the fox news effect: media bias and voter behaviorstaff/wrkshop_papers/2004_fall... ·...
TRANSCRIPT
The Fox News Effect: Media Bias and Voter Behavior∗
Stefano DellaVigna
UC Berkeley and NBER
Ethan Kaplan
UC Berkeley
This version: November 6, 2004.
Abstract
Does the media affect voting? We address this question by looking at the Fox News
effect. Between October 1996 and November 2000, the conservative news channel was intro-
duced in the cable programming of 20 percent of US towns. We investigate if Republicans
gained vote share in towns where Fox News was broadcast. We find no significant effect
of the introduction of Fox News on the vote share in Presidential elections between 1996
and 2000. We can rule out an effect of Fox News larger than 0.5 percentage points. We
also find no significant effect of Fox News on voter turnout. Our results imply that Fox
News convinced at most 2 percent of its viewers to vote Republican. This evidence suggests
that voters are sophisticated and filter out media bias. Alternatively, voters select into the
different media, so most Fox News viewers would already vote Republican.
∗PRELIMINARY VERSION. DO NOT QUOTE WITHOUT PERMISSION. Dan Acland, Shawn Banan-
zadeh, Jessica Chan, Marguerite Converse, Neil Dandavati, Tatyana Deryugina, Monica Deza, Dylan Fox,
Melissa Galicia, Calvin Ho, Sudhamas Khanchanawong, Richard Kim, Martin Kohan, Vipul Kumar, Jonathan
Leung, Clarice Li, Tze Yang Lim, Ming Mai, Sameer Parekh, Sharmini Radakrishnan, Rohan Relan, Chanda
Singh, Matthew Stone, Nan Zhang, Sibo Zhao, and Liya Zhu did an excellent job on the collection of the voting
and the cable data. Saurabh Bhargava, Avi Ebenstein, and Devin Pope provided excellent research assistance.
1 Introduction
Do the media provide biased information? Some researchers focus on alleged liberal media
bias of the New York Times and CNN (Puglisi, 2004; Mixon et al., 2004). Other researchers
point out the presumed conservative bias of Fox News (Groseclose and Milyo, 2004).
It is not clear, however, whether media bias has any effect on the behavior of the consumers.
If the consumers are sophisticated about the bias, they can at least partially recover the
unbiased information. Alternatively, if consumers enjoy listening to biased media that confirm
their priors, the media do not affect their thinking. In both cases, media bias may affect the
precision of the information, but does not affect beliefs on average.
On the other hand, if consumers underestimate the extent of media bias, are subject to per-
suasion (De Marzo, Vayanos, and Zwiebel, 2003) or to confirmatory bias (Rabin and Schrag,
1999), media slant affects systematically beliefs and voting behavior. In this latter case, changes
in rules of media concentration such as those discussed by the Federal Communications Com-
mission (FCC), may have large welfare impact.
Whether media bias affects behavior is an important question for research and policy. In
this paper, we bring empirical evidence to bear on this question. We focus on the introduction
of Fox News in local cable markets and consider the impact on voting behavior. We employ
a differences-in-differences estimator and compare changes in vote share for towns where Fox
News is introduced before the 2000 elections to towns where it is introduced later. We exploit
three key elements of the Fox News case, the fast expansion, the geographical differentiation,
and the widely-perceived conservative slant in its coverage.
The 24-hour channel Fox News was introduced by Rupert Murdoch in October 1996 in
order to compete with CNN. Like CNN, it is only offered via cable and, to a smaller extent,
via satellite. Thanks to an aggressive marketing campaign, a number of cable companies added
Fox News to their programming over the next four years. The geographical expansion of Fox
News was accompanied by a corresponding increase in its share of the audience. By June 2000,
17 percent of the US population listened regularly to the Fox News channel (Pew, 2004).
The nature of the cable industry also induces substantial geographical variation in the
access to Fox News. Cable markets are natural monopolies with constraints on the number of
channels. The availability of Fox News in a town depends on whether the local cable company
decides to add it to the programming, possibly at the expense of another channel. Cable
companies in two neighboring towns may take different decisions, creating variation in the
access to this media.
Even given the popularity of Fox News and the variation in Fox News diffusion, it is
unclear whether the addition of any single media source would have a significant impact on the
information received by voters. Fox News coverage, however, is unique among the television
media. Groseclose and Milyo (2004) use data on citations of think-tanks to rank the political
1
orientation of the media. They conclude that Fox News is significantly to the right of all
the other mainstream television networks (ABC, CBS, CNN, and NBC). The introduction of
Fox News into a cable market, therefore, is likely to affect systematically and significantly the
available political information in that cable market.
In order to analyze the impact of Fox News on voting, we assemble a new panel data set of
town-level election data and match it with town-level data on programming of cable companies.
The data set covers the federal elections of 1996 and 2000 in 24 US States. We compare the
change in the Republican vote share between 1996 and 2000 for the towns that had adopted
Fox News by 2000 with those that had not.
Our main result is that the entry of Fox News did not affect voting behavior. The estimates
of the no effect are very precise. We can rule out an effect of Fox News entry on vote share in
Presidential elections larger than .5 percentage points. Since Fox News in 2000 is available in
less than 50 percent of the households, the overall effect of Fox News on the 2000 elections is
estimated to be less than .25 percentage points, 250,000 votes, and possibly zero. The results
hold after the introduction of population controls, state, district, and county dummies, as well
as controls for features of the cable system. The results do not vary by Census region and are
robust to controls for time trends
As an additional test, we compare the effect of Fox News on races with national coverage
and on more local races. Since Fox News does not cover local candidates, the impact of
Fox News, if any, should be on the national races. As a first test, we compare the effect on
presidential elections to the effect on US House elections, that Fox News does not cover. We
find no evidence that Fox News differentially affected the Republican vote share for Presidential
and House elections.
We also focus on Senatorial races. While most races for the US Senate go unmentioned in
the Fox News programming, the Senatorial race in New York State between Hillary Clinton
and Rick Lazio attracted considerable attention. In particular, Fox News maintained a very
critical position toward Hillary Clinton. If Fox News affects viewers, therefore, its effect should
be highest in this race. Instead, we cannot reject the hypothesis that the effect on the New
York race is the same as on the other Senate races, that is, no effect.
Fox News entry in media markets does not appear to have had any significant effect on
the Republican vote share. This, however, does not imply that Fox News did not affect voting
behavior. It is possible that Fox News energized both Republicans and Democrats, inducing
both to turn out more to the polls. According to this scenario, the offsetting increases in
turnout lead to no overall effect on Republican vote share. We test for this possibility by
examining the effect of Fox News on two measures of voter turnout. For both measures, we
fail to reject the hypothesis of no effect of Fox News on turnout.
Overall, we find no effect of Fox News on either vote share for Republicans or turnout to
the polls. While the point estimate of this effect is essentially zero, it is possible that we are
2
not detecting it due to lack of power. Using a simple back-of-the-envelope calculation using
audience data, we show that Fox News convinced at most one listener out of 100 to go to the
polls. The point estimates suggest an even smaller effect.
These results contrast sharply with previous findings of the impact of Fox News on political
beliefs. A survey run by Kull et al. (2003) finds that Fox News watchers are 50 percent more
likely to believe (erroneously) that weapons of mass destruction were found in Iraq by October
2003, compared to viewers of other networks. Our findings suggest that these survey results
overstate the role of Fox News due to selection.
