Content uploaded by Seth C. Lewis
Author content
All content in this area was uploaded by Seth C. Lewis on Oct 30, 2015
Content may be subject to copyright.
Journalism & Mass Communication Quarterly
2015, Vol. 92(3) 723 –743
© 2015 AEJMC
Reprints and permissions:
sagepub.com/journalsPermissions.nav
DOI: 10.1177/1077699015589187
jmcq.sagepub.com
Survey Method
Are Demographics
Adequate Controls for
Cell-Phone-Only
Coverage Bias in Mass
Communication Research?
Brendan R. Watson1, Rodrigo Zamith2, Sarah
Cavanah1, and Seth C. Lewis1
Abstract
Cell-phone-only (CPO) households differ along key variables from non-CPO
households, creating potential coverage biases in landline-only random-digit-dialing
(RDD) surveys. Researchers have attempted to correct for this by weighting their data
based on demographic differences. Previous research, however, has not examined
CPO coverage biases in media-use surveys—an important oversight as cell phone
use is itself a media choice. This article presents a secondary analysis of Pew’s 2012
media consumption survey and concludes that demographics alone are not adequate
controls for the CPO bias in media-use surveys.
Keywords
coverage bias, cell phone, mobile, surveys, online surveys, sampling, methodology
The increasing number of cell-phone-only (CPO) households poses a challenge for
random-digit-dialing (RDD) telephone surveys. Among those challenges is the poten-
tial for coverage bias resulting from excluding CPO households, which researchers
have found to differ along key variables from the general population. Survey research-
ers want to understand the differences between CPO and non-CPO respondents to
reduce biases associated with undersampling CPO households, either by weighting
1University of Minnesota-Twin Cities, Minneapolis, USA
2University of Massachusetts-Amherst, USA
Corresponding Author:
Brendan R. Watson, School of Journalism & Mass Communication, University of Minnesota Twin Cities,
111 Murphy Hall, 206 Church St. SE, Minneapolis, MN 55455, USA.
Email: brwatson@umn.edu
589187JMQXXX10.1177/1077699015589187Journalism & Mass Communication QuarterlyWatson et al.
research-article2015
at University of Minnesota Libraries on October 26, 2015jmq.sagepub.comDownloaded from
724 Journalism & Mass Communication Quarterly 92(3)
survey data or including CPO households in the sample (or both). Previous research
on this topic has primarily focused on political and health surveys (Ansolabehere &
Schaffner, 2010; Blumberg, Ganesh, Luke, & Gonzales, 2013; Blumberg & Luke,
2009; Keeter, Kennedy, Clark, Tompson, & Mokrzycki, 2007; Lavrakas, Shuttles,
Steeh, & Finberg, 2007; Link, Battaglia, Frankel, Osborn, & Mokdad, 2007;
Mokrzycki, Keeter, & Kennedy, 2009). These studies concluded that the coverage bias
associated with not including a representative number of CPO households arises from
demographic differences between CPO and non-CPO respondents, which the research-
ers involved suggested could be controlled for by weighting the data based on those
demographic differences.
As the percentage of CPO households has grown (along with the associated cover-
age bias), the survey research field has reached consensus that landline-only surveys
cannot provide adequate coverage of the population and that telephone surveys must
include cell phone sampling frames (Hill, Tchernev, & Holbert, 2012). Following this
consensus, a number of sources of secondary data used in mass communication
research now include cell phones. For example, Pew began using cell phone sampling
frames in 2006 (Pew Research Center for the People and the Press, 2012a), and GfK
(formerly Knowledge Networks),1 which supplies lists to major surveys including the
American National Election Studies Panel and the National Annenberg Election
Survey, began using address-based sampling to help compensate for the CPO cover-
age bias in 2008, with full implementation in 2009 (Dennis & DiSorga, 2009). We
reviewed 17 journals affiliated, or sponsored by, the Association for Education in
Journalism and Mass Communication (AEJMC) from 2000 to 2013 to see whether the
mass communication discipline has widely adopted cell phone sampling frames.2 Of
the 104 articles we identified that used sampling methods that could potentially be
affected by the inclusion or exclusion of CPO households, 77% did not provide enough
detail to judge whether a cell phone sampling frame was used in the study. For exam-
ple, studies might have simply said, “A random national sample of 400 adults partici-
pated in telephone survey” (Avery, 2010) or “A national telephone survey using a
computer-assisted telephone interviewing system was conducted” (Bobkowski, 2009).
Scholars also continue to publish secondary analyses of Pew and other sources of sec-
ondary data that predate the use of cell phone sampling frames, without addressing the
potential coverage bias associated with a landline-only sample (Hmielowski, 2012;
Rittenberg, Tewksbury, & Casey, 2012; L. Wei & Hindman, 2011). The fact that the
majority of articles do not directly address the potential coverage bias associated with
CPO households suggests that this potential bias has received inadequate attention
within the mass communication literature (Hill et al., 2012).
The issue of coverage bias associated with CPO households, however, should be of
particular concern to mass communication researchers because cell phone use reflects
a communication and media-use choice. Previous technology adoption research sug-
gests the CPO media-use choice is also likely correlated with other patterns in CPO
households’ media preferences, particularly the use of a mobile phone to access news
(Chan-Olmsted, Rim, & Zerba, 2013). That is, the different media preferences between
CPO and non-CPO households may not necessarily be controlled for based on
at University of Minnesota Libraries on October 26, 2015jmq.sagepub.comDownloaded from
Watson et al. 725
demographic weighting alone. Thus, the potential coverage bias arising from CPO
households should be of unique concern to those using surveys to investigate respon-
dents’ media habits. Yet there has been insufficient research on the potential for CPO
coverage bias within mass communication research concerning audiences’ media-use
habits.
The present study fills that gap in the research by conducting a secondary data
analysis of the Pew Research Center’s 2012 biennial media consumption survey (Pew
Research Center for the People and the Press, 2012b). The Pew study uses dual-frame,
probability samples of both landline and cell phone numbers. The latter includes CPO
households, which can be further parsed out. Thus, we are able to estimate the signifi-
cant media-use differences of CPO and non-CPO households, and then examine
whether these differences can be controlled for based on demographics alone in stud-
ies that do not contain adequate samples of CPO households.
The present study first estimates what the potential coverage biases associated with
CPO households are. Then, we control for key demographic differences, drawn from
the literature, between CPO and non-CPO households to assess whether they account
for the difference in news media use—the use of television (TV), radio, newspapers,
and Internet as a source for news—between CPO and non-CPO households. If demo-
graphics serve as adequate controls for the coverage bias associated with CPO house-
holds after controlling for the demographic differences between the two groups, CPO
status should not be a significant predictor of media use. Conversely, if CPO status
remains a significant predictor of media use, it would suggest that demographics alone
are inadequate controls. While this study analyzes telephone survey data, the investi-
gation is also relevant to online survey data, as the same logic that underlies weighting
of telephone survey data is used to justify weighting online survey data to make them
“representative” (Correa, Hinsley, & de Zúñiga, 2010; Curran, Iyengar, Lund, &
Salovaara-Moring, 2009; de Zúñiga, Jung, & Valenzuela, 2012).
