More on the Groseclose-Milyo Measure of Media Bias

by John Sides on August 3, 2011 · 3 comments

in Media

This research note engages the current research on measuring media bias. I present a reanalysis of the results found in Groseclose and Milyo (2005) and show that the original parameter estimates of the ideological positions of media outlets are not stable over time. Using the same data but analyzed over diff erent periods of time, I fi nd a diff erent conclusion than the previous article. I examine four-year rolling time periods and find that the data produce diff erent parameter estimates in the early- to mid-1990s as compared to after 2000, with all analyzed outlets appearing more moderate or conservative in later time periods. My results indicate that the estimated positions are sensitive observations in the data and the time period of observation of the outlet.

From a forthcoming paper by John Gasper.  Find it here (pdf).  Here is a bit more:

The GM fi nding of a strong liberal bias seems suspect in light of my reanalysis. The fi ndings presented here indicate liberal media outlets during the early 1990s, but conservative media outlets by the beginning of the next decade.

And the results also change if one interest group (the National Taxpayers Union) is excluded from the estimation of the positions of media outlets:

Performing a pooled analysis without using citations of the National Taxpayers Union (NTU), the most frequently cited policy group by members of Congress, resulted in an average drop in nearly nine points on the ADA scale. In other words, excluding citations of the NTU from both members of Congress and media outlets results in all media outlets appearing substantially more conservative.

For more, see the paper as well as Andy’s post on the Groseclose book (and the comments), the Puglisi-Snyder paper, Andy’s comments on that paper, Brendan Nyhan’s post on the original Groseclose-Milyo paper, and my earlier post on that paper and the general subject of ideological bias in the news media.

{ 3 comments }

Erik Nisbet August 3, 2011 at 11:04 am

The idea that liberal/conservative bias of media outlets is contextual as this reanalysis suggests brings to mind the work by Bennett and others on “indexing” – if there is bias in the media, they would argue that it is pro-govt/pro-elite consensus – thus, a Democratic administration in the 1990s would lead to possible “liberal” bias and a Republican administration in the 2000s would lead to possible “conservative” bias. On a larger point, if you look at the scholarship on media bias writ large (ie. communication, media sociology, etc.), and not the rather narrow sliver found in political science – most scholarship would paint this is specific debate about whether the media is politically “liberal” or “conservative” as simplistic and childish and of more partisan than scholarly interest. “Bias” comes from the assymetrical distribution of power and a complex set of social and institutional interactions – whether political, economic, social, etc. Bias manifests in rather complex ways and is not easily categorized as simply “liberal” or “conservative”

Adam August 3, 2011 at 9:27 pm

I agree, Erik. It just occurred to me yesterday that indexing would be a useful lens through which to critique the reliance on congressional composition to measure media bias. In effect, the measure itself is “indexed” to the elite consensus, thus rendering it completely unable to measure the “biases that matter” (in Bennett’s words) that plague the American press. I also thought that the theory suggests the possibility that the GM measure would be extremely unstable over time — and I’m glad someone beat me to the empirical verification of that hunch.

Jestak August 4, 2011 at 5:47 pm

The Gasper paper confirms a point I made in the comments on Andrew Gelman’s post. Even if you accept that Groseclose & Milyo’s basic methodology is a valid approach to identifying media bias (and I don’t), their results have a huge sensitivity problem.

Comments on this entry are closed.

Previous post:

Next post: