University of Chicago professors Matthew Gentzkow and Jesse Shapiro propose an answer.
The study by Gentzkow and Shapiro has been around for a while in working paper form, though I only came across it on Friday when the latest issue of Econometrica in which it was finally published arrived in my mailbox.
Gentzkow and Shapiro propose to measure the slant of a particular newspaper by searching speeches entered into the Congressional Record and counting the number of times particular phrases were used by representatives of each party, mechanically identifying phrases favored by one party over the other. For example, a Democrat is more likely to use the phrase “workers rights” whereas a Republican is more likely to use the phrase “human embryos”. They then counted the number of times phrases of each type appeared in a particular newspaper to construct an index of the political slant of that newspaper. The Gentzkow-Shapiro index of slant (shown on the vertical axis in the diagram below) has a reasonable correlation with subjective measures such as ratings assigned by users of Mondo Times (horizontal axis). For example, both measures agree that the Washington Times is one of the most conservative papers and the Atlanta Constitution is one of the most liberal newspapers.
Gentzkow and Shapiro then asked what other factors help explain a newspaper’s slant. They found that the most important variable is the political orientation of people living within the paper’s market. For example, the higher the vote share received by Bush in 2004 in the newspaper’s market (horizontal axis below), the higher the Gentzkow-Shapiro measure of conservative slant (vertical axis).
On the other hand, the politics of the paper’s owner seem to matter much less. For example, once one controls for geographic factors, there is no statistically significant correlation between the newspaper’s own slant (horizontal axis below) and average slant of papers in other communities owned by the same owner (vertical axis).
Gentzkow and Shapiro conclude that papers to some degree are just giving their readers what the readers want so as to maximize the newspapers’ profits.
One detail I’m curious about is how slant gets implemented at the ground level by individual reporters. My guess is that most reporters know that they are introducing some slant in the way they’ve chosen to frame and report a story, but are unaware of the full extent to which they do so because they are underestimating the degree to which the other sources from which they get their information and beliefs have all been doing a similar filtering. The result is social networks that don’t recognize that they have developed a groupthink that is not centered on the truth.
Peter Gordon marvels that
There are intelligent adults in the world who go to bed each night believing that the other side tells lies, but their side is above all that.