Literature Review
CPO Households
Coverage bias occurs when some members of a population being studied are not in the
sampling frame and those excluded members differ significantly from those within the
population frame, creating a bias in the population parameter (Dillman, Smyth, &
Christian, 2008). The rise in the number of CPO households—in 2012, the National
Center for Health Statistics estimated that 38.2% of U.S. households relied exclusively
on cell phones, up four percentage points from the previous year (Blumberg et al.,
2013; Blumberg, Luke, Ganesh, Davern, & Boudreaux, 2012)—creates a significant
challenge for traditional landline-only, RDD surveys. Among the challenges involved
in including CPO households in surveys are increased cost, federal regulations requir-
ing that cell phone numbers be dialed manually, the increased likelihood of reaching a
minor when dialing a cell phone number, and a greater percentage of users who screen
calls or do not answer them (Pew Research Center for the People and the Press, 2011).
at University of Minnesota Libraries on October 26, 2015jmq.sagepub.comDownloaded from
726 Journalism & Mass Communication Quarterly 92(3)
Thus, many studies are limited in their ability to include respondents who rely primar-
ily on cell phones and opt instead to weight responses to improve the representative-
ness of the sample.
Data are typically weighted to match the population and demographic estimates
from the U.S. Census Bureau and patterns of CPO and combined landline/cell phone
households. In particular, CPO households tend to be younger, more urban, and gener-
ally non-White (Keeter, 2006; Keeter et al., 2007; Link et al., 2007; Mokrzycki et al.,
2009). CPO households are also more likely to have lived in their current residence for
less than a year, have rented their home, have lower incomes, be unmarried, and be
childless (Ansolabehere & Schaffner, 2010). That is, CPO households are more geo-
graphically mobile and have fewer ties to a community, the latter of which is also
associated with relying less on a newspaper for local public affairs information (Emig,
1995).
Previous studies of the coverage bias associated with CPO households have primar-
ily focused on the implications for political surveys (Ansolabehere & Schaffner, 2010;
Keeter, 2006; Keeter et al., 2007; Mokrzycki et al., 2009) and health surveys (Blumberg
& Luke, 2009; Link et al., 2007). For example, these studies have found that individu-
als who live in CPO households are less likely to vote (Ansolabehere & Schaffner,
2010) and more likely to vote for Democratic candidates (Mokrzycki et al., 2009). In
the health literature, CPO respondents have been found to be more likely to engage in
risky health behaviors (Link et al., 2007). A key finding in these studies, however, is
that this variation can be primarily attributed to demographic differences among CPO
and non-CPO households—that is, that CPO households are more likely to be urban,
younger, non-White, childless, and have lower incomes. Thus, weighting the data
would allow a researcher to account for the coverage bias associated with CPO
households.3
Although the problems posed by CPO households for survey researchers have been
well studied within political and health contexts, the issue has received almost no
attention by mass communication researchers. The exception is a study by Hill and
colleagues (2012) that examined differences in political media use between survey
participants reached via landline versus survey participants reached by cell phone.
(The study did not separate out CPO households.) Hill et al. measured media use based
on a 5-point scale that ranged from never (1) to all of the time (5). They found that
compared with the landline sample, the cell phone sample more frequently accessed
web news and more frequently watched MSNBC, political satire shows, and general-
interest satirical TV shows (i.e., “The Simpsons”). However, once age was introduced
as a control, they concluded those differences were primarily due to the fact that the
cell phone sample was significantly younger. Had Hill et al. separated out individuals
that could only have been reached by cell phone, they might have found additional
differences in media use between a landline and CPO sample. In addition, it is worth
noting that their survey was conducted in late 2009 and early 2010. As previously
indicated, the percentage of CPO households in the United States has increased dra-
matically in the last several years, and there may be differences between earlier and
later CPO adopters.
at University of Minnesota Libraries on October 26, 2015jmq.sagepub.comDownloaded from
Watson et al. 727
Thus, this study heeds Hill et al.’s (2012) call for additional research exploring the
implication of a growing number of CPO households for media consumption surveys.
Understanding media use is important not only for understanding the changing prefer-
ences among audiences but also for understanding how individuals access information
about topics like politics (Dahlgren, 2009; Graber, 2009; McLeod, Scheufele, & Moy,
1999) and health (Noar, 2006).
Technology Adoption and Use
Previous literature on the adoption of cell phones and related technologies suggests
there are distinct characteristics and motivations that distinguish CPO households
from non-CPO households and that these differences extend beyond demographic
variation (Leung & Wei, 2009; Rice & Katz, 2003; van Biljon & Kotzé, 2007). The
literature also suggests that cell phone adoption is correlated with other types of media
use, such as reading news on a mobile device (Chan-Olmsted et al., 2013). If this were
the case, then simply weighting CPO respondents based on demographic characteris-
tics would likely continue to produce biased estimates of media use.
Rice and Katz’s (2003) survey of American households found that early adopters of
the mobile phone were younger than late adopters and that early adopters were less
likely to be married, which would suggest that demographics might be adequate con-
trols for the CPO coverage bias. Technology adoption models, however, suggest there
are influences beyond demographics that affect whether one adopts a particular tech-
nology, including the cell phone. Previous studies have shown that adoption of cell
phones is related to various factors that are both intrinsic and extrinsic to the individual
user, thus extending beyond demographics. Extrinsic factors include access to the
infrastructure that supports use, such as access to a cell phone and to cell phone ser-
vice, and design factors of the technology, such as whether it is attractive, easy to
learn, and easy to use (Lu, Yu, Liu, & Yao, 2003). Intrinsic factors include the per-
ceived ease of use of the technology, perceived usefulness of the technology, enjoy-
ment derived from using the technology, one’s desire to learn new skills, peer influence,
and even one’s perception of whether using mobile technologies makes personal data
less secure (Conci, Pianesi, & Zancanaro, 2009).
While these studies have investigated the general use of cell phones, the decision to
adopt the cell phone as one’s only form of telephone access is likely to be related to
intrinsic and extrinsic factors. For example, if one does not enjoy talking on a cell
phone, finds a cell phone difficult to use, or does not have access to good cell service,
one is not likely to rely on cell phones for sole telephone access. Furthermore, mobile
phone usage is positively associated with other types of media use, including using
cell phones for mobile web browsing and news consumption (R. Wei, 2008). Mobile
news consumption is also associated with perceived usefulness and perceived ease of
use, factors that also affect cell phone adoption (Conci et al., 2009). Indeed, as Chan-
Olmsted and colleagues (2013) note, “It seems that mobile news adopters have certain
distinctive media usage patterns and news preferences” (p. 127). Some of those prefer-
ences may include the use of Twitter for news consumption; of Twitter users in the
at University of Minnesota Libraries on October 26, 2015jmq.sagepub.comDownloaded from
728 Journalism & Mass Communication Quarterly 92(3)
United States, 85% report getting news on Twitter via a mobile device, compared with
64% of Facebook news consumers (Mitchell & Guskin, 2013). Nevertheless, the lit-
erature is still emerging in articulating how news use via social media—for example,
using social networks to “friend” and follow journalists or otherwise gather and dis-
seminate news—might be connected to mobile media adoption and activity in particu-
lar (see, for example, Weeks & Holbert, 2013).