We provide three interpretations of our results, based on selection, rational decoding, and
superficiality. The selection story says that Fox News does not impact voter behavior because
only conservative voters watch it. The decoding story is that voters display a (perhaps sur-
prising) ability to interpret media coverage and are not swayed (on average) by media bias.
The superficiality story suggests that media bias affects voter beliefs, but this is not ultimately
what drives the voting decision. The selection story reconciles the survey evidence with the
results in this paper, except for the turnout results. The decoding story is consistent with all
the facts in this paper, and can explain the survey evidence if combined with the selection
story. Finally, the superficiality story can explain all facts, but at the cost of assuming that
political beliefs do not impact voting.
These results suggest that persuasion bias (De Marzo, Vayanos, and Zwiebel, 2003) and
confirmatory bias (Rabin and Schrag, 1999) may be more limited than previously thought.
The results are most consistent with models of media bias such as. According to these models,
the exogenous introduction of a new information source does not systematically affect voter
beliefs.
Our paper also relates to the evidence of shifts in media exposure on voter turnout. George
and Waldfogel (2004) show that, in areas where the New York Times circulation expanded in
the ’90s, voter turnout in local election decreases in particular between more educated voters
who were more likely to read the Times. Gentzkow (2004) finds a similar effect on voter
turnout from the expansion of television. Finally, Prat and Stromberg (2004) find that the
introduction of a private TV in Sweden increases voter turnout. Unlike in these papers, we
find no effect on voter turnout of the arrival of a new information source.
The remainder of the paper is structured as follows. In Section 2 we provide an overview
of the cable industry and of the history of the expansion of Fox News. In Section 3 we discuss
the voting and the cable data. In Section 4 we present the empirical results of the paper, first
on the Republican vote share and then on turnout. In Section 5 we present interpretations
and in Section 6 we conclude.
3
2 Cable Industry and Fox News
Cable industry. The cable industry is a local natural monopoly. Once one company has
paid the fixed cost to lay the cables in a town, it is uncommon for a second company to pay
the fixed cost as well and enter the local market. In our sample, only ten percent of the towns
have two competing cable companies, and only one percent have three or more companies.
The towns with multiple cable systems are on average larger.
A second important feature of the cable industry is the technological constraint on the
number of channels. Channels, therefore, are rationed, and consumers have to take as given
the programming choice of the cable company. This generates substantial variation across
towns in the programming provided. Established channels like CNN are offered in almost all
towns. New channels like Fox News, instead, have to convince local cable companies to add
them, often at the expense of other channels being dropped.
Cable companies pay a monthly fee, typically between 10 cents and 40 cents, to the networks
that they carry. Cable companies also pay fees to the town that grants them the right to
broadcast. These fees are typically set as part of a 10- to 15-year contract between the cable
company and the town. Finally, cable companies get their revenue from the monthly subscriber
fees. The amount of the subscriber fee is partly regulated and varies between $10 and $60,
depending on the cable company and on the tier of service.
Fox News history. In March 1996 Rupert Murdoch, CEO of News Corp., announced
the introduction of a 24-hour-a-day cable news channel. The new channel, Fox News Chan-
nel, was created to compete with CNN. Prior to the launch of the Fox News Channel, news
broadcasts took up a small share of programming of the Fox channels, such as Fox Sport, Fox
Entertainment, and Fox Family. There was no broadcast news at a national level, and prime
time programming on Fox did not include news. The main television sources of news before
1996 were the three major broadcast networks–ABC, CBS, and NBC–, in addition to CNN,
distributed via cable.
The distribution of Fox News started in October 1996 in a limited number of cable markets.
In order to facilitate the spread of Fox News, the Fox network took the unusual move to offer
a one-time payment of $10 per subscriber to cable companies that included Fox News in its
programming. TCI was one of the first cable company to sign a contract with Fox News in
1996. After the initial contract was signed by the parent company, local affiliates of the cable
company decided whether to include Fox News among the channels transmitted. By November
2000, AT&T Broadband, which acquired TCI Cable, offered Fox News in 33 percent of the
1,538 towns served by AT&T Broadband affiliates (estimate from our sample). Adelphia
Communications, instead, had a late agreement with Fox News. By November 2000 only 6.3%
of the 1,301 towns in our sample served by Adelphia affiliates included Fox News in their
broadcast.
4
In addition to the cable programming, short news segments from the Fox News channel
are distributed by the local TV stations that are affiliates of Fox Broadcasting. However, the
complete programming of Fox News is only available via cable and, for a much smaller number
of subscribers, via satellite.
In the expansion of Fox News, the most relevant year for this study is year 2000. In our
sample of 24 States, Fox News is present in the cable programming of 20.3 percent of the towns
with cable service. Since the towns reached by Fox News in 2000 were, on average, twice as
large, Fox News reached over 33 percent of the population of these States. Overall, in 2000
Fox News reached 50 million subscribers.1 Moreover, in April 2000 17 percent of Americans
watched regularly the Fox News Cable Channel and 28 percent watched it sometimes (Pew
Report, 2004).
Fox News content and programming. A key feature of Fox News for the purpose
of this study is the significant differentiation in political coverage relative to CNN and the
network news. Groseclose and Milyo (2004) use data on citations of think-tanks to rank the
political orientation of the news from the different media. Groseclose and Milyo impute an
ADA score for the media based on the ADA score of the members of Congress that refer to
the same think-tanks.2 Their maximum likelihood estimation results assign an imputed ADA
score for Fox News of 29.0. This score is significantly lower than the score of any of the other
mainstream television media (ABC, CBS, NBC), with scores between 52.8 (ABC) and 60.8
(CBS). The difference in political ranking between Fox News and the networks is three times
as large as the maximum distance between the rankings themselves.
Interestingly, the study by Groseclose and Milyo may, if anything, underestimate the degree
of political differentiation of Fox News. The two Fox News shows with the highest ratings, The
O’Reilly Factor and Hannity & Colmes, are likely more conservative than the Special Report,
which Groseclose and Milyo use to estimate the ADA citations. Fox’s top-rated show, The
O’Reilly Factor, is named after its confrontational anchor, Bill O’Reilly. The show, which has
aired since the beginning of Fox News, now occupies the popular 8pm spot. The show, which
hosts mainly journalists and politicians, deals mostly with political topics. It is not meant to
be unbiased; in fact, every segment of the show begins with a “Talking Points” memo, in which
Bill O’Reilly shares his opinion on a leading news story. The second most popular Fox News
programs is Hannity & Colmes, a talk show hosted by conservative Sean Hannity and liberal
Alan Colmes. The aggressive Hannity typically prevails over the calmer Colmes. This show
has been part of the programming of Fox News since 1996 and is aired at 9pm.
1The last figure is from from Fox News website.2The ADA score is a measure of political orientation created by Americans for Democratic Action.
5
3 Data
Cable data. We obtained information on programming and contracts in local cable markets
from the Television and Cable Factbook, published by Warren Communications. We use the
2001 edition, which contains information updated up to November 2000, that is, right until the
2000 elections. We did not collect information for the 1996 period, since we know that in this
period Fox News was not available in any market3. In Appendix A we present an example of a
cable data entry for year 2000. Each observation is identified by an account number, a principal
community, a name, and an owner. A cable company serves up to 100 additional communities
in the neighboring region, all listed in the entry. The contract structure includes the price for
Basic Service and the price for Expanded Basic Service 1, 2, and 3 (when available). Though
basic service programming varies from company to company, the basic service typically includes
channels such as C-Span and ABC Family Channel. In addition, cable companies are required
by law to broadcast local television stations as part of the basic service. Expanded basic service
offers additional channels such as Comedy Central and Fox Sports for an additional fee over
the basic service fee. In our sample, two thirds of the subscribers purchased expanded basic
service when available. Finally, cable companies also offer a la carte channels, such as HBO,
for an additional fee.
