The Fox News Eect:Media Bias and Voting S. DellaVigna and E. Kaplan (2007) Anna Airoldi Igor Cerasa IGIER Visiting Students Presentation March 21st, 2014
Research Questions Does the media have an impact of voting behavior?
Research Questions Does the media have an impact of voting behavior? Can biased information alter individuals' beliefs?
Research Questions Does the media have an impact of voting behavior? Can biased information alter individuals' beliefs? Are individuals able to lter out information bias that is repeated over time?
Answers Rational Expectations Theories Voters lter out bias in reporting without being persuaded on average (Bray and Kreps, 1987)
Answers Rational Expectations Theories Voters lter out bias in reporting without being persuaded on average (Bray and Kreps, 1987) Behavioral Theories Voters are subject to media persuasion (De Marzo, Vayanos and Zwiebel, 2003)
Previous studies Field Experiments (Green and Gerber, 2004) Households are randomly selected and assigned to right-wing or left-wing newspapers The eect is evaluated in a post-election survey Increase in the shares of Democratic voters
Previous studies Field Experiments (Green and Gerber, 2004) Households are randomly selected and assigned to right-wing or left-wing newspapers The eect is evaluated in a post-election survey Increase in the shares of Democratic voters Lab Experiments (Ansolabehere and Iyengar, 1995) Experimental subjects are exposed to thirty-second political advertisements supporting a candidate Beliefs and voting intentions at the end of the experiment Positive eect on the stated voting share
Previous studies (cont.) Surveys (Kull, Ramsay and Lewis (2003) Use of surveys to assess the impact of the media Fox News watchers are more likely to believe that weapons of mass destruction were found in Iraq in 2003 (33% v. 22%)
Consequences Dierent explanations allows for dierent outcomes Deregulation of media markets may have a large impact on political outcomes Answers are a matter of empirical results
Consequences Dierent explanations allows for dierent outcomes Deregulation of media markets may have a large impact on political outcomes Answers are a matter of empirical results Natural Experiment Impact of the introduction of Fox News Channel on Republican vote share in 1996 and 2000
The Fox News Channel "Fox News" was introduced in October 1996
The Fox News Channel "Fox News" was introduced in October 1996 24-hours cable news channel
The Fox News Channel "Fox News" was introduced in October 1996 24-hours cable news channel News broadcast had only a small share
The Fox News Channel "Fox News" was introduced in October 1996 24-hours cable news channel News broadcast had only a small share Signicantly to the right of all the other mainstream television networks (Groseclose and Milyo, 2005)
The Fox News Channel "Fox News" was introduced in October 1996 24-hours cable news channel News broadcast had only a small share Signicantly to the right of all the other mainstream television networks (Groseclose and Milyo, 2005) Strong eect on the available political information
The Cable Industry Local natural monopoly due to the xed cost of laying cables
The Cable Industry Local natural monopoly due to the xed cost of laying cables Only 10 % of cities has more than 1 companies
The Cable Industry Local natural monopoly due to the xed cost of laying cables Only 10 % of cities has more than 1 companies Technological costraint on the number of channels
The Cable Industry (cont.) By the year 2000, Fox News was present in 20% of the sample towns accounting for 34% of the population Fox News also distributed short news segments to local TV Stations aliated to Fox Broadcasting
Proportion of town with Fox News in each county RED = 0 BLUE = 1
The Data Local cable companies => Paper copy of the Television and Cable Factbook, 2001 edition. No info about 1996. Election Data => Election Division of the Secretary of State for each state + Federal Election Project (year 2000) + ROAD Project (year 1988) + Atlas Election Data (year 2004). Census Data => For the 28 US states collect demographics from the 2000 and the 1990 census at the level of Place.
The Data (cont.) SAMPLE = 28 US states, the ones for which aggregation at the town level is possible. 1 Match cable, election and Census data by the State, county and town name 2 Drop towns with multiple cable systems (some with FN some without), not oering CNN or with likely voting data problems => Final sample of 9256 towns with comparable Fox News availability and lower Republican vote share relative to the initial sample.
Selection Need to investigate the nature of selection and estimate which town-level variables predict the availability of Fox News If Republican areas were becoming more Republican between 1996-2000 than the Fox News Eect may just be capturing political trends.