Overall, the technology adoption literature suggests that cell phone users differ
from non-cell phone users—and presumably, CPO and non-CPO respondents—based
on factors that go beyond demographics. Furthermore, previous studies of mobile
media use suggest that these technology adoption factors are associated with other
types of media use. Thus, one would expect that weighting survey data of CPO and
non-CPO respondents based solely on demographic characteristics would not com-
pletely correct the potential coverage bias associated with not including a representa-
tive sample of CPO respondents in a study of individuals’ media use.
Hypotheses and Research Questions
The previously cited literature provides two key sets of assumptions. First, that there
are significant differences between CPO and non-CPO households (Keeter, 2006;
Keeter et al., 2007; Link et al., 2007; Mokrzycki et al., 2009) and that those differences
are not independent of media-use choices. Second, the literature on technology adop-
tion and use indicates these differences may not be entirely accounted for by demo-
graphic differences (Rice & Katz, 2003; van Biljon & Kotzé, 2007; R. Wei, 2008). We
therefore hypothesize the following:
Hypothesis 1 (H1): There will be a significant coverage bias associated with CPO
households in estimating individual media use; that is, CPO status will be a signifi-
cant predictor of media use.
Hypothesis 2 (H2): After controlling for demographic differences between CPO
and non-CPO households, CPO status will remain a significant predictor of media
use.
Method
Data Source
Data were obtained from the Pew Research Center for the People and the Press’ (2012)
Media Consumption Survey. This data set was chosen because the Pew biennial media
consumption survey is a frequently cited source of media-use data for researchers in
the social sciences. (To wit: A search for “Pew Biennial Media Consumption Survey”
in Google Scholar returned more than 2,000 results.) The 2012 Media Consumption
Survey data were collected by Princeton Survey Research Associates International on
behalf of Pew. Between May 9 and June 3, 2012, 3,003 respondents were interviewed
by telephone. These individuals were contacted through RDD of both landline
at University of Minnesota Libraries on October 26, 2015jmq.sagepub.comDownloaded from
Watson et al. 729
(n = 1,801) and cell phone (n = 1,202) numbers using samples provided by Survey
Sampling International.
Interviews were conducted in both Spanish and English. For the landline respon-
dents, the interviewer requested to speak with the youngest male or female—depending
on a random rotation—currently present in the home. With the cell phone sample,
interviews were conducted with whoever answered the phone. All interviewees had to
be aged 18 years or older to participate. Response rates were 11% for landline numbers
and 7% for cell phone numbers (American Association for Public Opinion Research
[AAPOR], Response Rate 3 [RR3]).
As with any survey, the Pew’s media consumption survey contains both measure-
ment and representation error (Groves & Lyberg, 2010). With regard to measurement
error, self-reported media use often contains a degree of measurement error (e.g.,
respondents often overestimate news media use; Prior, 2009). With regard to represen-
tation error, a low response rate raises particular concerns about nonresponse error.
The population in the Pew survey skews older, better educated, wealthier, and more
female than the population as a whole. To the extent that the goal of this article is to
provide an accurate estimate of media use, these sources of error would be more prob-
lematic. Such estimation, however, is not the goal of this article. Rather, we seek to
explore the differences in media use between CPO and non-CPO households within
the survey’s sample and whether those differences can be explained by demographic
differences alone.
Independent variables. Based on the previous literature on CPO coverage bias in sur-
vey research (Ansolabehere & Schaffner, 2010; Keeter, 2006; Keeter et al., 2007; Link
et al., 2007; Mokrzycki et al., 2009) in addition to whether or not the respondent lived
in a CPO household, our analysis focused on seven demographic variables: age, edu-
cation level, income level, marital status, parental status, race and ethnicity, and sex.
Age was measured continuously, with an endpoint of 97 or older. The education level
variable was measured ordinally, ranging from 0 (less than high school) to 7 (post-
graduate or professional degree). The income level variable was also measured ordi-
nally, ranging from 0 (less than US$10,000) to 8 (US$150,000 or more). Marital status
was measured through six nominal categories; this variable was recoded to reflect
whether an individual had never married (0) or been married at some point (1), which
included widowers and those who were divorced. Parental status recorded whether a
respondent was the parent or guardian of children younger than 18 currently living in
his or household. Race and ethnicity were determined by Pew through a combination
of questions, and we recoded to reflect whether the respondent was either White and
non-Hispanic (1) or in another racial or ethnic group (0). Sex was measured as either
male (1) or female (0). Last, the primary mode of telephone communication in the
home was determined by Pew; this response was recoded into a CPO status variable
measuring whether a respondent lived in a CPO (1) or non-CPO (0) household.
Dependent variables. We analyzed 10 media-use variables from the 2012 biennial
media consumption survey (Table 2).4 Each author independently reviewed the survey
at University of Minnesota Libraries on October 26, 2015jmq.sagepub.comDownloaded from
730 Journalism & Mass Communication Quarterly 92(3)
questionnaire, and the variables were selected through group consensus. Specifically,
we were looking for questions that reflected respondents’ use of different types of
media in the context of news. This led us to questions about the use of (a) newspapers,
(b) TV, (c) radio, and (d) networked technologies and applications, thus enabling us to
consider both “legacy” media as well as “new” media. Furthermore, we focused on
questions about general media use, rather than the consumption of specific programs
(e.g., NBC Nightly News), because the central contention of this article is that cell
phone use is itself a media-use choice that should correlate with other general media
uses. Last, to make our analysis more parsimonious, we focused the analysis on
dichotomous media-use questions (e.g., “Do you happen to read any daily newspaper
or newspapers regularly, or not?”).
Because the purpose of this study is to examine differences between the CPO and
non-CPO samples, not necessarily to generalize media use to the general population,
we used Pew’s unweighted survey responses.
Results
General Sample Characteristics
A total of 2,490 respondents provided responses to all of the questions associated with
the independent variables. The average respondent, as measured by the sample’s
median, was aged 52 years, had received some college education but no degree, earned
between US$40,000 and US$50,000 per year, was married, did not have a child
younger than 18 years in the household, was White and non-Hispanic, and was female.