Since no electronic version of the cable data exists for year 2000, we collected the most
relevant information from a paper copy. We restricted the collection to the 24 states for which
town-level election data is also available (Appendix Table 1). For all principal communities
in these States, we collected information on communities served, company name, cable owner,
number of subscribers, and price for each tier of contract. In particular, we use the information
on the owner of the cable company to instrument for Fox News availability in a town. For each
cable account, we collected information on whether CNN and Fox News were included in the
cable package and, if so, as part of which contract (Basic, Expanded Basic 1, 2, or 3). We also
collected information on whether at least one of the local TV stations broadcasts news by ABC,
CBS, and Fox, respectively. Finally, we estimated the total number of channels (excluding the
local TV channels) offered by the cable company. Overall, we have such information for over
10,000 towns.
Election data. We collected information on Presidential, US House, and US Senate
election for all the US States for which the data was available. For the year 2000 we mostly
used data from the Federal Election Project (Lublin and Voss, 2001). For the year 1988, we
used the data from the Record of American Democracy (ROAD) project (King et al., 1997).
For most of the other years, and specially for 1996, we obtained the information directly
3Technically, Fox News entered the programming in a small number of markets in October 1996. We assume
that the first month of programming (with low ratings) had no impact.
6
from the Election Division of the Secretary of State of each state4. An additional source of
presidential voting data for the years 1992 and 1996 is the Atlas Election data (Leip, 2004).
Since the information on cable is available at the town level, we have generated a data
set of voting information at the town level. A first group of states–California, Connecticut,
Maine, Massachusetts, New Hampshire, Rhode Island, and Vermont–directly provides voting
information at the town level. A second group of states–Minnesota, Michigan, Montana,
Pennsylvania, Wyoming–provides precinct-level voting information with an indication of the
town to which each precinct belongs; in this case, we just aggregated the voting information
to the town level. Finally, a third group of states–Alaska, Alabama, Arkansas, Iowa, Idaho,
Missouri, Ohio, South Carolina, Tennessee, Utah, Virginia, and Wisconsin–only has precinct-
level voting information, with precinct names that usually include the name of the town.
Examples of precinct names are ”02 - Concord Elem School” and ”Hot Springs Retirement
Hm”. For these states, we recover the town name from the precinct name by elimination of
numbers and common works, such as ”School” and ”Church”5. Overall, 24 US states fall in
one of these groups and have voting information available for at least years 1996 and 2000.
This group of States forms the sample used in this paper. The remaining 26 States either do
not have electronic precinct-level or town-level voting information available, or have numeric
precincts with no conversion table to town available from the State.6 Overall, we have ***
observation for towns.
Matched data. From the matched sample of 10,500 towns, we drop towns that are
missing data on number of channels offered (38 observations). We also drop towns that we
could not match to data on population from the Census 2000 (490 observations). Finally,
we drop observations for which the voting data on presidential elections is missing in either
1996 or 2000 (1,050 observations). The final sample includes 8,927 towns, covering 189 House
districts in 1,121 different counties.
Variables. The main two variables of interest from the cable data are the indicator
variables dFOXk,2000 and dCNNk,2000. The variable d
FOXk,2000 indicates that at least one cable systems in
town k in year 2000 includes Fox News in either the Basic package or one of the Expanded
Basic packages. Similarly, the variable dCNNk,2000 indicates that at least one of the cable systems
in town k makes CNN available. As controls, we use the population of town k in year 2000,
pk,2000, the estimated number of channels available in town k, nck,2000, and the total population
reached across towns by the cable company present in town k, ptk,2000.7
4Mostly this data was available online but in some cases, hard copies or electronic files were sent.5The Stata code that translates precinct names into town names is available upon request.6We did no collect voting information when only available through the individual counties.7For towns with more than one cable system, the number of subscribers refers to the cable system with the
highest number of subscribers. The total number of channels is an estimate. In order to save coding time, the
variable is recorded as the total number of lines in the cable book devoted to the channels in the Basic and
Expanded Basic 1, 2, and 3 Services. We then obtain the estimated number of channels by multiplying this
7
Turning to the voting data, for each town k in year t we consider the number of votes cast
for race j, where j equals Pres. for a Presidential race, House for a US House race, and Sen.
for a US Senate race. We denote the total number of votes cast in race j for the Republican
candidate as V R,jk,t and the total number of votes cast for the Democratic candidate as V D,j
k,t .
We aggregate the total votes cast for other parties into V O,jk,t . The main voting variable in
the paper is the two-party republican vote share vR,jk,t = V R,jk,t /
³V R,jk,t + V D,j
k,t
´.As a robustness
check, we also use the all-party republican vote share v0R,jk,t = V R,jk,t /
³V R,jk,t + V D,j
k,t + V O,jk,t
´.
Both measures are set to missing if the number of votes for either the Republican or the
Democratic candidate is zero, usually reflecting an incumbent running unchallenged.
We also define two measure of voter turnout. The main measure of turnout normalizes the
total number of votes by the population of the town in year 2000: tjk,t = V TOT,jk,t /Popk,2000 is the
Republican turnout in town k in year t for race j. For robustness, we also construct a measure
of turnout that uses as a denominator the number of registered voters in the appropriate year.
Unfortunately, town-level data on number of registered voters is available for only a subsample
of states.
4 Empirical results
4.1 Summary statistics
Column 1 of Table 1 presents summary statistics on the 8,927 towns in our sample. Fox News
is present in the cable programming of 20.30 percent of the towns, while CNN is present in 96.7
percent of the towns. By the year 2000, Fox News has expanded to a significant but limited
number of towns, while CNN is present in almost all towns. The average cable company offers
over 28 channels in the Basic and Expanded Basic programming. The average town population
is 9,838 people, with a median town population of 2,389. The average total population reached
by a cable system is 87,316, with a median of 24,303. The average two-party vote share for
Republicans in Presidential elections is .471 in the 1996 elections and .540 in the 2000 elections.
The Republican vote share is higher than the national average because these averages are not
weighted by population and smaller towns are substantially more likely to vote Republican.
Regarding the geographical distribution of the data, the two regions with the greatest number
of towns in the sample were the Northeast Census region (42.1 percent) and the Midwest region
(42.8 percent). The South (8.8 percent) and the West (6.3 percent) follow.
Columns 2 and 3 show the same statistics for the towns that offered Fox News in their
programming (Column 2) and towns that did not (Column 3). Towns that offered Fox News
were more likely to offer CNN in their programming (99.8 percent versus 95.9 percent) and
had a substantially higher number of channels offered (45.7 versus 24.3). In addition, towns
number by 2.6.
8
that offered Fox News had twice as large a population and were served by cable channels that
reached three times as many people. Turning to the voting data, towns served by Fox News
were somewhat more likely to vote Republican in 1996 (48.1 percent versus 46.9 percent), but
not in year 2000 (54.2 percent vs. 54 percent). Towns in the West and Northeast were more
likely to offer Fox news than towns in the South.
