Sunday, February 26, 2006

More reality TV voting

My brother emails an interesting question. Essentially, he wonders if all this reality TV voting affects individuals' propensity to vote in "real elections." This is an interesting question. I do not have strong priors about which way I expect the effects to go. The easiest, and perhaps most likely, answer is that they are totally separate activities, and, as such, no relationship exists between them. However, one could argue that viewers of these shows internalize the importance (or futility) of voting and are thus more (less) likely to vote in regular elections.

While I don't know if I should expect a positive, negative, or neutral relationship between reality TV viewing/voting and regular voting, I do think that it is possible to look for this relationship empirically. So let's practice the empirical methodology I discussed in class last week.

We want to examine the relationship between reality TV viewing (or voting) and voting in regular elections. What constitutes simple evidence?

A reasonable start consists of a simple comparison of the fraction of reality TV viewers (or voters) who vote in a regular election to the fraction of non-viewers who vote in the regular election.

What data are necessary to compute even this simple relationship?

We need individual level data which includes information about each individuals' viewing of reality TV and about their voting in elections.

Are there obvious problems with the simple comparison?

Clearly, the set of people who watch (and vote) for reality TV shows differs from the set of people who don't watch. As such, it is likely that there are some omitted variables which bias the simple comparison. E.g., reality show viewers are much younger than the overall population. Young people, independent of reality shows, vote less in regular elections. As such, the simple approach attributes some of the effect of age on voting to watching reality shows. This is clearly undesirable.

What is the ideal evidence?

At the very least, we would like to compare very similar people who differ only in their viewing of reality TV. That is, we would like to control for a variety of factors like age, race, gender, etc. Yet, even with a large set of controls, a selection problem still exists. What if the unobserved characteristic(s) which explains why otherwise similar people differ in their TV viewing also explain why they differ in their voting behavior? In this case, we falsely attribute differences in voting to reality TV viewing when they should be attributed to these unobserved characteristics.

As such, ideally, we would randomly assign people into a groups which do and do not watch the reality show and then compare the voting behaviors between these two groups.

Running an experiment seems like it would be pretty hard, is there something else we might try?

Yes, we can look for a natural experiment or instrumental variable -- something which randomly changes two otherwise similar people's propensity to watch a voting reality show, but does not change their propensity to vote (except through the change in TV viewing). I would suggest that, in the case of a show like American Idol, we can use Idol finalists' hometowns as instruments.

I would be shocked if Idol ratings do not shoot up in the hometowns of those who make the finals. The fact that an individual lives in the same area as someone who makes the finals (or semi-finals since that is before anyone votes) is random. Thus, I expect that living in the same place as someone who makes the finals randomly increases the propensity to watch American Idol. As such, we can, essentially, compare these people who were "randomly" exposed to Idol with similar people who were not exposed. If their propensities to vote in regular elections differ (all else equal), then these differences are the result of watching Idol.

Wow, that sounds very cool, can I do this for my term paper?

Yes, you can. However, there are a couple of remaining problems. To my knowledge, none of the data sets which ask people if they voted in, say, the last presidential election ask people if they watch American Idol. The National Election Study certainly doesn't. Maybe one of the PEW Center's datasets does (I believe they have questions about which news channel you watch, so maybe they have stuff like this as well). Whether or not one voted in the last election is a common survey question, the trick will be to find a dataset which has it and information about TV viewing. You will also need data on where the survey respondent lives, and you will need a significant number of respondents to be from the same areas as the Idol finalists. This could be an even more significant problem.

As such, I think a slightly different approach is more feasible. Gather panel data on voter turnout for communities (this is actually not that difficult) and then relate changes in voter turnout to having an Idol finalist in the period between elections. This is what as known as a differences in differences approach. You compare voter turnout in communities before and after they had an Idol finalist to a similar set of communities without Idol finalists. Any differences in the changes in voter turnout between the two sets of communities across the two periods is attributed to the presence of an Idol finalist (assuming a few things I won't go into here).

Let me know if you are interested in working on this.

Comments: Post a Comment

Subscribe to Post Comments [Atom]

<< Home

This page is powered by Blogger. Isn't yours?

Subscribe to Posts [Atom]