lnlnlnlnln . (6)
If the politics of the newspaper had no impact on consumer registration with the federal DNC
registry, then NRD . If, however, RD , NR , or DN , then newspapers of
different political slant had differential impacts on consumer registrations.
Table 1 reports summary statistics of the number of news reports from Republican,
Democratic, and neutral newspapers. Table 7 reports descriptive statistics of the content of
the newspaper reports by the politics of the newspaper. Neutral newspapers published more
reports about the DNC registry than Republican newspapers, which in turn, published more
reports than Democratic papers, but the differences were not statistically significant. Based
21
on the various measures of content, there seemed to be no statistically significant difference
in the content of reports in newspapers by political affiliation.
<<Insert Table 7 here>>
With regard to model (6), in the first estimate, we included the number of reports of
the federal DNC registry in Republican newspapers, the Republican vote, and the interaction
of the two political variables. As reported in Table 8, column (i), news reports in Republican
papers had no significant effect on DNC registrations, but Republican voters were more
likely to register for DNC. The effect of Republican voting preference was robust – it
continued to be significant in the full specification including reports in newspapers of all
political affiliations. This factor in DNC registration was not noticed in previous
demographic studies (Varian et al. 2004; Goh et al. 2009).
As reported in Table 8, column (ii), news reports in Democratic papers had no
significant effect on DNC registrations, while Democratic voters were less likely to register
for DNC (the latter effect, however, was not robust to the inclusion of reports in newspapers
of all political affiliations).
<<Insert Table 8 here>>
As reported in Table 8, column (iii), news reports in neutral newspapers (those that
endorsed neither major party candidate in the 2004 Presidential election) had a significant
and positive effect on DNC registration. The estimated elasticity, 0.020 (±0.008), was almost
identical to that in the basic regression, 0.021 (±0.007) (Table 3, column (ii)). This result was
similar to that in the full specification, including reports in newspapers of all political
affiliations (Table 8, column (iv)). However, the estimated elasticity of DNC registration
with respect to the number of reports in neutral newspapers was somewhat smaller, 0.016 (±
0.009), and statistically less significant.
22
Taken together, the results in Table 8 imply that in the context of DNC registrations,
the political bias of newspapers mattered – only reports in neutral newspapers had a
significant impact on DNC registrations. It is striking that we found such a difference in
DNC registrations, which was apparently a non-political issue that was supported by both
major parties (see, for instance, the Congressional votes reported in the Washington Post
2010). However, we should caution that our sample included only five neutral newspapers.
An obvious explanation would be that news reporting differed between newspapers
according to political affiliation. However, as reported in Table 7, there seemed to be no
(statistically significant) objective differences in coverage or content. Another possibility is
that neutral newspapers gave more prominent or more favorable coverage to the DNC
registry than Republican or Democratic newspapers. For instance, neutral newspapers might
have placed reports of the DNC registry on the front page as compared with an inside page.
Or, neutral newspapers might have reported about the DNC registry with a more pro-
government or pro-consumer slant.
In future research, it would be interesting to investigate whether neutral newspapers
report about commercial and government services with systematically different prominence
and slant as compared with Republican or Democratic newspapers. The findings would have
obvious implications for public policy and managerial practice.
6. Implications
Through a series of structural estimations exploiting information from multiple sources on
newspaper and county characteristics, we found robust evidence that newspaper reports of the
federal DNC registry affected consumer behavior. Our results suggest guidance on the
magnitude of the consumer response to newspaper publicity (i.e., the sales-response function).
While we found that newspaper reports had a significant economic impact, we must caution
23
that DNC registration was free of charge. Consumers would have been more responsive to
publicity about the DNC registry than to publicity about a commercially-provided item that
must be paid for. Intuitively, we expect the demand for commercial items to be less sensitive
to newspaper publicity, and we interpret our results to provide an upper bound to the sales-
response function.
An immediate avenue for further research would be to estimate the sales-response
functions for the various consumer products with respect to the various media. Using the
incremental margin (price less marginal cost), it would be straightforward to adapt the
Dorfman-Steiner condition to calculate the profit maximizing level of publicity expenditure
(Png and Lehman 2007: 208-211). Such analyses would shed light on whether the U.S.
expenditure of more than $19.42 billion per year on publicity exceeds or falls below the
profit-maximizing level.