For each congressional district, we compute the share of towns in the district which offered
Fox News in the cable programming. Figure 2a plots the resulting Fox News share for the 188
districts in the sample. Out of these, 133 districts include towns that offered Fox News and
towns that did not. In our differences-in-differences specification with district fixed effects, the
effect of Fox News is estimated exclusively on this subgroup of districts. The median district
in this subsample of 7,091 towns has 19.4 percent of towns offering Fox News. Columns 4
and 5 of Table 1 provide summary statistics for observations in this subsample, respectively
for the subsample of towns with and without Fox News. Towns in this subsample are smaller
in population but have comparable political behavior and geographical distribution as in the
overall sample.
We also consider the distribution of Fox News at the finer geographical level of the county.
As above, we compute the share of towns that offer Fox News within each of the 1,116 counties.
Figure 2b plots this distribution. Only 278 counties out of 1,116 incorporate both towns that
have Fox News and towns that do not. In our empirical specification with county fixed effects,
the effect of Fox News is estimated on this subgroup of counties. Within these counties, 41.9
percent of towns offer Fox News. Columns 6 and 7 of Table 1 provide summary statistics for
observations in this subsample of 3,369 towns, respectively for the subsample of towns with
and without Fox News. Observations in this subsample have similar population and political
data as in the subsample with District variation, and are more concentrated geographically in
the Northeast and less in the South. The population and voting data is more similar across
towns with and without Fox News than in the original sample. This reflects the fact that towns
with and without Fox News in this subsample are close geographical neighbors and therefore
are more closely matched on observables.
4.2 Selection
The empirical results in this paper rely on comparing towns with Fox News in their program-
ming in year 2000 to towns without Fox News. Since the assignment of towns into these two
groups is not exogenous, we investigate the nature of the selection. Table 1 shows that towns
with Fox are substantially more populated, have cable systems with more channels, and are in
somewhat different areas of the country. The most important type of sorting, though, is with
respect to electoral outcomes. Fox News may well have expanded more in Republican areas,
9
since demand for its services is likely to be higher in these areas8. If Republican vote share
in 1996 is correlated with changes in vote share between 1996 and 2000, the estimates of the
differences-in-difference estimates are biased.
We investigate this form of selection by regressing the Republican vote share in presidential
elections in 1996 (that is, pre-Fox News) on the Fox News dummy and a set of controls. We
employ the same empirical specification as in the rest of the paper:
vR,Presk,1996 = α+ βFdFOXk,2000 + βCd
CNNk,2000 + ΓXk,2000 + εk. (1)
In Column 1 are the results of a specification with exclusively a constant and the dummies
for Fox News and CNN. The constant α = .4874 reflects the fact that Democratic party won
the 1996 elections. Given that our sample is not weighted by population, the Republican vote
share in our sample, 48.74, is higher than the actual one in the 1996 Presidential elections. The
estimated βF = .0131 suggests that in towns with Fox News in 2000 the Republican vote share
is 1.31 percentage points higher, a significant difference. The estimated βC , the effect of CNN,
is negative and significant. In Column 2, we add controls for population, total population
in the cable system, and number of channels. In addition, we add state fixed effects. With
these controls, the effect of Fox News changes sign and becomes insignificant. In Column 3
we add district fixed effects and in Column 4 we add both district and county fixed effects. In
both cases there is no evidence that towns with Fox News in 2000 have significantly different
vote share for Republicans in 1996. Towns with CNN have significantly lower Republican vote
share, except with county fixed effects. However, one should not give too much importance
to this result; only 3 percent of towns do not have CNN in 2000, and these towns are highly
unrepresentative. Moreover, unlike the Fox News dummy dFOX , dCNN did not equal zero in
1996. While βF captures the effect of adding Fox News to the programming between 1996 and
2000, βC captures the effect of having CNN over time.
In Column 5 we replicate the specification of column 3 with the addition of an extra control,
the growth in Republican vote share in Presidential elections between 1988 and 2002. Since
town-level data for 1992 is hard to find, this reduces the sample to 3,198 towns. Towns with
Fox News have a (marginally significant) higher Republican vote share by one percentage
point. The effect disappears, both in magnitude and significance, once county fixed effects are
introduced (Column 7).
Overall, we find no evidence that the towns that adopted Fox News by the year 2000 had
significantly different political preferences in 1996. Similarly, we find no evidence that the
towns that adopted Fox News by the year 2000 had different change in vote share between
1996 and 1992 or different turnout in 1996. (results not shown) This suggests that the sorting
of Fox News into towns was not mainly driven by political factors, and alleviates the concerns
about selection on political variables.
8Pew (2004) shows that Republicans are more likely to watch Fox News.
10
4.3 Baseline results
The baseline empirical specification is a standard difference-in-difference estimation. We com-
pare town where Fox News entered the cable market by the year 2000 and towns where Fox
News was not available by the year 2000. We consider the impact of the entry of Fox News
on the change in Republican vote share between 1996 and 2000. This strategy exploits the
timing of the entry of Fox News. By November 1996 Fox News had been launched in only a
few markets, and just one month before the elections. By November 2000 Fox News had an
audience that was smaller but nonetheless comparable to that of CNN. The baseline empirical
specification is
vR,Presk,2000 − vR,Presk,1996 = α+ βFdFOXk,2000 + βCd
CNNk,2000 + ΓXk,2000 + εk. (2)
The controls include fifth-order polynomials in the following variables: population in year
2000, the total population covered by the cable system in year 2000, and the number of
channels in year 2000. Column 1 in Table 3 presents the result of specification (2) without
any controls. The estimated average change in vote share α is .0807, indicating on average
an 8 percentage point increase of Republican vote share for Republicans between the 1996
and the 2000 elections. The estimate for βF suggests that towns with Fox News become more
Democratic by one percentage point relative to towns without Fox News. The estimate for βC
has similar sign and magnitude.
Column 2 presents the results with controls and state fixed effects. The effect of the
controls (not shown) is as follows: increases in town population, total population, and number
of channels all reduce the Republican vote share, with some non-linearities. The introduction
of these controls and state fixed effects renders the coefficient βF on Fox News small (.0023)
and insignificantly different from zero. The point estimate suggests that the introduction of
Fox News increases the Republican vote share by 2 tenth of a percentage point, a very small
effect. We then introduce district (Column 3) and district and county fixed effects (Column
4). In these specifications the identification of βF depends on comparison of neighboring towns
with and without Fox News. In Column 3, the comparison of neighboring towns takes place
within a US House district, a geographical unit inhabited by on average 640,000 people. In
Column 4, the comparison takes place within a county inside a district, a substantially finer
geographical unit. The estimate of the effect of Fox News remains small, -.0017 and +.0015
respectively.
Robustness. In the next Columns, we replicate the results of Columns 3 and 4 with
additional controls and specifications. In Columns 5 and 6 we add the change in Republican
vote share between 1988 and 1992 as an additional control. In this specification, the number
of observations is reduced to 3,198, since town-level voting data in year 1992 is available only
for a limited sample of states. The point estimates of βF are small (and significantly negative
11
in column 5) and comparable to the benchmark ones in Columns 3 and 4. Moreover, the
change in vote share between 1988 and 1992 is not significant. This suggests that it is unlikely
that the null effect of Fox News is due to Fox News settling in areas that are becoming more
Democratic.
In Columns 7 and 8 we consider an alternative specification of the dependent variable.
Rather than using the two-party Republican vote share vR,Presk,t = V R,Presk,t /
³V R,Presk,t + V D,Pres
k,t
´,
we use the all-party republican vote share v0R,Presk,t = V R,Presk,t /
³V R,Presk,t + V D,Pres
k,t + V O,Presk,t
´.