As shown in Table 1, respondents in CPO households were considerably younger than
non-CPO households, less educated, had a lower annual income, were less likely to
have been married at some point, were more likely to have children younger than 18
in the household, were less likely to be White and non-Hispanic, and were more likely
to be male. To assess each media-use variable, a subsample was taken that included
only respondents who provided a response to the respective question in addition to all
of the independent variables; thus, there was some variation in the sample characteris-
tics for each media-use variable.
CPO as a Single Factor
The first hypothesis predicted that there would be a significant coverage bias associ-
ated with CPO households in estimating individual media use. To assess this hypoth-
esis, a series of generalized linear models using a logit link were fitted to each question,
using CPO status—whether someone lived in a CPO household or not—as the lone
predictor. As shown in Table 2, CPO status was found to be a statistically significant
predictor for 6 of the 10 media-use variables: reading a daily newspaper, watching
news on TV, listening to news on the radio, reading news on Twitter, following jour-
nalists on Twitter, and reading news on general social networking sites. Thus, the first
hypothesis received partial support.
at University of Minnesota Libraries on October 26, 2015jmq.sagepub.comDownloaded from
Watson et al. 731
Table 1. Sample Characteristics of Respondents to the 2012 Media Consumption Survey
Who Responded to All Independent Variables.
Variable
CPO (n = 535) Non-CPO (n = 1,955)
F(1, 2488) Cohen’s dM SD M SD
Age 38.57 15.32 54.35 17.00 377.17*** 0.948
Education level 3.32 1.80 3.86 1.88 34.25*** 0.286
Income level 3.20 2.42 4.30 2.41 86.91*** 0.455
Marital status 0.67 0.47 0.85 0.36 88.80*** 0.460
Parental status 0.47 0.33 0.25 0.44 10.63** −0.159
Race/ethnicity 0.60 0.49 0.79 0.41 84.20*** 0.448
Sex 0.61 0.49 0.44 0.50 49.70*** −0.344
Note. For education level, 3 = some college, but no degree and 4 = 2-year associate degree. For income level,
3 = US$30,000 to US$40,000 and 4 = US$40,000 to US$50,000. For marital status, 0 = never married,
1 = married at some point. For parental status, 0 = does not have a child younger than 18 years living in the
household, 1 = has child younger than 18 years living in the household. For race/ethnicity, 0 = some other race,
1 = White, Non-Hispanic. For sex, 0 = female, 1 = male. CPO refers to cell-phone-only households.
**p < .01. ***p < .001.
Table 2. Results From Fitting Main-Effects Models Utilizing CPO Status as a Single Factor
and With Control Variables.
Quest Label Question text Single factor With controls
Q3 News Do you happen to read any daily
newspaper or newspapers
regularly, or not?
Yes Yes
Q4 TV Do you happen to watch any TV
news programs regularly, or
not?
Yes Yes
Q5 Radio Do you listen to news on the
radio regularly, or not?
Yes No
Q25 Mobile Many national and local TV news
programs are available online
and on mobile devices. Did you
watch any TV news programs
on a computer, tablet, cell
phone or other device
yesterday, or not?
No No
Q68 App Have you ever downloaded an
application or “app” that allows
you to access news or news
headlines on a cell phone, tablet
or other mobile handheld
device, or not?
No No
Q70 Email Did you get any news or news
headlines by email yesterday, or
not?
No No
(continued)
at University of Minnesota Libraries on October 26, 2015jmq.sagepub.comDownloaded from
732 Journalism & Mass Communication Quarterly 92(3)
More specifically, as shown in Table 3, the odds that a respondent in a CPO house-
hold would read the newspaper (Q3 on survey form) were 53% lower than those of a
respondent in a non-CPO household; similarly, the odds of watching news on TV (Q4)
and listening to news on the radio (Q5) were 54% and 24% lower, respectively, for
CPO respondents. Thus, not having a representative sample of CPO and non-CPO
households would likely lead to an overestimation of those media-use choices. In con-
trast, the odds that a respondent in a CPO household read news on Twitter (Q75), fol-
lowed journalists on Twitter (Q77), or read news on general social networking sites
(Q82) were 157%, 174%, and 42% higher, respectively, than those of a respondent in
a non-CPO household. Thus, not having a representative sample of CPO and non-CPO
households would likely lead to an underestimation of those media-use choices.
Put differently, as shown in Figure 1, the model estimated that respondents in a
CPO household would respond “yes” to reading a newspaper 41% of the time (in con-
trast to 60% of the time for non-CPO households), to watching news on TV 63% of the
time (vs. 79%), to listening to news on the radio 40% of the time (vs. 47%), to reading
news on Twitter 47% of the time (vs. 25%), to following journalists on Twitter 61% of
the time (vs. 36%), and to reading news on general social networking sites 45% of the
time (vs. 37%). The findings thus appear to indicate that CPO households are consid-
erably less likely to use so-called “legacy media” (e.g., newspapers and radio) and far
more likely to use “new media” (e.g., Twitter and other social networking sites) for the
purposes of news consumption.
It should be noted, however, that CPO status was not a statistically significant pre-
dictor for four media-use variables: watching TV news on a computer or mobile
Quest Label Question text Single factor With controls
Q75 Twitter News Did you see any news or news
headlines on Twitter yesterday,
or not?
Yes Yes
Q77 Twitter Follow Do you follow any news
organizations or journalists on
Twitter, or not?
Yes Yes
Q82 Social Did you see any news or news
headlines on social networking
sites yesterday, or not?
Yes No
Q87 Podcast Did you watch or listen to a
news podcast yesterday, or not?
No No
Note. The control variables were the respondents’ age, education level, income level, marital status,
parental status, race and ethnicity, and sex. Question numbers refer to the original designator for the
2012 biennial media consumption survey. The label is intended to facilitate the identification of models
in the manuscript and in other tables. “Yes” and “No” refer to whether CPO status was found to be
statistically significant. For reasons of parsimony, we do not report the “With Controls” models for
variables in which the single-factor CPO status predictor was found to be statistically insignificant. No
effects were found in those models. CPO = cell phone only; TV = television.
Table 2. (continued)
at University of Minnesota Libraries on October 26, 2015jmq.sagepub.comDownloaded from
733
Table 3. Presentation of Logistic Models Regressing Media Use as a Function of CPO Status (Single Factor).