Government regulation of media and particularly policy to ensure diversity of media
is typically justified by the need to ensure that the public has good information and so can
make well-informed choices (Gentzkow and Shapiro 2008). In an apparently non-political
context, we have found that the political neutrality of newspapers matters for consumer
choice. To the extent that our finding generalizes to commercially-provided goods and
services, it is important that there be sufficient circulation of neutral media so that consumers
would be well-informed. Our implication contrasts somewhat with political economy
research, the findings of which point public policy towards media diversity. Our results
suggest that increasing media diversity, if this means increasing the number of politically
affiliated media, would not affect consumer behavior.
For managers, our findings suggest that they must attend to the political bias of media
in directing their public relations effort. Instead of merely considering the reach (number of
impressions) and demographics of readership, marketing managers must also attend to the
24
politics of the media, and how the political orientation of the media interact with consumer
preferences. For example, our results show that only politically neutral newspapers affected
consumer decisions. To the extent that this generalizes to other goods and services,
marketing managers would achieve greater impact by focusing their publicity efforts at such
newspapers.
7. Limitations
This study is subject to several limitations. First, as we highlighted in Section 5.4, more data
are needed to overcome multicollinearity so as to distinguish the impact due to the various
dimensions of newspaper content. Second, our Factiva search yielded reports in only 117 of
1,197 U.S. newspapers audited by the Audit Bureau of Circulations. To the extent that
reports in various newspapers are correlated and these are correlated with reports in television
and other media – for example, through their publication of the same wire service reports –
our estimates might have exaggerated the effect of news reports on consumer registrations.
Third, the impact of newspaper reports could be more precisely estimated if we had
observations of consumer behavior at the level of individual newspapers. The challenge is
that it is difficult to observe the characteristics of those who registered for DNC, and to
associate each newspaper report with a particular individual/household.
Fourth, we did not have any data on newspaper readership other than circulation. If
the news reports exerted impact through self-selection of readership with characteristics
different from the overall county demographics, then our estimates could have been biased.
It would be helpful to control for such unmeasured heterogeneity among the studied counties
and newspapers, perhaps by augmenting the data-set with readership information.
8. Conclusions
25
Using observations of federal DNC registrations and comprehensive information on
newspaper reports and other data that we compiled from multiple sources, we found robust
evidence that newspaper reports significantly affected DNC registrations. Our estimate of
0.021 (±0.007) for the elasticity of the demand for DNC registration with respect to
newspaper reports established an upper bound for the sales-response function with respect to
media publicity. Further, we found that the political neutrality of newspapers mattered for
the impact of newspaper reports on DNC registration.
This study reveals that an important marketing function – media publicity – interacts
with politics. Even for an apparently non-political service, the impact of news reports
depended on the political neutrality of the media. This suggests that, for public policy, media
neutrality may be important for consumers to be well-informed even about apparently non-
political matters, and that media diversity as such may not necessarily help consumers. A
business implication is that managers should pay close attention to the political slant of any
media used as publicity channels.
In future research, it would be interesting to investigate whether media of different
political affiliation report about commercial and government services with systematically
different prominence and slant. The findings would have obvious implications for public
policy and managerial practice.
Another interesting extension to this research is to study newspapers that had changed
their political affiliations between the 2000, 2004, and 2008 presidential elections. Our
conclusions would be strengthened if the impact due to the news reports changed according
to the changing political affiliations of the newspapers. Also, future research into advertising
could study the impact of political affiliation of media on the sales-response to advertising.
Does the effectiveness of advertising also vary with the political bias of the media? This is
26
an obviously important question in marketing scholarship and an important direction for
future research.
27
References
Ahluwalia, R., R.E. Burnkrant, and H.R. Unnava, “Consumer Response to Negative Publicity: The
Moderating Role of Commitment,” Journal of Marketing Research, Vol. 37, No. 2, 2000, 203-
214.
Alsem, K.J., S. Brakman, L. Hoogduin and G. Kuper, “The Impact of Newspapers on Consumer
Confidence: Does Spin Bias Exist?” Applied Economics, Vol. 40, 2008, 531-539.
Berry, S.T., “Estimating Discrete-Choice Models of Product Differentiation,” RAND Journal of
Economics, Vol. 25, No. 2, 1994, 242-262.
Bertrand, M., E. Duflo and S. Mullainathan, “How Much Should We Trust Differences-in-Differences
Estimates?” Quarterly Journal of Economics, Vol. 119, No. 1, 2004, 249-275.
Besanko, D., S. Gupta, and D.C. Jain, “Logit Demand Estimation Under Competitive Pricing
Behavior: An Equilibrium Framework,” Management Science, Vol. 44, No. 11, 1998, 1533-1547.
