WikiSummary, the Social Science Summary Database

Bartels: Messages received

From WikiSummary, the Free Social Science Summary Database

 

Bartels. 1993. Messages received: The political impact of media exposure. APSR 87: 267-285.

The Puzzle

Using survey data, political scientists have found that the media's effects on public opinion are "minimal." But this finding is at odds with the opinions of political campaign managers and the findings of experimental research (Iyengar and Kinder 1987). Why are political scientists unable to empirically demonstrate media effects on public opinion?

The Model

The problem with assessing media effects is to distinguish the effect of media exposure from the effect of previously held opinions and information. Using a Bayesian updating model allows the weighting of old and new information in people's current opinions. Only new information that contradicts prior opinions can produce observable opinion change in direct proportion to the strength (uncertainty) of the prior opinions.

Research Design

Data and Variable Measurement

Uses 1980 NES data where a panel of respondents was interviewed at three time points: right before the presidential primary season, between the end of primary season and the national nominating conventions, and during the first month of the general election campaign.

This repeated measurement allows Bartels to estimate the magnitude of measurement error using a variant of the Wiley and Wiley model, and then incorporates this measurement error into errors-in-variable parameter estimates. He then compares a simple OLS model's estimates with the errors-in-variables estimates.

  • X: Bartels uses the respondents' answers to questions on TV news viewing and newspaper readership as proxies for media exposure (the independent variable),
  • Y: and uses a bunch of questions on candidates (thermometer ratings) and issue positions as the dependent variable.

Findings

The OLS estimates of media impact were substantially smaller than the errors-in-variables estimates of media impact, suggesting that the failure to account for measurement error was responsible for some of the "minimal effects" results of earlier studies.

Also, allowing for the effects of measurement error shows that opinions are much more stable over the course of a campaign season than previously thought. This means that new information from the media must compete with a much greater mass of prior information than the earlier studies supposed.

Conclusions

  • Adjusting for measurement error increases the apparent impact of media exposure on opinion change in a presidential campaign setting.
  • Analysis which focuses on opinion change over short periods of time (like a campaign season) will find modest media effects, because opinions at the beginning of the campaign season are already strongly held and media messages during the campaign season are only occasionally sharply inconsistent with those preexisting opinions. Bayesian logic would predict modest or no change under these circumstances.
  • This suggests that in trying to estimate media impacts on opinions, political scientists should look for candidates or issues where prior opinions are weak (such as with new issues).
    • One reason earlier studies found no impact on public opinion from media exposure was their focus on issues with stable prior opinions.
  • This also suggests that studies of media impacts should take place over long time periods. One reason earlier studies found no impact of media exposure is that there time periods were to short.