Variable
News
(n = 2,482)
TV
(n = 2,486)
Radio
(n = 2,485)
Mobile
(n = 1,000)
App
(n = 1,253)
Email
(n = 1,161)
Twitter news
(n = 2,171)
Twitter follow
(n = 238)
Social
(n = 1,102)
Podcast
(n = 944)
B SE Exp (B)B SE Exp (B)B SE Exp (B)B SE Exp (B)B SE Exp (B)B SE Exp (B)B SE Exp (B)B SE Exp (B)B SE Exp (B)B SE Exp (B)
CPO status −0.75 0.10 0.47 −0.77 0.11 0.46 −0.27 0.10 0.76 0.24 0.19 1.27 0.03 0.13 1.03 0.01 0.15 1.01 0.94 0.29 2.57 1.01 0.29 2.74 0.35 0.14 1.42 0.23 0.18 1.25
(p < .001) (p < .001) (p < .01) (p < .01) (p < .001) (p < .05)
(Intercept) 0.39 0.05 1.48 1.31 0.06 3.72 −0.14 0.05 0.87 −1.54 0.09 0.21 −0.14 0.07 0.87 −0.54 0.07 0.59 −1.07 0.18 0.34 −0.56 0.16 0.57 −0.55 0.07 0.58 −1.37 0.09 0.25
Model evaluation
Deviance 3350.37 2717.12 3413.60 953.30 1731.71 1529.48 289.12 315.44 1465.90 973.53
AIC 3354.37 2721.12 3417.60 957.30 1735.71 1533.48 293.12 319.44 1469.90 977.53
BIC 3366.01 2732.76 3429.23 967.11 1745.98 1543.60 300.07 326.39 1479.91 987.23
Pseudo R2
Adj. McFad. .015 .016 .000 −.005 −.003 −.004 .015 .021 .000 −.005
Efron’s .024 .022 .003 .002 .000 .000 .045 .054 .006 .002
Nagelkerke .031 .030 .004 .002 .000 .000 .060 .070 .008 .002
Note. p values calculated using the likelihood ratio test statistic for the one-degree-of-freedom χ2. p values only reported for variables with p < .05. Adj. McFad. refers to the Adjusted
McFadden pseudo R2. Because these models are each evaluating a different outcome variable, they should be compared with the corresponding models in Table 4, and not to one another.
Subsample sizes differ because some respondents did not answer that question. CPO = cell phone only; TV = television; AIC = Akaike information criterion; BIC = Bayesian information
criterion.
at University of Minnesota Libraries on October 26, 2015jmq.sagepub.comDownloaded from
734
Table 4. Bivariate Correlation of Independent Variables.
CPO Age Education Income Marital status Parental status Race/ethnicity Sex
CPO 1.000
Age −.363 1.000
Education −.117 .078 1.000
Income −.184 .006 .487 1.000
Marital status −.186 .463 .086 .206 1.000
Parental status .065 −.341 .022 .109 .166 1.000
Race/ethnicity −.181 .259 .165 .205 .201 −.102 1.000
Sex .140 −.104 .010 .116 −.055 −.006 −.068 1.000
Note. The reference categories are as follows: never married (marital status); not a parent (parental status); not “White, non-Hispanic” (race/ethnicity); and
female (sex). CPO = cell phone only.
at University of Minnesota Libraries on October 26, 2015jmq.sagepub.comDownloaded from
Watson et al. 735
computing device (Q25), downloading an “app” to access news on a mobile computing
device (Q68), accessing news headlines over email (Q70), and listening to news on
podcasts (Q87). Thus, certain media-use choices did not differ significantly across
households.
Demographics as Controls
The second hypothesis predicted that, after controlling for demographic differences
between CPO and non-CPO households, CPO status would remain a significant pre-
dictor of media use. To assess this hypothesis, a series of generalized linear models
using a logit link were fitted to all media-use questions for which CPO status had been
found to be a statistically significant predictor by itself; in contrast to the previous
models, however, the demographic variables were included in the model, in addition
to CPO status. As shown in Table 2, CPO status remained a statistically significant
predictor for four of the six media-use variables: reading a daily newspaper, watching
news on TV, reading news on Twitter, and following journalists on Twitter. Thus, the
second hypothesis also received partial support.
More specifically, as shown in Table 5, even after controlling for demographic dif-
ferences, the odds that a respondent in a CPO household would read the newspaper
(Q3) or watch news on TV (Q4) were 29% and 25% lower, respectively, than those of
a respondent in a non-CPO household. Thus, not having a representative sample of
CPO and non-CPO households would likely lead to an overestimation of those media-
use choices. In contrast, the odds that a respondent in a CPO household read news on
Figure 1. Predicted probability of respondents answering “Yes” to 10 media-use questions
as a function of whether the respondent lives in a CPO household or a non-CPO household.
Note. CPO status was found to be a statistically significant predictor for news, TV, radio, Twitter news,
Twitter follow, and social. CPO = cell phone only; TV = television.
at University of Minnesota Libraries on October 26, 2015jmq.sagepub.comDownloaded from
736
Table 5. Presentation of Logistic Models Regressing Media Use as a Function of CPO Status and Seven Control Variables (With Controls).
Variable
News
(n = 2,482)
TV
(n = 2,486)
Radio
(n = 2,485)
Twitter news
(n = 2,171)
Twitter follow
(n = 238)
Social
(n = 1,102)
B SE Exp (B)B SE Exp (B)B SE Exp (B)B SE Exp (B)B SE Exp (B)B SE Exp (B)
CPO status −0.35 0.11 0.71 −0.28 0.12 0.75 −0.05 0.11 0.95 0.91 0.33 2.49 1.10 0.32 3.00 0.14 0.15 1.15
(p < .01) (p < .05) (p < .01) (p < .001)
Age 0.03 0.00 1.03 0.04 0.00 1.04 0.01 0.00 1.01 0.00 0.01 1.00 0.00 0.01 1.00 −0.03 0.01 0.97
(p < .001) (p < .001) (p < .01) (p < .001)
Education level 0.18 0.03 1.20 −0.06 0.03 0.94 0.11 0.03 1.12 0.25 0.10 1.28 0.15 0.09 1.16 0.15 0.04 1.16
(p < .001) (p < .001) (p < .05) (p < .001)
Income level 0.04 0.02 1.04 0.01 0.02 1.01 0.09 0.02 1.10 0.01 0.08 1.01 0.11 0.07 1.12 −0.01 0.03 0.99
(p < .001)
Marital status −0.18 0.14 0.84 0.07 0.15 1.08 0.01 0.13 1.01 −0.10 0.47 0.90 −1.01 0.45 0.37 0.21 0.20 1.23
(p < .05)
Parental status 0.20 0.11 1.23 0.17 0.12 1.18 0.50 0.11 1.65 0.34 0.37 1.42 0.76 0.36 2.15 0.29 0.15 1.33
(p < .001) (p < .05)
Race/ethnicity 0.12 0.10 1.13 −0.72 0.13 0.49 −0.07 0.10 0.93 −0.20 0.34 0.82 −0.16 0.31 0.85 0.10 0.15 1.10
(p < .001)
Sex 0.36 0.09 1.43 −0.11 0.10 0.89 0.19 0.09 1.21 0.71 0.31 2.03 −0.12 0.28 0.89 −0.03 0.13 0.97
(p < .001) (p < .05) (p < .05)
(Intercept) −1.97 0.20 0.14 −0.21 0.22 0.81 −1.65 0.20 0.19 −2.59 0.62 0.08 −1.29 0.55 0.28 −0.25 0.27 0.78
Model evaluation
Deviance 3,167.53 2,500.67 3,291.55 271.80 301.16 1,415.60
AIC 3,185.53 2,518.67 3,309.55 289.80 319.16 1,433.60
BIC 3,237.88 2,571.03 3,361.91 321.05 350.45 1,478.64
Pseudo R2
Adj. McFad. .065 .089 .032 .026 .022 .025
Efron’s .093 .111 .051 .114 .110 .050
Nagelkerke .124 .152 .068 .154 .144 .068
Note. p values calculated using the likelihood ratio test statistic for the one-degree-of-freedom χ2. p values only reported for variables with p < .05. Models were only fitted for
questions where CPO status was found to be statistically significant as the sole predictor. Adj. McFad. refers to the Adjusted McFadden pseudo R2. The reference categories are as
follows: never married (marital status); not a parent (parental status); not “White, non-Hispanic” (race/ethnicity); and female (sex). Because these models are each evaluating a different
outcome variable, they should be compared with the corresponding models in Table 3, and not to one another. Subsample sizes differ because some respondents did not answer that
question. CPO = cell phone only; TV = television.
at University of Minnesota Libraries on October 26, 2015jmq.sagepub.comDownloaded from
Watson et al. 737
Twitter (Q75) or followed journalists on Twitter (Q77) were 149% and 200% higher,
respectively, than those of a respondent in a non-CPO household. Thus, not having a
representative sample of CPO and non-CPO households would likely lead to an under-
estimation of those media-use choices.
Put differently, as shown in Figure 2, when holding all demographic variables con-
stant at their mean, the model estimated that respondents in a CPO household would
respond “yes” to reading a newspaper 49% of the time (in contrast to 58% of the time
for non-CPO households), to watching news on TV 74% of the time (vs. 79%), to
reading news on Twitter 45% of the time (vs. 24%), and to following journalists on
Twitter 62% of the time (vs. 35%). After adding in the demographic variables, CPO
status was no longer a statistically significant predictor for two of the media-use vari-
ables: listening to news on the radio and reading news on general social networking
sites. Thus, the findings indicate that, in the majority of the cases, the differences in
media-use behaviors between CPO and non-CPO households cannot be accounted for
by demographic variables alone.
Discussion
This study found that there are significant differences in media use between CPO and
non-CPO households. There were significant differences between CPO and non-CPO
households’ use of 6 of 10 media. Demographics proved to be adequate controls for
only two of these variables: regularly listening to a radio news program and receiving
news headlines on any social networking site within the previous day. However, even
after controlling for demographic differences, CPO respondents were significantly
less likely to regularly read a daily newspaper or watch a TV news program, and sig-
nificantly more likely to receive news headlines on Twitter or have followed a news
organization or journalist on the social networking site. The fact that demographics do
not adequately control for the media-use differences between CPO and non-CPO
households suggest there are additional characteristics that distinguish these two
groups. It is possible additional extrinsic and intrinsic factors that influence technol-
ogy (i.e., cell phone) adoption may both explain the media choice to rely completely
on a cell phone for communication and that choice may be correlated with other media
choices. The Pew survey did not probe why households rely solely on cell phones.
Thus, it is not possible to use these data to further examine how the CPO choice may
be related to other aspects of individuals’ media use. Further research should investi-
gate this relationship to deepen our understanding of how a CPO coverage bias
uniquely affects mass communication researchers.
That said, this study did illustrate that media-use surveys that do not include a rep-
resentative sample of CPO households, even if the data are weighted to approximate
the demographics of a known population, are likely to overestimate regular daily
newspaper reading and TV news program watching, and underestimate the use of
Twitter to receive news headlines or follow news organizations or journalists. Put dif-
ferently, the failure to properly sample CPO households may lead to findings that sug-
gest that, for the purposes of news consumption, “legacy media” use is higher than it
at University of Minnesota Libraries on October 26, 2015jmq.sagepub.comDownloaded from
738 Journalism & Mass Communication Quarterly 92(3)
actually is, and that some types of “new media” use are lower than they actually are.
This is an important finding in light of the growing numbers of individuals—at least
in the United States—who are “cutting the cord” (Blumberg et al., 2013). In particular,
scholars should be very careful when considering sampling strategies and not become
overreliant on statistical procedures to account for sampling deficiencies. To account
for the coverage bias associated with CPO-only households, it is best to use a dual-
frame design that includes a representative sample of both telephone and cell phone
(including CPO) numbers.
Because of various hurdles associated with conducting RDD surveys, mass com-
munication scholars are eschewing telephone surveys for various forms of web-based
surveys. Thus, one might question how important the potential coverage bias associ-
ated with CPO households in telephone-based surveys continues to be. However, the
political and health literature that has suggested that demographic weighting of survey
data has been presumed to control for biases in nonrepresentative telephone samples
(i.e., samples that do not include a representative number of CPO households) is also
Figure 2. Predicted probability of respondents answering “Yes” to six media-use questions
as a function of whether the respondent lives in a CPO household or a non-CPO household,
holding demographic variables constant at their mean.
Note. CPO status was found to remain a statistically significant predictor for news, TV, Twitter news, and
Twitter follow. CPO = cell phone only; TV = television.
at University of Minnesota Libraries on October 26, 2015jmq.sagepub.comDownloaded from
Watson et al. 739
used to rationalize weighting web survey data to achieve a “representative” sample
(see, for example, Correa et al., 2010; Curran et al., 2009; de Zúñiga et al., 2012). The
results of this study demonstrate that demographics alone were not adequate controls
for all of the sampling biases associated with CPO households in mass media tele-
phone surveys. Just as the choice to use a telephone or cell phone is a media choice that
may be correlated with other media preferences, so is the choice to use the web. Thus,
there are also likely differences in media preferences between web and nonweb sam-
ples that cannot be controlled for based on demographics alone. Establishing that this
is the case is beyond the scope of this study. But because this study showed that poten-
tial CPO coverage biases pose a unique concern for mass communication scholars, the
political and health communication literature on weighting telephone samples should
not continue to justify weighting media-use survey data gathered online based solely
on demographics to make it “representative” without further research on how different
survey modes (e.g., telephone, cell phone, web, etc.) affect media-use variables.