Blattberg, R.C., R. Briesch, and E.J. Fox, “How Promotions Work,” Marketing Science, Vol. 14, No.
3, 1995, G122-G132.
Cameron, A. C. and P. K. Trivedi. Microeconometrics: Methods and Applications, 2005. New York:
Cambridge University Press.
Chevalier, J.A. and D. Mayzlin, “The Effect of Word of Mouth on Sales: Online Book Reviews,”
Journal of Marketing Research, Vol. 43, No. 3, 2006, 345-354.
Chintagunta, P.K., V. Kadiyali, and N.J. Vilcassim, “Endogeneity and Simultaneity in Competitive
Pricing and Advertising: A Logit Demand Analysis,” Journal of Business, Vol. 79, No. 6, 2006,
2761-2787.
DellaVigna, S. and E. Kaplan, “The Fox News Effect: Media Bias and Voting,” Quarterly Journal of
Economics, Vol. 122, No. 3, 1187-1234.
DellaVigna, S. and M. Gentzkow, “Persuasion: Empirical Evidence,” NBER Working Paper 15298,
August 2009, http://www.nber.org/papers/w15298 [Accessed January 20, 2010].
Falconi, T.M., “How Big Is Public Relations (and Why Does It Matter)? The Economic Impact of Our
Profession,” Institute for Public Relations, 2006,
http://www.instituteforpr.org/files/uploads/Falconi_Nov06.pdf [Accessed January 20, 2010].
Federal Communications Commission, “In the Matter of Rules and Regulations Implementing the
Telephone Consumer Protection Act of 1991: Annual Report on the National Do-Not-Call
Registry”, CG Docket No. 02-278, December 15, 2004.
Federal Trade Commission. “National Do Not Call Registry -Total Registrations As of September 16,
2003”, September 2003, http://www.ftc.gov/opa/2003/09/030917dncstates.pdf [Accessed January
25, 2010].
28
Federal Trade Commission. Annual Report to Congress for FY2003 and 2004 Pursuant to the Do Not
Call Implementation Act on Implementation of the National Do Not Call Registry, September
2005a, http://www.ftc.gov/reports/donotcall/051004dncfy0304.pdf [Accessed January 25, 2010].
Federal Trade Commission. The Truth about Cell Phones and the Do Not Call Registry, January 19,
2006, http://www.ftc.gov/opa/2006/01/dnccellphones.shtm [Accessed January 25, 2010].
Federal Trade Commission. The Truth about Cell Phones and the National Do Not Call Registry,
April 15, 2005b, http://www.ftc.gov/opa/2005/04/dnc.shtm [Accessed January 25, 2010].
Fornell, C., W.T. Robinson, and B. Wernerfelt, “Consumer Experience and Sales Promotion
Expenditure,” Management Science, Vol. 31, No. 9, 1985, 1084-1105.
Gentzkow, M.A. and J.M. Shapiro, “Media Bias and Reputation,” Journal of Political Economy, Vol.
114, No. 2, 2006, 280-316.
Gentzkow, M.A. and J.M. Shapiro, “Media, Education and Anti-Americanism in the Muslim World,”
Journal of Economic Perspectives, Vol. 18, No. 3, 2004, 117-133.
Gentzkow, M.A., “Television and Voter Turnout,” Quarterly Journal of Economics, Vol. 121, No. 3,
2006, 931-972.
Gentzkow, M.A., E.L. Glaeser and C.D. Goldin, “The Rise of the Fourth Estate: How Newspapers
became Informative and Why it Mattered,” in Corruption and Reform: Lessons from American’s
Economic History, Chap. 6, eds. E.L. Glaeser and C.D. Goldin, Chicago, IL, 2006, 187-230.
George, L.M. and J. Waldfogel, “The New York Times and the Market for Local Newspapers,”
American Economic Review, Vol. 96, No. 1, 2006, 435-447.
George, L.M. and J. Waldfogel, “Who Affects Whom in Daily Newspaper Markets?” Journal of
Political Economy, Vol. 111, No. 4, 2003, 765-784.
Gerber, A.S., D. Karlan and D. Bergan, “Does the Media Matter? A Field Experiment Measuring the
Effect of Newspapers on Voting Behavior and Political Opinions,” American Economic Journal:
Applied Economics, Vol. 1, No. 2, 2009, 35-52.
Goh, K.Y., K.L. Hui, and I.P.L. Png, “Consumer Heterogeneity, Privacy, and Personalization:
Evidence from the Do-Not-Call Registry”, October 1, 2009, http://ssrn.com/abstract=1119166.