The estimated coefficient βF with district (Column 7) and with county fixed effects (Column
8) is respectively -.00006 and .0015, neither significant. The results of the baseline specification
replicate.
In Table 4 we examine whether the effect of Fox News exposure on the Republican vote
share depends on the geographical area or the previous vote share. In Columns 1 and 2 we
allow the result of Fox News to differ between the four Census regions, New England (the
omitted category), the Midwest, the West, and the South. We do not find any evidence that
the effect of Fox News differs by US region, except for the coefficient on the Midwest that is
marginally significantly lower.
In Columns 3 and 4 we allow the effect of Fox News to differ based on the Republican vote
share in 1996. In particular, we generate a dummy for districts with above-median republican
vote share, that is, vote share above .473. We associate this dummy to all the towns in the
district. The estimates in Column 3 suggest that in Democratic districts Fox News exposure
significantly decreases the Republican vote share by .43 percentage points. In Republican
districts, instead, Fox News exposure increases the Republican vote share by an insignificant
-.43+.53=.1 percentage points. Interestingly, Fox News exposure has a significantly more
positive effect on Republican vote share in Republican areas. This result, however, does not
hold with county fixed effects (Table 4), in which case there is no significant effect of Fox News
in either region and the point estimates are essentially zero.
Overall, across the different specifications the entry of Fox News into a cable market by year
2000 does not appear to have increased the Republican vote share in Presidential elections.
The result is precisely estimated and robust to controls and alternative specifications.
Magnitudes. Any null effect can derive from lack of power. This is not the case, though,
in these specifications. The benchmark estimates in Columns 3 and 4 have standard errors of
.0015 and .0018. This implies tight confidence intervals for the effect of Fox News: (-.0047,
.0013) in Column 3 and (-.0021,.0051) in Column 4. In the specification with district fixed
effects, therefore, the 95 percent confidence interval allows us to reject an effect of Fox News on
the Republican vote share larger than .13 percentage points. In the specification with county
fixed effects, we can reject an effect larger than .0051, half of a percentage point.
Given this, we can compute bounds on how large the Fox News effect on the 2000 Pres-
12
idential election is likely to have been. We assume that the treatment effect estimated here
holds for the 26 States for which we do not have data. The highest treatment effect that is
consistent with the data, as we said, is .51 percentage points. In 2000 Fox News had 50 million
household subscribers9. Therefore, at most half of the US population was in an area with
Fox News exposure, that is, was part of the treatment group. The upper bound on the effect
of Fox News entry on the Republican vote share is .5*(.0051)=.0025, that is, .25 percentage
points, approximately 250,000 votes. This is the upper bound of the Fox News effect, the point
estimates suggest a smaller, potentially zero effect.
4.4 Fox News issues
The baseline specification shows that overall the introduction of Fox News into cable markets
did not significantly affect the propensity of voters to favor the Republican candidate. In
this Subsection, we examine if Fox News entry into a town had an effect on the Republican
vote share for elections that received more direct coverage by Fox News. We use two different
measures of Fox News coverage. First, we compare Presidential elections to House elections.
Fox News gives substantial coverage to Presidential candidates, but no coverage to US House
candidates. Second, we compare Senatorial races that received substantial attention in Fox
News to Senatorial Races that did not. In both cases, the impact of Fox News on the political
orientation of listeners should be higher for races that are explicitly discussed.
President vs. House. In this modified specification, we consider how Fox News affects the
Republican vote share for Presidential election, relative to the vote share for House elections.
The triple-difference specification is:³vR,Presk,2000 − vR,Presk,1996
´−³vR,Housek,2000 − vR,Housek,1996
´= α+ βFd
FOXk,2000 + βCd
CNNk,2000 + ΓXk,2000 + εk. (3)
This specification is equivalent to baseline specification (2) except that the dependent variable
is the double difference of change in vote share for Presidential and House elections. Table
5 reports the findings. In Columns 1 and 2 we present the results of estimation of (3) with
the usual set of controls and district (Column 1) and county (Column 2) fixed effects. The
point estimates for βF are small and insignificantly different from zero, with standard errors
that are 60-80 percent higher than in the baseline specification. In Columns 3 and 4 we
obtain the similar results after including a 1988-1992 time trend. The effect of Fox News on
the Republican vote share is significantly more positive for Republican districts (Column 6),
although it is not significantly positive even for Republican districts.
Finally, we replicate the specification of Columns 1 and 2 using the all-party vote share as
an alternative dependent variable. In the specification with county fixed effects (Column 8)
9Data from www.foxnews.com.
13
we find no effect for Fox News as in the baseline specification. However, in the specification
with district fixed effects only (Column 7) the data suggests a significant and moderately large
effect of Fox News, .88 percentage points.
Overall, we find little support for a differential effect of Fox News exposure on Presidential
elections, relative to House elections.
Featured Senate Races. Senate races are a middle ground between the national Pres-
idential elections and the local House elections. While most Senate races fail to get national
coverage, some do get national attention. In 2000, the Senate race that by a wide margin got
the most coverage in Fox News was the Hillary Clinton-Rick Lazio race in New York State.
The names of these two candidates got 99 overall mentions in the O’Reilly Factor and the
Hannity & Colmes show in the two months prior to the 2000 elections. All the other Sen-
ate candidates running in the 2000 campaign combined got a total of 73 mentions, with Joe
Lieberman getting a lion share of these mentions. In addition, the mentions of the race are to
a large extent critical of Hillary Clinton, and supportive of Rick Lazio.10
We examine whether Fox News had a differential effect for the Clinton-Lazio race. We
denote by dFeature the indicator variable for featured Senate races. We then estimate the
specification
vR,Senk,2000 − vR,Presk,1996 = α+ βFdFOXk,2000 + φFd
FOXk,2000 ∗ dFeature,2000 + βCd
CNNk,2000 + ΓXk,2000 + εk. (4)
where the coefficient φF indicates whether the entry of Fox News has a differential effect for
featured Senate races as compared to non-featured Senate races. Notice that the dependent
variable is the difference of the Senatorial vote share in 2000 and the Presidential vote share in
1996. This specification maximizes sample size: only half of the States with a senatorial race
in 2000 have a senatorial race in 1996.
Table 6 reports the results. The effect of fox News on non-featured races is zero, with
standard errors of .0020 (Column 1) and .0022 (Column 2). The differential effect of Fox News
on the New York race is positive (.0033 and .0059) but not significant. These effects are not
large, but could be consistent with a sizeable effect of one percentage point. When we introduce
as an additional control the change in presidential vote share between 1988 and 1992 we lose
New York state observations, so we cannot identify the effect for the featured races. The effect
on the other races is still not significantly different from zero. Finally, we present the results
with the all-party vote share as dependent variable (Columns 5 and 6). In the specification
with county dummy there is a significant 1 percent effect of Fox News on the New York race.
Overall, we find mixed evidence of a large effect of Fox News for featured races.
10Two excerpts from the ”O’Reilly Factor” of 10/31/2000: ”Now, I am saying to THE FACTOR viewers, and
everyone else who hears what we have to say on this program, that Mrs. Clinton has trouble with honesty”;
”Mr. Gore does have some honesty issues about campaign finance, but they pale beside the deceit factory the
Clintons have set up”.
14
4.5 Voter turnout
So far, we have focused on the effect of Fox News entry on the Republican vote share. However,
the absence of an effect of Fox News on the Republican vote share could occur for two reasons.