Linear Probability Model Where: d F k,2000 OX res = α + βvr,p k,1996 + β res T tpk,1996 + Γ 2000X k,2000 + d F k,2000 OX +Γ 00 90 X k,00 90 + Γ C C k,2000 + ε k is equal to 1 if all cable systems in town k include Fox News and to 0 if no cable system includes Fox News; R,P res vk,1996 t P k,1996 res is the two-party Republican vote share; is the voter turnout, ie the log of votes cast as share of population; Γ are the coecients of controls for census and features of the cable system
Once we control for geographic heterogeneity and size of the cable system, availability of Fox News is uncorrelated with both political outcome and demographic characteristics; Conditional on cable market size and within country/districts, assignment to towns within an area is RANDOM!
A Natural Experiment Random assignment of a treatment (FOX NEWS) to a population (The American population) thanks to idiosyncratic dierences in access due to decentralization of cable industry; Can be used to study the eects on presidential elections, voter turnout, persuasion rates; Since Fox News is signicantly to the right of all other mainstream television networks, signicant eect on the available political information in that cable market.
Presidential Elections - Regression Framework Evaluate the change in Republican voting share from 1996 and 2000 elections R,P res νk,2000 νr,pres k,1996
Presidential Elections - Regression Framework Evaluate the change in Republican voting share from 1996 and 2000 elections R,P res νk,2000 νr,pres k,1996 Compare towns that had Fox News in 2000 d F OX k,2000 = 1
Presidential Elections - Regression Framework Evaluate the change in Republican voting share from 1996 and 2000 elections R,P res νk,2000 νr,pres k,1996 Compare towns that had Fox News in 2000 d F OX k,2000 = 1 With respect to towns which did not d F OX k,2000 = 0
Presidential Elections - Regression Framework (cont.) Baseline specication R,P res νk,2000 νr,pres k,1996 = α + β F d F k,2000 OX + Γ 2000X k,2000 + +Γ 00 90 X k,00 90 + Γ C C k,2000 + ε k
Presidential Elections - Regression Framework (cont.) Baseline specication R,P res νk,2000 νr,pres k,1996 = α + β F d F k,2000 OX + Γ 2000X k,2000 + +Γ 00 90 X k,00 90 + Γ C C k,2000 + ε k Control variables refer to town-level demographics in absolute values (X k,2000 ), changes (X k,00 90 ) together with cable variables (C k,2000 )
Presidential Elections - Results
Presidential Elections - Results
Robustness
Robustness
Robustness
Interactions
Interactions
Interactions
Placebo Specications
Voter Turnout Two possible explanations for the impact of Fox News on the Republican vote share
Voter Turnout Two possible explanations for the impact of Fox News on the Republican vote share 1 Fox News entry convinced Democratic voters to vote Republican 2 Fox News attracted new Republican voters
Voter Turnout - Baseline Regression We estimate the following equation t P k,2000 res res tpk,1996 = α+β F d F k,2000 OX +γ[log(p op k,2000) log(p op k,1996 )]+ +Γ 2000 X k,2000 + Γ 00 90 X k,00 90 + Γ C C k,2000 + ε k
Voter Turnout - Baseline Regression We estimate the following equation t P k,2000 res res tpk,1996 = α+β F d F k,2000 OX +γ[log(p op k,2000) log(p op k,1996 )]+ +Γ 2000 X k,2000 + Γ 00 90 X k,00 90 + Γ C C k,2000 + ε k Where t P k,t res T OT,P res = log(vk,t ) P op k,t refers to the voting-age town population in year t
Voter Turnout - Results
Voter Turnout - Results
Senate Elections Is the Fox News eect candidate-specic or a general ideological shift?
Senate Elections Is the Fox News eect candidate-specic or a general ideological shift? We test the hypotheses using data from Senate Elections
Senate Elections Is the Fox News eect candidate-specic or a general ideological shift? We test the hypotheses using data from Senate Elections 1 Senate races that failed to get national coverage (72 mentions in total) 2 The Clinton-Lazio race in New York State (99 mentions)
Senate Elections - Regression Framework We estimate the following equation ν R,Sen R,P res k,2000 = α + α P νk,1996 + β F d F k,2000 OX + φ F d F k,2000 OX d NY + +Γ 2000 X k,2000 + Γ 00 90 X k,00 90 + Γ C C k,2000 + ε k Where β F indicates the eect of Fox News on Senate races other than New York and φ F indicates the dierential eect for the featured New York race
Senate - Results
Senate - Results
Persuasion Rate PERSUASION RATE = try to evaluate the magnitude of the Fox News Eect by estimating the share of audience that was convinced by Fox News To do so, compare treatment town T, where Fox News is available, with control towns C, where Fox News is not available via cable.