It must be noted that this study examined one telephone survey data set. Arguably
the Pew Research Center is among the most respected sources of survey data, repre-
senting the “gold standard” of telephone survey methods. Pew’s biennial media con-
sumption survey is also among the most important sources of media-use data, which
is why we chose to use this data set for our study. Nonetheless, in addition to the cover-
age error that was the focus of this article, there are other sources of representation
(i.e., how well does the sample represent the population of interest) and measurement
errors (i.e., how well does the survey capture the desired construct) that are associated
with any survey (Groves & Lyberg, 2010). For example, the low response rate (11%
for landlines and 7% for cell phones) raises concerns about possible nonresponse
error.5 On the measurement side, our secondary analysis focused on dichotomous yes/
no measures that may mask meaningful differences in the frequency with which cell
phone and non-CPO households use different media. Furthermore, this study relies on
self-reported media use, which may be overestimated, particularly for younger demo-
graphics (Prior, 2009). This study does not attempt to ignore the potential biases in
self-reported data (nor does it seek to produce an accurate estimate of media use, per
se). Rather, the study’s findings suggests that in addition to being concerned with the
accuracy of self-reported survey data, researchers should be concerned with potential
biases associated with CPO households in media-use surveys. In addition, it is possi-
ble that the instructions provided to the interviewee about whom to speak with (i.e.,
with the youngest adult for landline surveys and the individual who answered for cell
phone surveys) influenced the Pew sample, and consequently the results. Although the
landline-only sample was considerably older than the CPO sample (medians of 57 and
34, respectively), it is possible that this difference in sampling procedure artificially
makes the samples appear to be more similar than they actually are, thus potentially
underestimating the differences found in this study.
That said, future research should seek to validate these findings by examining dif-
ferent data sets and different variables that extend beyond simple media use to exam-
ine, for example, using online media as tools for civic engagement, participation, and
political discussion. In addition, future researchers may also wish to consider
at University of Minnesota Libraries on October 26, 2015jmq.sagepub.comDownloaded from
740 Journalism & Mass Communication Quarterly 92(3)
cell-phone-mostly homes—that is, homes that still have a landline but rarely make use
of it—as this growing phenomenon poses similar challenges for mass communication
researchers and is also a media choice. It would also be helpful to examine whether
demographics can control for differences in responses across multiple survey modes,
including comparing RDD with landline and cell phone samples with online surveys.
The present study should, however, give pause to mass media scholars who trust
demographic variables alone to control for the coverage bias associated with the dif-
ferences between CPO and non-CPO households.
Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship,
and/or publication of this article.
Funding
The author(s) received no financial support for the research, authorship, and/or publication of
this article.
Notes
1. GfK Knowledge Networks is primarily known for online panels, but their lists were con-
structed using RDD sampling prior to 2009, when address-based sampling was added.
American National Election Studies and the National Annenberg Election Survey both
have incorporated Knowledge Networks samples for different individual phone-based,
phone-plus-web, and face-to-face-plus-web surveys (GfK Knowledge Networks, n.d.).
2. Journals were searched for the following keywords/phrases: survey, questionnaire, panel,
and secondary data.
3. Weighting involves assigning a weight, based on a known value for a population, greater
than one to individuals who represent characteristics underrepresented in a sample and a
weight of less than one to individuals who represented characteristics that are overrep-
resented in a sample. Actual weights depend on how skewed the distribution of a given
sample statistic is compared with a known population parameter
4. Specifically, these questions were (using the designators from the original codebook) Q3,
Q4, Q5, Q25, Q68, Q70, Q73, Q75, Q77, Q82, and Q87.
5. Cultural and linguistic differences may also be a source of nonresponse error among
Hispanics, who may also be among the low-wage, transient, cell-phone-only households.
While it is difficult to address the fact that some Hispanics may be wary of unknown
surveyors, the Pew Research Center did conduct interviews in English and Spanish to
overcome linguistic barriers.
References
Ansolabehere, S., & Schaffner, B. F. (2010). Residential mobility, family structure, and the cell-
only population. Public Opinion Quarterly, 74, 244-259. doi:10.1093/poq/nfq018
Avery, E. (2010). Contextual and audience moderators of channel selection and message recep-
tion of public health information in routine and crisis situations. Journal of Public Relations
Research, 22, 378-403. doi:10.1080/10627261003801404
at University of Minnesota Libraries on October 26, 2015jmq.sagepub.comDownloaded from
Watson et al. 741
Blumberg, S. J., Ganesh, N., Luke, J. V., & Gonzales, G. (2013). Wireless substitution: State-
level estimates from the National Health Interview Survey, 2012. National Health Statistics
Reports. Retrieved from http://www.cdc.gov/nchs/data/nhsr/nhsr070.pdf
Blumberg, S. J., & Luke, J. V. (2009). Reevaluating the need for concern regarding noncoverage
bias in landline surveys. American Journal of Public Health, 99, 1806-1810. doi:10.2105/
AJPH.2008.152835
Blumberg, S. J., Luke, J. V., Ganesh, N., Davern, M. E., & Boudreaux, M. H. (2012). Wireless
substitution: State-level estimates from the National Health Interview Survey, 2010–2011.
Retrieved from http://www.cdc.gov/nchs/data/nhsr/nhsr061.pdf
Bobkowski, P. S. (2009). Adolescent religiosity and selective exposure to television. Journal of
Media and Religion, 8, 55-70. doi:10.1080/15348420802670942
Chan-Olmsted, S., Rim, H., & Zerba, A. (2013). Mobile news adoption among young adults
examining the roles of perceptions, news consumption, and media usage. Journalism &
Mass Communication Quarterly, 90, 126-147. doi:10.1177/1077699012468742
Conci, M., Pianesi, F., & Zancanaro, M. (2009). Useful, social and enjoyable: Mobile phone
adoption by older people. Human-Computer Interaction—INTERACT 2009, 5726, 63-76.
Retrieved from http://link.springer.com/chapter/10.1007/978-3-642-03655-2_7
Correa, R., Hinsley, A. W., & de Zúñiga, H. G. (2010). Who interacts on the Web? The inter-
section of users’ personality and social media use. Computers in Human Behavior, 26,
247-253. doi:10.1016/j.chb.2009.09.003
Curran, J., Iyengar, S., Lund, A. B., & Salovaara-Moring, I. (2009). Media system, public
knowledge and democracy: A comparative study. European Journal of Communication,
24, 5-26. doi:10.1177/0267323108098943
Dahlgren, P. (2009). Media and political engagement: Citizens, communication and democ-
racy. Cambridge, UK: Cambridge University Press.
Dennis, M., & DiSorga, C. (2009). Meeting the challenge of cell phone-only households, young
adults and minorities: Introducing address-based sampling to KnowledgePanel®. Retrieved
from http://www.knowledgenetworks.com/accuracy/spring2009/Dennis-DiSogra-Graham-
spring09.html
de Zúñiga, H. G., Jung, N., & Valenzuela, S. (2012). Social media use for news and individuals’
social capital, civic engagement and political participation. Journal of Computer-Mediated
Communication, 17, 319-336. doi:10.1111/j.1083-6101.2012.01574.x
Dillman, D. A., Smyth, J. D., & Christian, L. M. (2008). Internet, mail, and mixed-mode sur-
veys: The tailored design method (3rd ed.). Hoboken, NJ: John Wiley.
Emig, A. G. (1995). Community ties and dependence on media for public affairs. Journalism &
Mass Communication Quarterly, 72, 402-411. doi:10.1177/107769909507200212
GfK Knowledge Networks. (n.d.). Description of selected past projects. Retrieved from http://
www.knowledgenetworks.com/ganp/past-projects.html
Graber, D. A. (2009). Mass media and American politics (8th ed.). Washington, DC: CQ
Press.