First, Fox News entry did not have any effect on average on political behavior. Second, Fox
News entry had a polarizing effect on Republicans and Democrats alike and it increases the
turnout simmetrically. In this latter case, Fox News would affect political behavior, but not
the vote share. We now use measures of turnout to test these hypotheses.
The baseline regression for Republican turnout is:
tPresk,2000 − tPresk,1996 = α+ βFdFOXk,2000 + βCd
CNNk,2000 + ΓXk,2000 + εk, (5)
where tPresk,t is the ratio of the total number of votes cast in a town divided by population
in year 2000. For the analysis of the turnout data, we need to be careful about States in
which the town-level election data is generated from precinct-level data. Since the precinct
names change frequently over time, the precincts aggregated into a town in 2000 could differ
from the precincts that generate the town observation in 1996. While this is unlikely to affect
substantially the estimation of vote share11, it may generate substantial noise in the turnout
data. Therefore, we drop towns with turnout lower than 10 percent or larger than 100 percent,
as well as towns such that the number of precincts in 2000 differs by more than 20 percent
from the number of precincts in 1996. The resulting sample includes 7,764 towns.
Table 7 shows the results. The average change in turnout between 1996 and 2000 is of 3.84
percentage points, reflecting a much tighter race in 2000. The estimates in Column 1 and 2
suggest that Fox News has not had any effect on voter turnout. The point estimates for βF
are small and negative (-.0006 and -.0017), with small standard errors (.0019 and .0022). The
addition of a linear trend of turnout over the period 1988-1992 does not affect the estimates
(Columns 3 and 4). Finally, we find that the impact of Fox News is marginally more positive
for Republican areas in the specification with county fixed effects (Column 6), although not in
the specification with district fixed effects alone (Column 5).
Overall, Fox News entry into a market does not appear to have mobilized new voters.
5 Interpretation
The results in this paper show that one of the most dramatic shifts in media orientation of
the past 20 years has had little or no effect on political outcomes. The entry of Fox News into
local cable markets does not appear to have affected the vote share for Republicans. We fail to
detect an effect also for the races to which Fox News programming gave most visibility. Maybe
even more surprisingly, the entry of Fox News has not affected turnout to the polls.
11All the results above replicate in the smaller sample in which the number of precincts does not vary over
time.
15
A possible interpretation if these results is just that fox News did not have a high enough
viewership to affect voters. However, a Pew survey of a random sample of population (Pew,
2004) shows that in June 2000, right before the elections, 17 percent of the people surveyed
state that they listen regularly to the Fox News Cable station. An additional 27 percent of
people say that they listen to it sometimes. In addition, in June 2000, only 50 percent or
fewer of US households had access to the Fox News cable channel. This implies that, in the
group that has access to cable news, the share of people that regularly listen to Fox News is
34 percent.
We can use this to do a back-of-the-envelope calculation of the treatment on the treated,
that is, how many of the people that actually listened to Fox News changed their voting
behavior. The change in vote share is of at most .5 percent points, and the group treated is 34
percent of the population. Of this 34 percent, approximately one quarter went to the polls in
1996 and voted Republican. This implies that Fox News convinced at most .005/ (3/4) ∗ .34 ≈.02, that is, 2 percent of its listeners that were not already voting Republican to do so. We
should also repeat that this is an upper bound from our estimates, the point estimates of the
findings suggest no effect. A similar calculation for turnout shows that Fox News appears to
have convinced at most 3 percent of people that were not already voting to do so.
Given the evidence on the diversity of the fox News message from the message of all the
other television media (Groseclose and Milyo, 2004), this suggests that media bias does not
have a large impact on people’s behavior.
6 Conclusion
This paper looks at the impact of media bias upon voting. We take advantage of the differential
entry of the conservative Fox News channel in different cable markets. We estimate the impact
of availability of Fox News into the cable programming on the vote share of Republicans and
on voter turnout.
We find, perhaps surprisingly, that there appears to be no effect of exposure to Fox News
on voting behavior. The effect is precisely estimated, so the results are unlikely to be due to
lack of power. A simple calculation using audience data suggests that exposure to Fox News
induced at most 2 percent of its viewers to start voting for the Republican party.
We provide two main explanations of the results. A first one is simply that voters that self-
select into watching Fox News are already voting conservative, and therefore Fox News exposure
cannot have an effect on voting. The second explanation is that viewers are sophisticated
and take into account the bias of the source of information. Viewers know that Fox news is
conservative and rationally use that knowledge when watching the Fox news programming.
16
References
[1] Baron, David, ”Persistent Media Bias”, mimeo, 2004.
[2] De Marzo, Peter, Dimitri Vayanos, and Jeffrey Zwiebel, ”Persuasion Bias, Social Influence,
and Uni-Dimensional Opinions”, Quarterly Journal of Economics. Volume 118, Issue 3,
pp. 909-968, 2001.
[3] Gentzkow, Matthew. “Television and Voter Turnout”, mimeo, GSB Chicago, 2004.
[4] George, Lisa and Joel Waldfogel, “Who Affects Whom in Daily Newspaper Markets?”
Journal of Political Economy, Vol. 11, pp. 765-785, 2003.
[5] George, Lisa and Joel Waldfogel, “Does the New York Times Spread Ignorance and Ap-
athy?”, American Economic Review, forthcoming.
[6] Groseclose, Timothy and Jeff Milyo, ”A Measure of Media Bias”, mimeo, 2004.
[7] Gary King; Bradley Palmquist; Greg Adams; Micah Altman; Kenneth Benoit; Claudine
Gay; Jeffrey B. Lewis; Russ Mayer; and Eric Reinhardt. 1997. “The Record of American
Democracy, 1984-1990,” Harvard University, Cambridge, MA, Ann Arbor, MI: ICPSR.
[8] Kull, Steven; Clay Ramsay; Stefan Subias; Evan Lewis; Phillip Warf. Misperceptions, The
Media, And The Iraq War. The PIPA/Knowledge Networks Poll, 2003
[9] Lee, David S., Enrico Moretti and Matthew J. Butler, ”Do Voters Affect or Elect Policies?
Evidence from the US House”, Quarterly Journal of Economics. Volume 119, Issue 3, pp.
807-859, 2004.
[10] Leip, David. Dave Leip’s Atlas of U.S. Presidential Elections.
http://www.uselectionatlas.org, 2004.
[11] David Lublin and D. Stephen Voss. 2001. ”Federal Elections Project.” American Univer-
sity, Washington, DC and the University of Kentucky, Lexington, KY.
[12] J. Wilson Mixon, Jr., Amit Sen, and E. Frank Stephenson. ”Are the Networks Biased?
Calling states in the 2000 election” Public Choice. Volume 118, pp. 53-59, 2004.
[13] Mullainathan, Sendhil, and Andrei Shleifer. ”The Market for News”, American Economic
Review, forthcoming.
[14] Pew Report, ”News Audiences Increasingly Politicized”, April 2004.
[15] PIPA (Program on International Policy Attitudes), ”Misperceptions, the Media and the
Iraq War” http://www.pipa.org/OnlineReports/Iraq/Media 10 02 03 Report.pdf
17
[16] Prat, Andrea and Stromberg, David. ”State Television and Voter Information”, mimeo,
2004.
[17] Puglisi, Riccardo. ”Being the New York Times: the Political Behavior of a Newspaper”,
mimeo, 2004.
[18] Rabin, Matthew and Joel Scharg, ”First Impressions Matter: A Model of Confirmatory
Bias”, Quarterly Journal of Economics 114(1), 37-82, February 1999.