Persuasion Rate - The Model Hypothesis: v j = r + (1 r)e j f r + d + (1 r d)e j f 1 Since towns have similar political outcomes in the pre-fox News period conditional on controls, we assume that d and r are the same in T and C 2 Exposure e j in town j to Fox News is independent of political aliation => PROBLEMATIC
Persuasion Rate - The Model (cont.) With some calculations, we get that v T v C = (e T e C )fd t C t T The implied persuasion rate is f = v T v C (1 r)t C t T (e T e C )(1 r) d The rst term is the inuence rate for treated population, that is the shift in republican vote share due to the availability of Fox News via cable over the share of population at risk of treatment. Second term: dierential convincing eect of Fox News on a Democrat and a non-voter. Only when a democrat shifts the democrats lose a vote. The larger the ratio of non-republicans to Democrats, the bigger the convincing impact for a given vote share change.
Estimating Audience Scarborough research = representative panel of households to collect demographic variables and two audience measures for each TV channel surveyed. Recall = share of respondents who answer yes to the question on whether they watched a given channel in the past seven days Diary = share of respondents who watched a channel for at least one full half-an-hour block according to a seven-day diary of TV watching. SAMPLE: 105,201 respondents from August 2000- March 2001. Using only the subsample of respondents for wich we observe a zipcode of residence that can be matched with cable data on availability of Fox news and aggregating it at town level, get a sample of 528 towns.
Estimating Audience (cont.) e T e C is the dierential exposure e F k OX = α + β F d F k,2000 OX + Γ 2000X k,2000 + +Γ 00 90 X k,00 90 + Γ C C k,2000 + ε k Where: e F k OX = fraction of town k residents in the Fox New audience; = availability of Fox News in town k; d F OX k,2000 β F = dierential Fox News diary audience due to Fox News' availability via cable in town k.
Estimating Persuasion Rates Using the expression: f = persuasion rate, where: v T v C (1 r)t C t T (e T e C )(1 r) d we calculate the e T e C is the coecient β F of the previous regression; t C,t T,r and d are evaluated using sample averages of the 2000 elections, weighted by total votes cast in 2000. Recall Multiply β F by a conversion rate that is the ratio of the CNN recall audience to the CNN diary audience. Estimates for dierential exposure are 12% and 8% with district and county xed eects. The persuasion rate is 3.39% with congressional district xed eects, 8.27% with county xed eects. Diary The persuasion rate with congressional district xed eect is 11.62% and 28.29% with county xed eects.
Explanations Estimates imply that Fox News convinced a signicant portion of its audience to vote Republican. There are three explanations for this nding: Endogeneity bias = ndings might be spurios, induced by entry of Fox News in towns that were independently becoming more conservative. This explanation is not consistent with the fact that entry of Fox News does not predict political voting trends between 1992 and 1996, but only the ones between 1996 and 2000. Rational learning = if voters are uncertain about the bias of Fox News, exposure will have a temporary eect on beliefs and voting. However, by 2000 the conservative slant of Fox News should have been clear. Moreover, the eect seems to be persistent over time.
Explanations (cont.) Nonrational Persuasion = viewers do not discount media bias strongly enough and therefore are subject to nonrational persuasion upon exposure => permanent alteration of voting behaviour. This is associated with a general ideological shift towards the right.
Main Findings Fox News had a signicant impact on the 2000 elections: according to estimates, it increased Republican vote share in presidential election by 0.4-0.7%. It was therefore decisive in the close presidential election, which was decided by around 500 votes in Florida; Fox News impact extends to general political beliefs and is stronger where the population "at risk" is larger; Fox News increased turnout to the polls; Fox News conviced between 3 and 28% of its non-republican viewers to vote Republican; => Media can have a sizeable political impact
Final Observations The assumption that exposure to Fox News is uncorrelated with political aliation is problematic. If Republicans self-selected into watching Fox News the persuasion rate of non-republicans could be biased downwards; The time persistence of the eect is not really well explained in the paper; This paper presents interesting evidence on the voting of persuasion-biased agents, which systematically underestimate the extent of the bias present in the information they receive when updating their beliefs.
Four hostile newspapers are more to be feared than a thousand bayonets (Napoleon Bonaparte) Whoever controls the media, controls the mind (Jim Morrison)