Local media tone, economic conditions, and the evaluation of US governors

ABSTRACT

Connections between media coverage, economic conditions, and performance evaluations of political leaders have seldom been explored in specific local media markets, due to the challenges of measuring media tone and content for a large number of media outlets or markets. In this paper, we develop a measure of media tone by comparing the economic evaluations of local media consumers and national media consumers within the same media market. We then use this measure to evaluate the relationship between media tone and objective economic conditions. We find that positive media tone increases the probability that individuals will approve of the governor’s performance in office, and that tone also attenuates the negative relationship between unemployment and gubernatorial approval.

Acknowledgements

The authors are grateful for the assistance and feedback from the following individuals: Matthew Jarvis, Steve Rogers, Meredith Crane, Makayla Wendland, and Christopher Wlezien.

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes on contributors

Nicholas Clark is an assistant professor in the Department of Political Science at Susquehanna University. His research interests focus on voting behavior, public opinion, and elections in advanced industrial democracies.

Todd Makse is an assistant professor in the Department of Political Science and International Relations at Florida International University. His research interests include state politics, legislative politics, redistricting, and political behavior.

Notes

1 The same principle has been extended to the local level. For example, local officials are often held accountable for their performance on relevant metrics, such as school board officials being rewarded or punished based on the performance of district schools (Berry and Howell 2007).

2 Like Reeves and Gimpel (2012), we find nearly identical results using an ordered logit model.

3 In fact, Heckman selection models indicate that the two processes are, in fact, non-independent. Many of the same factors which predict media choice also predict national economic evaluations, and the null hypothesis of independent equations is easily rejected (p < .001). Since we are only interested in predicting national economic evaluations in order to create this metric, and not to test theoretical claims about the national evaluations per se, modeling the two types of media consumers separately accounts for the selection issue without modeling it directly.

4 Some studies have argued theoretically and demonstrated empirically that it is change in unemployment, rather than the contemporaneous level of unemployment, that should influence retrospective voting. However, we find no evidence of this pattern in the 2010 CCES data. Even before accounting for the factors that are the focus of this paper, the correlation between year-over-year change in state-level unemployment and gubernatorial approval is positive (r = 0.09), not negative as would be expected.

5 While other studies have examined other economic indicators such as GDP change, gas prices, and foreclosure rates (Reeves and Gimpel 2012; Ansolabehere, Meredith, and Snowberg 2014), unemployment rates are still the most common metric in the literature on gubernatorial evaluations.

6 We considered other county-level covariates such as the “normal vote”, operationalized here as the two-party presidential vote from 2008, or a “friends-and-neighbors” effect (Key 1949; Rice and Macht 1987), measured with an indicator of whether the county is the governor’s home county (the county they lived in at the time they were elected governor).

7 We also considered alternative models that measured state-level unemployment, rather than county-level unemployment. In these models, we found support for Hypotheses 1 and 2, but no interactive effect between unemployment and media tone.

8 Although this measure is also used in the construction of the economic portrayal measure, there is no risk of overlap between the two measures since the media market-level measure is aggregated across many individuals and compared to predicted scores, as described in the previous section. In fact, the correlation between the two measures at the individual level is negative (r = −0.10).

9 For the sake of comparison, we also find a similar correlation between national economic evaluations and approval of the state legislature.

10 Pure independents are coded as a “0” on this measure, and a separate dummy variable for these individuals is included in the model. Republican and Democratic-leaners are treated as partisans.

11 Although the partisanship variables should capture most of the propensity of individuals to support governors of their own party, we also considered the possibility that the effects of demographics would differ depending on the governor’s party, insofar as those demographics are capturing some variance associated with partisanship. For only two of these variables (age and education) did we find such an effect. Thus, we also include interactions between each of these variables and the GOP governor dummy.

12 Of course, it is possible that even individuals who do not consume local media are influenced, at least indirectly, by the tone of local media portrayals of the economy. Moreover, the CCES question asks respondents only about their consumption of local and national media in the past 24 hours, allowing the possibility that people consume a given type of media in general, but not in the past 24 hours. Thus, while we should expect the direct and conditioning effects of local media to be weaker for national media consumers, we need not expect a zero effect.

13 We also find evidence that the effects of media portrayals, unemployment, and the interaction between the two differ substantially across Democratic and Republican governors. For Democratic governors, media portrayals have a significant direct effect but do not condition the relationship between unemployment and gubernatorial approval. For Republicans, the interactive effect is statistically significant while the main effect is not.

14 Note that in discussing the results (and presenting Figures 1 and 2), we transform unemployment rates and economic portrayal scores back to their uncentered values.

15 This finding – differential effects when unemployment is high and low – is consistent with the broader research on economic voting. For example, Van Der Brug, Van Der Eijk, and Franklin (2007) find a similar ceiling effect regarding the effect of economic conditions on support for governing parties. That effect may well be due to uneven media coverage of the economy; the media tends to cover the economy and pre-benchmark more during periods of decline (Anduiza and Marinova 2015; Kayser and Peress 2015).