18
20
All Fox News No Fox News Fox News No Fox News Fox News No Fox News
Fox News 0.203 1 0 1 0 1 0(4.02) 0 0 0 0 0 0
CNN 0.967 0.998 0.959 0.999 0.957 0.997 0.966(0.179) (0.047) (0.199) (0.035) (0.204) (0.053) (0.180)
Channels 28.6 45.7 24.3 45.8 24.3 46.2 25.2(15.9) (17.3) (12.1) (17.5) (12.5) (17.8) (12.7)
Potential Subscribers 87316 190408 61061 163490 50470 159744 64518(190587) (344306) (109548) (287531) (91865) (285236) (106768)
Population 9838 16252 8204 14822 6292 14559 8773(54335) (102328) (32024) (103916) (17184) (106645) (30299)
1996 0.471 0.481 0.469 0.485 0.475 0.479 0.477(0.124) (0.127) (0.124) (0.127) (0.123) (0.129) (0.122)
2000 0.54 0.542 0.54 0.547 0.551 0.541 0.546(0.126) (0.129) (0.126) (0.128) (0.123) (0.130) (0.122)
1996 0.433 0.418 0.437 0.416 0.436 0.414 0.42(0.142) (0.125) (0.146) (0.125) (0.146) (0.117) (0.131)
2000 0.465 0.448 0.47 0.447 0.467 0.446 0.448(0.139) (0.124) (0.142) (0.124) (0.142) (0.118) (0.126)
Northeast 0.421 0.434 0.418 0.456 0.412 0.498 0.558(0.494) (0.496) (0.493) (0.498) (0.492) (0.500) (0.497)
Midwest 0.428 0.432 0.428 0.426 0.447 0.426 0.336(0.495) (0.495) (0.495) (0.495) (0.497) (0.495) (0.472)
South 0.088 0.061 0.095 0.053 0.082 0.029 0.029(0.283) (0.239) (0.293) (0.224) (0.274) (0.167) (0.167)
West 0.063 0.073 0.06 0.065 0.059 0.047 0.078(0.242) (0.261) (0.238) (0.248) (0.236) (0.211) (0.267)
N 8927 1812 7115 1668 5423 1429 1940
Regions
Table 1. Summary Statistics
Notes: Standard deviations in parenthesis. Towns with district variation are towns for which the district that they are in has at least one town that does not get Fox News. Towns with county variation is the same except at the county level. Potential Subscribers is the total population that a cable provider for a given town reaches. Republican 2 party vote share is the votes received by the republican candidate in the presidential election divided by the votes received by both the republican and democratics candidates. Voter turnout is measured by the total number of voters that turned out in a given town divided by the total population of the town.
* significant at 10%; ** significant at 5%; *** significant at 1%
Availability in 2000
Number in 2000
Total Voter Turnout
Rep. 2-Party Vote Share
Towns w/ County Var.Towns w/ District Var.All Towns
21
(1) (2) (3) (4) (5) (6)
0.4874 0.514 0.5014 0.5018 0.4927 0.4954(0.0072)*** (0.0209)*** (0.0185)*** (0.0169)*** (0.0217)*** (0.0208)***
0.0131 -0.0037 0.0018 -0.0021 0.0104 -0.0023(0.0033)*** (0.0038) (0.0036) (0.0041) (0.0061)* (0.0076)
-0.0196 -0.0281 -0.0168 -0.0064 -0.0229 -0.018(0.0074)*** (0.0081)*** (0.0073)** (0.0076) (0.0108)** (0.0110)
Population X X X X X
Potential Subscribers X X X X X
No. of Channels X X X X X
X X X X X
X X X X
X X
X X
0.0024 0.1743 0.3753 0.5795 0.3471 0.5546
N = 8927 N = 8927 N = 8927 N = 8927 N = 3198 N = 3198N
Notes: The observations are 1996 towns for which the data is available. The dependent variable is the two-party republican vote share for the 1996 presidential election (Votes for Dole/Votes for Dole + Votes for Clinton). Fox News 2000 and CNN News 2000 are binary variables that equal one if Fox News and CNN respectively were availbable in the town's local cable package. Potential Subscribers is the population for which the local cable system reaches. The number of channels is the number of channels that are avaible in the cable pacakage. For regressions (5)-(6), the linear trend between 1992 and 1988 presidential election vote share was controlled for. * significant at 10%; ** significant at 5%; *** significant at 1%
District Fixed Effects
County Fixed Effects
1992-1988 Linear Trend
R2
Fox News 2000
CNN News 2000
Fifth Order Polynomials in:
State Fixed Effects
Table 2. Selective Penetration of Fox News
Dependent Variable: Republican vote share in 1996 presidential election
Constant
22
(1) (2) (3) (4) (5) (6) (7) (8)
0.0875 0.128 0.1248 0.1264 0.1226 0.1204 0.1124 0.1142(0.0032)*** (0.0084)*** (0.0078)*** (0.0073)*** (0.0103)*** (0.0098)*** (0.0084)*** (0.0079)***
-0.01 0.0023 -0.0017 0.0015 -0.007 -0.001 -0.0006 0.0015(0.0014)*** (0.0015) (0.0015) (0.0018) (0.0029)** (0.0036) (0.0016) (0.0019)
-0.0168 -0.0035 -0.004 -0.0067 -0.011 -0.018 -0.0067 -0.0101(0.0032)*** (0.0032) (0.0031) (0.0033)** (0.0051)** (0.0052)*** (0.0034)** (0.0035)***
Population X X X X X X X
Potential Subscribers X X X X X X X
No. of Channels X X X X X X X
X X X X X X X
X X X X X X
X X X
X X
0.0091 0.3159 0.4206 0.5968 0.4031 0.5964 0.4881 0.6435
N = 8927 N = 8927 N = 8927 N = 8927 N = 3198 N = 3198 N = 8927 N = 8927
R2
N
Notes: The observations are 2000-1996 differences for towns which the data is available. For columns (1) - (6), the dependent variable is the two-party republican vote share for the 2000 presidential election minus the two-party republican vote share for the 1996 presidential election. For columns (7) & (8) all-party republican vote share is used instead of the two-party republican vote share. Fox News 2000 and CNN News 2000 are binary variables that equal one if Fox News and CNN respectively were part of the town's local cable package in 2000. Potential Subscribers is the population for which the local cable system reaches. Columns (5) & (6) use a linear trend between 1992 and 1988 republican vote share in presidential elecions as an additional control.
