People who learn that a newspaper does not suppress information exhibit a lower demand for news from it; the idea that people read partisan news because they see those papers as more informative seems wrong

Do People Value More Informative News? Felix Chopra, Ingar Haaland, Christopher Roth. March 2, 2019. https://www.briq-institute.org/wc/files/people/chris-roth/working-papers/do-people-value-more-informative-news.pdf

Abstract: We examine how people’s perceptions of media bias affect their demand for news. Drawing on a large representative sample of the US population, we measure and experimentally manipulate people’s beliefs about the extent to which newspapers suppress information. Inconsistent with the“more-information-is-better principle,” we find that people who learn that a newspaper is less likely to suppress information have a lower demand for news from this newspaper. Our results demonstrate that people have a demand for biased news, consistent with a desire to confirm pre-existing beliefs.

Keywords: Information, Belief polarization, Media Bias, News Consumption,Motivated Beliefs

---
1 Introduction

What drives people’s demand for news? A core principle in economics is that more information is always better. While people’s demand for news articles should thus be strictly increasing in the informativeness of the news, a large literature has documented that newspapers report news in a biased way by slanting their news stories towards the beliefs of their readers (Gentzkow and Shapiro,2010). There are several ways to rationalize why people tend to read slanted news (Xiang and Sarvary, 2007). First, it could reflect a desire for better informationas they perceive news that are closer to their prior beliefs as more informative(Gentzkow and Shapiro, 2006). Second, it could reflect that people have other motives for reading the news that conflict with expanding their knowledge. For instance, people might receive utility from reading news that confirm their pre-existing beliefs (Golman et al., 2016; Loewenstein and Molnar, 2018).

Causally identifying people’s people’s motivation for reading news is difficult. First, to understand people’s motivations for reading biased news articles, oneneeds data on subjective perceptions of biases in reporting. Second, one needsexogenous variation in these perceptions to rule out omitted variable bias andreverse causality. For example, people may distort their stated beliefs to justifytheir news consumption habits. Third, one needs to measure people’s demandfor real-world news and their actual consumption of this news, holding constanttheir information set about news articles. We address these challenges by usingan experimental approach with real news articles which allows us to test whetherconsumers indeed value more informative news in a setting with high externalvalidity.

Drawing on a large representative sample of Americans, we first elicit peo-ple’s beliefs about the extent to which theNew York Timessuppresses information. For that purpose, we tell our respondents that the Congressional Budget Office(CBO), Congress’s official nonpartisan provider of cost and benefit estimates for legislation, published a report about the “Trump Healthcare Plan” (the AmericanHealth Care Act of 2017). We then tell them that the CBO estimated that thiswould (i) decrease the federal deficit by $119 billion and (ii) leave 23 millionmore people uninsured. We truthfully tell our respondents that Republicansclaimed that the the plan would decrease the federal deficit—but not increase thenumber of people without health coverage—while Democrats claimed that theplan would not decrease the deficit and increase the number of people withouthealth coverage. Subsequently, we ask our respondents to estimate the percentchance that theNew York Timesreported only the figure on the number of unin-sured people, only the figure on the deficit decrease, or both figures. This allowsus to quantify people’s beliefs about the extent of media bias in theNew YorkTimes. To introdude exogenous variation in people’s perceptions of media bias,we inform a random subsample of our respondents that theNew York Timesreported both estimates from the CBO. Finally, we measure our respondents’demand for news from theNew York Timesby asking them whether they wouldlike to read an article in the newspaper about the Trump Tax Plan based onestimates from the CBO. The “more-information-is-better principle” predictsthat people’s demand for news about the CBO should increase for respondentswho learn that the newspaper is less likely to suppress information from CBOreports.

The key finding of this paper is that respondents who learn that the NewYork Times does not suppress information significantly reduce their demand for reading an article in this newspaper by 3.4 percentage points. This corresponds to a reduction in the demand for news of 12 percent. The time spent reading the article does not vary significantly across treatment arms, suggesting that the treatment did not affect how carefully people read the article. The reduction indemand for news is driven by respondents who initially thought that the New York Timeswas more likely to suppress information and is absent for respondents with more accurate pre-treatment beliefs about the extent of media bias in the2 New York Times. Consistent with models of motivated beliefs, our results aredriven by respondents who—in light of their prior beliefs about the directionof the bias in reporting and their political affiliation—have a stronger motive toavoid news from an unbiased source. For example, among Republican-leaning respondents the reduction in the demand for news is driven by those respondents who initially thought that the New York Timesis more right-wing biased.

We leverage two tailored measures of beliefs about newspaper reportingto shed light on mechanisms. We provide evidence that treated respondents significantly update their beliefs about the biasedness of the reporting of the  NewYork Times. Our treated respondents are 6.9 percentage points more likely to think that theNew York Times does not suppress any information about the CBO reporton the Trump Tax Plan. Respondents are also 3.7 percentage points less likely to think that theNew York Times did not cover a CBO report highlighting the negative budget consequences of granting citizenship to young undocumented immigrants. We also provide evidence that our results are inconsistent witha series of alternative explanations: Respondents do not update their beliefsabout the technicality of reporting, the complexity of the article, or about thecharacteristics of the CBO. Several patterns in our data are inconsistent withalternative mechanisms, such as cognitive constraints, uncertainty about sourcequality, curiosity, and motives for diversifying news sources.

We contribute to the literature on media bias (Allcott and Gentzkow, 2017;DellaVigna and La Ferrara, 2015; DellaVigna and Kaplan, 2007; Enikolopov etal., 2011; La Ferrara et al., 2012; Gentzkow and Shapiro, 2006, 2010; Gentzkowet al., 2015, 2018; Gerber et al., 2009; Mullainathan and Shleifer, 2005; Qin etal., 2018) and the demand for slanted news (Durante and Knight, 2012; Garzet al., 2018). Gentzkow and Shapiro’s (2010) seminal work introduces a newindex of media slant that measures the similarity of a news outlet’s language tothat of a congressional Republican or Democrat. Their model-based estimatesreveal that readers have a strong preference for like-minded news, but this pattern3 is consistent both with rational Bayesian updating about the informativenessof news (Gentzkow and Shapiro, 2006) and a behavioral preference for beliefconfirmation (Golman et al., 2016). We contribute to this literature by providingthe first causal evidence on the question of whether people value more informativenews. Specifically, we provide evidence that people who learn that the New York Times does not suppress information exhibit a lower demand for news from this newspaper. Our results are inconsistent with the idea that people read partisannews because they perceive partisan newspapers as more informative, as proposedby Gentzkow and Shapiro (2006).1


6    Conclusion

Our paper provides novel evidence on whether people value more informativenews. The main finding of this paper is that respondents who learn that theNewYork Timesdoes not suppress information reduce their demand for articles fromthis newspaper. This is inconsistent with the normative benchmark prediction ofthe “more-information-is-better principle.” Our results are driven by individualswith initially larger biases in beliefs about the extent of media bias, and thosewho in expectation should receive the largest negative belief utility shock when reading an unbiased article. Our empirical findings are consistent with models of motivated beliefs according to which people mainly consume news in order to confirm their prior beliefs,and inconsistent with models according to which people mainly consume news to receive better information. Our findings have important policy implications: Our evidence suggests that transparency about media bias might backfire and actually increase political belief polarization by shifting people’s consumption of news towards more biased sources.