Media Bias Research and the Endogeniety Problem

Color me your color, baby
Color me your car
Color me your color, darling
I know who you are
--Blondie

The endogeniety problem is often present in studies seeking to establish relationships between media outlet exposure and erroneous views on issues.

The hypothesis goes like this. I suspect that outlet X misinforms its viewers. So if I survey viewers of various media outlets about their knowledge of a particular 'fact,' and viewers of X are more 'incorrect' on the fact than the viewers of other outlets, then I can conclude that X is misinforming its viewers in a politically slanted way.

Before we get to the endogeniety problem here, studies like this one have other obvious issues. Take, for example, the 'facts' that are employed in the study. One must come up with 'facts' that are not contestable, which is harder than you might think. Just because something is claimed as true by a group of people, e.g., George Bush lied about WMD in Iraq, does not make it a fact. Moreover, one would think that samples of media stories surrounding such facts require similar exposure across media outlets, lest the issue could plausibly relate to programming choices by outlet managers rather than distortion of facts actually presented.

Even if the researcher was able to compensate for the 'fact factors,' the endogeniety problem remains. People who view X may do so because they prefer the X's content and how it is delivered. Viwers may be attracted to X because they have particular beliefs and views of the world, and X's output may confirm those beliefs (this is the confirmation bias mechanism at work). These beliefs and views may cause large numbers of viewers of X to interpret information differently than those with other beliefs and views.

Because viewers self-select X, it will be difficult to discern whether subjects our hypothetical study significantly differ in their grasp of 'facts' because of the X treatment, or because of some other factor endogenous to the subjects that influences their intrepretive mechanisms.

If/when such studies are published, they should raise more questions than answers.