* significant at 10%; ** significant at 5%; *** significant at 1%
State Fixed Effects
District Fixed Effects
County Fixed Effects
1992-1988 Linear Trend
Constant
Fox News 2000
CNN News 2000
Fifth Order Polynomials in:
Table 3. The Effect of Fox News on the Difference Between 2000 & 1996 Presidential Election Vote Share
Dependent Variable: Republican vote share difference between 2000 & 1996 presidential elections Two-Party Vote Share All-Party Vote Share
23
(1) (2) (3) (4)
0.0028 0.0042 -0.0043 0.0012-0.0023 (0.0025)* (0.0020)** -0.0023
-0.0068 -0.0103 -0.0041 -0.0067(0.0034)** (0.0035)*** -0.0031 (0.0033)**
-0.0065 0.0031(0.0054) (0.0079)
-0.0058 -0.0064(0.0031)* (0.0036)*
-0.0074 -0.0079(0.0058) (0.0067)
0.0053 0.0006Share In District > .473) (0.0026)** -0.0031
Population X X X X
Potential Subscribers X X X X
No. of Channels X X X X
X X X X
X X X X
X X
0.4884 0.6437 0.4208 0.5968
N = 8927 N = 8927 N = 8927 N = 8927
Notes: The observations are 2000-1996 differences for towns which the data is available. The dependent variable is the two-party republican vote share for the 2000 presidential election minus the two-party republican vote share for the 1996 presidential election. Fox News 2000 and CNN News 2000 are binary variables that equal one if Fox News and CNN respectively were part of the town's local cable package in 2000. Potential Subscribers is the population for which the local cable system reaches. For regressions (1) - (2) region interaction terms are used with the northeast being the ommitted region. For regressions (3) - (4) Fox is interacted with a dummy that equals 1 for all districts which have a Republican vote share above the median for 1996, .473. * significant at 10%; ** significant at 5%; *** significant at 1%
District Fixed Effects
County Fixed Effects
R2
N
Fox * (Average 1996 Rep. Vote
Fifth Order Polynomials in:
State Fixed Effects
CNN News 2000
Fox * South
Fox * Midwest
Fox * West
Table 4. Fox News Effect by Region and by Vote Share
Dep Var: Two-party republican presidential vote share difference between 2000 & 1996
Fox News 2000
24
(1) (2) (3) (4) (5) (6) (7) (8)
0.0022 -0.0015 -0.0042 -0.0086 0.0037 -0.0053 0.0088 0.0019(0.0027) (0.0029) (0.0040) (0.0048)* (0.0053) (0.0051) (0.0029)*** (0.0030)
0.0037 -0.005 -0.0081 -0.0102 0.0015 -0.0057 -0.0019 -0.0095(0.0053) (0.0051) (0.0069) (0.0069) (0.0032) (0.0035) (0.0056) (0.0052)*
0.0018 0.011Share In District > .473) (0.0047) (0.0052)**
Population X X X X X X X X
Potential Subscribers X X X X X X X X
No. of Channels X X X X X X X X
X X X X X X X X
X X X X X X X X
X X X X
X X
0.7065 0.828 0.6697 0.7874 0.7065 0.8281 0.735 0.8564
N = 6178 N = 6178 N = 3040 N = 3040 N = 6178 N = 6178 N = 6625 N = 6625
Notes: The observations are 2000-1996 differences in differences for towns which the data is available. For columns (1)-(6), the dependent variable is the two-party republican vote share for the 2000 presidential election minus the two-party republican vote share for the 2000 house election all minus the same difference for the 1996 elections. For columns (7) & (8) the dependent variable is the same except that it is the all-party republican vote share rather than the two-party republican vote share. Fox News 2000 and CNN News 2000 are binary variables that equal one if Fox News and CNN respectively were part of the town's local cable package in 2000. For regressions (3)-(4) a linear trend using the 1992 - 1988 presidential election vote shares was added as an additional control. For regressions (5) -(6) Fox news is interacted with a dummy that equals 1 for all districts which have a Republican vote share above the median for 1996, .473.
* significant at 10%; ** significant at 5%; *** significant at 1%
County Fixed Effects
1992-1988 Linear Trend
R2
N
Fifth Order Polynomials in:
State Fixed Effects
District Fixed Effects
Fox News 2000
CNN News 2000
Fox * (Average 1996 Rep. Vote
Table 5. The Effect of Fox News on the Difference between 2000 & 1996 Presidential Minus House Vote Share
Dependent Var.: Presidential minus house Republican vote share difference between the 2000 & 1996 elections Two-Party Vote Share All-Party Vote Share
25
(1) (2) (3) (4) (5) (6)
0 0.0001 -0.0056 -0.0008 0.0002 -0.001(0.0020) (0.0022) (0.0036) (0.0043) (0.0020) (0.0022)
0.0002 -0.0042 -0.0135 -0.0195 -0.0016 -0.0053(0.0042) (0.0042) (0.0072)* (0.0070)*** (0.0043) (0.0042)
0.0033 0.0059 0 0 0.0057 0.0107(0.0046) (0.0049) 0.0000 0.0000 (0.0047) (0.0050)**
-0.0006 0.0127 0 0 0.004 0.0133(0.0131) (0.0124) 0.0000 0.0000 (0.0133) (0.0126)
Population X X X X X X
Potential Subscribers X X X X X X
No. of Channels X X X X X X
X X X X X X
X X X X X X
X X X
X X
0.8116 0.8787 0.8862 0.9211 0.839 0.897
N = 7837 N = 7837 N = 2461 N = 2461 N = 7837 N = 7837
Notes: The observations are 2000-1996 differences in differences for towns which the data is available. For columns (1)-(4), the dependent variable is the two-party republican vote share for the 2000 senatorial election minus the two-party republican vote share for the 1996 senatorial election. For columns (5) & (6) the dependent variable is the same except that it is the all-party republican vote share rather than the two-party republican vote share. Fox News 2000 and CNN News 2000 are binary variables that equal one if Fox News and CNN respectively were part of the town's local cable package in 2000. For regressions (3)-(4) a linear trend using the 1992 - 1988 presidential election vote shares was added as an additional control. For all regression, Fox News and CNN are interacted with New York's senatorial race between Hillary Clinton and Rick Lazio. * significant at 10%; ** significant at 5%; *** significant at 1%
County Fixed Effects
1992-1988 Linear Trend
R2
N
CNN * New York Race
Fifth Order Polynomials in:
State Fixed Effects
District Fixed Effects
Fox News 2000
CNN News 2000
Fox News * New York Race
Table 6. The Effect of Fox News on the Difference between 2000 & 1996 Senatorial Races
Dependent Variable: Difference in republican senatorial vote share between 2000 & 1996Two-Party Vote Share All-Party Vote Share
26
(1) (2) (3) (4) (5) (6)
-0.0006 -0.0017 0.0018 -0.0054 -0.0003 -0.0049(0.0019) (0.0022) (0.0034) (0.0044) (0.0024) (0.0029)*
0.0106 0.0078 -0.0017 -0.0064 0.0105 0.0075(0.0039)*** (0.0042)* (0.0061) (0.0066) (0.0038)*** (0.0041)*
-0.0007 0.0063Share In District > .473) (0.0032) (0.0038)*
Population X X X X X X
Potential Subscribers X X X X X X
No. of Channels X X X X X X
X X X X X X
X X X X X X
X X X
X X
0.3745 0.5531 0.2342 0.4392 0.379 0.5597
N = 7764 N = 7764 N = 2834 N = 2834 N = 7764 N = 7764
R2
N
Notes: The observations are 2000-1996 differences in differences for towns which the data is available. The dependent variable is the total votes in 2000 for a given town divided by its population minus a similar measure for 1996. Fox News 2000 and CNN News 2000 are binary variables that equal one if Fox News and CNN respectively were part of the town's local cable package in 2000. For regressions (3)-(4) a linear trend using the 1992 - 1988 presidential election vote shares was added as an additional control. For regressions (5) -(6) Fox news is interacted with a dummy that equals 1 for all districts which have a Republican vote share above the median for 1996, .473. * significant at 10%; ** significant at 5%; *** significant at 1%
State Fixed Effects
District Fixed Effects
County Fixed Effects
1992-1988 Linear Trend
CNN News 2000
Fox * (Average 1996 Rep. Vote
Fifth Order Polynomials in:
Table 7. The Effect of Fox News on the Difference between 2000 & 1996 Voter Turnout
Dependent Variable: Difference between 2000 and 1996 presidential election voter turnout
Fox News 2000