Groves, R. M., & Lyberg, L. (2010). Total survey error: Past, present, and future. Public Opinion
Quarterly, 74, 849-879. doi:10.1093/poq/nfq065
Hill, M. R., Tchernev, J. M., & Holbert, R. L. (2012). Do we need to go cellular? Assessing
political media consumption using a single-frame landline/cellular survey design. Mass
Communication and Society, 15, 284-306. doi:10.1080/15205436.2011.642926
Hmielowski, J. D. (2012). Intramedia moderation, electoral ambivalence, and electoral decision
making. Mass Communication and Society, 15, 454-477. doi:10.1080/15205436.2011.616640
at University of Minnesota Libraries on October 26, 2015jmq.sagepub.comDownloaded from
742 Journalism & Mass Communication Quarterly 92(3)
Keeter, S. (2006). The impact of cell phone noncoverage bias on polling in the 2004 presidential
election. Public Opinion Quarterly, 70, 88-98. doi:10.1093/poq/nfj008
Keeter, S., Kennedy, C., Clark, A., Tompson, T., & Mokrzycki, M. (2007). What’s missing
from national landline RDD surveys? The impact of the growing cell-only population.
Public Opinion Quarterly, 71, 772-792. doi:10.1093/poq/nfm053
Lavrakas, P. J., Shuttles, C. D., Steeh, C., & Finberg, H. (2007). The state of surveying cell
phone numbers in the United States 2007 and beyond. Public Opinion Quarterly, 71,
840-854. doi:10.1093/poq/nfm054
Leung, L., & Wei, R. (2009). Who are the mobile phone have-nots? Influences and conse-
quences. New Media & Society, 1, 209-226. doi:10.1177/1461444899001002003
Link, M. W., Battaglia, M. P., Frankel, M. R., Osborn, L., & Mokdad, A. H. (2007). Reaching
the U.S. cell phone generation comparison of cell phone survey results with an ongo-
ing landline telephone survey. Public Opinion Quarterly, 71, 814-839. doi:10.1093/poq/
nfm051
Lu, J., Yu, C., Liu, C., & Yao, J. E. (2003). Technology acceptance model for wireless Internet.
Internet Research, 13, 206-222. doi:10.1108/10662240310478222
McLeod, J. M., Scheufele, D. A., & Moy, P. (1999). Community, communication, and partici-
pation: The role of mass media and interpersonal discussion in local political participation.
Political Communication, 16, 315-336. doi:10.1080/105846099198659
Mitchell, A., & Guskin, E. (2013). Twitter news consumers: Young, mobile and educated.
Retrieved from http://www.journalism.org/2013/11/04/twitter-news-consumers-young-
mobile-and-educated/
Mokrzycki, M., Keeter, S., & Kennedy, C. (2009). Cell-phone-only voters in the 2008 exit
poll and implications for future noncoverage bias. Public Opinion Quarterly, 73, 845-865.
doi:10.1093/poq/nfp081
Noar, S. M. (2006). A 10-year retrospective of research in health mass media cam-
paigns: Where do we go from here? Journal of Health Communication, 11, 21-42.
doi:10.1080/10810730500461059
Pew Research Center for the People and the Press. (2011). Cell phone surveys. Retrieved from
http://www.people-press.org/methodology/collecting-survey-data/cell-phone-surveys/
Pew Research Center for the People and the Press. (2012a). About the media consumption
survey data. Retrieved from http://www.people-press.org/2012/09/27/about-the-media-
consumption-survey-data/
Pew Research Center for the People and the Press. (2012b). 2012 media consumption sur-
vey. Retrieved from http://www.people-press.org/2012/06/03/2012-media-consumption-
survey/
Prior, M. (2009). The immensely inflated news audience: Assessing bias in self-reported news
exposure. Public Opinion Quarterly, 73, 130-145. doi:10.1093/poq/nfp002
Rice, R. E., & Katz, J.E. (2003). Comparing internet and mobile phone usage: Digital divides
of usage, adoption, and dropouts. Telecommunications Policy, 27, 597-623. doi:10.1016/
S0308-5961(03)00068-5
Rittenberg, J., Tewksbury, D., & Casey, S. (2012). Media preferences and democracy: Refining
the “relative entertainment preference” hypothesis. Mass Communication and Society, 15,
921-942. doi:10.1080/15205436.2011.622065
van Biljon, J., & Kotzé, P. (2007). Modeling the factors that influence mobile phone adoption.
In Proceedings of the 2007 Annual Research Conference of the South African Institute
of Computer Scientists and Information Technologists on IT Research in Developing
Countries (pp. 152-161). ACM, New York, NY. doi:10.1145/1292491.1292509
at University of Minnesota Libraries on October 26, 2015jmq.sagepub.comDownloaded from
Watson et al. 743
Weeks, B. E., & Holbert, R. L. (2013). Predicting dissemination of news content in social media:
A focus on reception, friending, and partisanship. Journalism & Mass Communication
Quarterly, 90, 212-232. doi:10.1177/1077699013482906
Wei, L., & Hindman, D. B. (2011). Does the digital divide matter more? Comparing the
effects of new media and old media use on the education-based knowledge gap. Mass
Communication and Society, 14, 216-235. doi:10.1080/15205431003642707
Wei, R. (2008). Motivations for using the mobile phone for mass communications and enter-
tainment. Telematics and Informatics, 25, 36-46. doi:10.1016/j.tele.2006.03.001
Author Biographies
Brendan R. Watson (Ph.D., North Carolina) is an assistant professor in the School of Journalism
& Mass Communication at the University of Minnesota-Twin Cities. His research interests
include community information needs, how digital technologies including social and mobile
media are/are not changing communicaiton about public affairs issues, and quantitative com-
munication research methods. More about his research is available at http://brendanwatson.net.
Rodrigo Zamith (Ph.D., Minnesota) is an assistant professor in the Journalism Department at
the University of Massachusetts, Amherst. His research focuses on the reconfiguration of jour-
nalism in a changing media environment as well as the development of digital research methods.
More about his research is available at http://rodrigozamith.com.
Sarah Cavanah is a Ph.D. student in the School of Journalism & Mass Communication at the
University of Minnesota-Twin Cities. Her research interests include community news and infor-
mation and scholastic journalism.
Seth C. Lewis (Ph.D., Texas) is an associate professor and Mitchell V. Charnley Faculty Fellow
in the School of Journalism & Mass Communication at the University of Minnesota–Twin
Cities. His research explores the digital transformation of journalism, with a focus on conceptu-
alizing human–technology interactions associated with data, code, algorithms, and social media.
More about his research is available at http://sethlewis.org.
at University of Minnesota Libraries on October 26, 2015jmq.sagepub.comDownloaded from