Johnson, J.P. and D.P. Myatt, “On the Simple Economics of Advertising, Marketing, and Product
Design,” American Economic Review, Vol. 96, No. 3, 2006, 756-784.
Kalaitzandonakes, N., L.A. Marks, and S.S. Vickner, “Media Coverage of Biotech Foods and
Influence on Consumer Choice,” American Journal of Agricultural Economics, Vol. 86, No. 5,
2004, 1238-1246.
Kotler, P. and K.L. Keller, Marketing Management, Prentice Hall, 2008.
Liu, Y. “Word of Mouth for Movies: Its Dynamics and Impact on Box Office Revenue,” Journal of
Marketing, Vol. 70, No. 3, 2006, 74-89.
29
Mullainathan, S. and A. Shleifer, “The Market for News,” American Economic Review, Vol. 95, No. 4,
2005, 1031-1053.
Narasimhan, C., “Competitive Promotional Strategies,” Journal of Business, Vol. 61, No. 4, 1988,
427-449.
Oberholzer-Gee, F. and J. Waldfogel “Media Markets and Localism: Does Local News en Español
Boost Hispanic Voter Turnout?” American Economic Review, Vol. 99, No. 5, December 2009,
2120-2128.
Pew Research Center, “Principles of Journalism,” Project for Excellence in Journalism, 2009,
http://www.journalism.org/resources/principles [Accessed January 19, 2010].
Png, I. and D. Lehman, Managerial Economics, Malden, MA: Blackwell, 2007.
Sasieni, M.W., “Optimal Advertising Strategies,” Marketing Science, Vol. 8, No. 4, 1989, 358-370.
Schulhofer-Wohl, S. and M. Garrido, “Do Newspapers Matter? Evidence From the Closure of The
Cincinnati Post”, Discussion Paper in Economics #236, Woodrow Wilson School, Princeton
University, 2009.
Sethuraman, R. and G. Tellis, “An Analysis of the Tradeoff Between Advertising and Price
Discounting,” Journal of Marketing Research, Vol. 28, No. 5, 1991, 160-174.
Shea, John, “Instrument Relevance in Multivariate Linear Models: A Simple Measure”, Review of
Economics and Statistics, Vol. 79, No. 2, May 1997, 348–352.
Sigué, S.P. and P. Chintagunta, “Advertising Strategies in a Franchise System,” European Journal of
Operational Research, Vol. 198, No. 2, 2009, 655-665.
Smith, B.G., “Representing PR in the Marketing Mix: A Study on Public Relations Variables in
Marketing Mix Modeling,” Institute for Public Relations, 2008,
http://www.instituteforpr.org/files/uploads/BG_SmithKetchum.pdf [Accessed January 20, 2010].
Stock, J.H. and M. Yogo, “Testing for Weak Instruments in Linear IV Regression”, in D.W.K.
Andrews and J.H. Stock, eds., Identification and Inference for Econometric Models: Essays in
Honor of Thomas Rothenberg, Cambridge: Cambridge University Press, 2005, 80–108.
Strömberg, D. “Radio’s Impact on Public Spending,” Quarterly Journal of Economics, Vol. 119, No.
1, 2004, 189-221.
Tetlock, Paul C., Giving Content to Investor Sentiment: The Role of Media in the Stock Market,
Journal of Finance, Vol. 62 No. 3, June 2007, 1139-1168.
TNS Media Intelligence. TNS Media Intelligence Reports U.S. Advertising Expenditures Declined
4.1 Percent in 2008, May 4, 2009, http://www.tns-mi.com/news/05042009.htm [Accessed January
20, 2010].
Wooldridge, J.M. Introductory Econometrics: A Modern Approach, 3d edition. Thomson
Southwestern, Mason, OH, 2006.
30
Xiang, Y. and M. Sarvary, “News Consumption and Media Bias,” Marketing Science, Vol. 26, No. 5,
2007, 611-628.
Van Heerde, H.J., P.S.H. Leeflang, and D.R. Wittink, “Decomposing the Sales Promotion Bump with
Store Data,” Marketing Science, Vol. 23, No. 3, 2004, 317-334.
Washington Post. Votes Database, 2010,
http://projects.washingtonpost.com/congress/108/senate/1/votes/365/ [Accessed, January 20,
2010].
Wikipedia, “Fourth Estate”, http://en.wikipedia.org/wiki/Fourth_Estate [Accessed, January 27, 2010].
Wooldridge, J.M. Introductory Econometrics: A Modern Approach, 3d edition. Thomson
Southwestern, Mason, OH, 2006.