Thursday, May 04, 2006

The effect of internet on beliefs like tolerance


Ed Glaeser’s 2002 “The Political Economy of Hatred” introduces a model of hatred that is capable of explaining how an equilibrium can arise with a particular level of hatred given certain supply and demand. A large portion of his explanation deals with information; therefore, one might suspect the internet, one of today’s largest sources of information, to play some sort of role in the market for attitudes/beliefs towards tolerance for others (specifically we will look at false/negative stereotypes/prejudice).

Specifically, the internet plays a role in determining one’s level of tolerance for other beliefs in two ways. For people who hold beliefs of intolerance, two key forces play a role. The first is the cost of separation. If a particular group that holds an intolerant belief is ostracized by the rest of the community, this is a negative force to their utility. The internet lowers the cost of communicating with people of your same belief group: you can join online communities and groups with others of your belief pretty much anonymously, and this will strengthen your confidence in your current view. The second force involves exposure to contradictory information: one of the reasons that people hold intolerant beliefs is because they are not exposed to the factual information about the outgroup, whether it be because of limited access or because of choice. Once again, because the cost of communication is lowered on the internet (ie anyone can post their opinion/view), there is an increased probability that people will encounter information that contradicts with their current view, forcing them to reevaluate their belief. These two forces act in opposite directions, and it may be hard to say which one outweighs the other.

However, I would predict that the first force would outweigh the second. Although the lower cost of information transmission (ie for the outgroup to contact the group with the prejudice belief) plays a role, it seems that the ability for a person to almost totally (except for these random interactions with the outside group) segregate into a community and set of information that confirmed his belief without any cost of social pressure would make his beliefs become more polarized. For example, in the real world, expressing extremist opinions might lead to ostracizing, but on the internet, the expression of these beliefs would lack such consequences.

Data/Empirical Strategy

The ideal experiment would be to randomly provide people with internet access and see how their beliefs changed. This, of course is not possible. The simple regression would be to regress beliefs on internet use, but this has large OVB problems: determinants of internet use could very likely be determinants of the beliefs we are looking at. Even after controlling for demographics, there could very well be some lurking variable that explains why people who use the internet also have certain beliefs. Thus, we turn to instrumental variables. Although internet access is not itself randomly assigned, there is a random element to the cost of internet access supply. In other words, access to broadband internet (which empirically is a strong determinant of internet use; people with broadband use the internet more frequently than people with dial-up) is determined in a large part by a factor called “last-mile” cost. This last-mile cost is essentially the marginal cost of extending broadband access to an area from a backbone. The location of these backbones is largely geographically determined, and there is reason to believe that the determinants of backbone locations are at least more random that the determinants of actual broadband access in an area. So, in terms of instrument erogeneity, last mile cost seems to be a decent instrument in that it may be fairly uncorrelated with the error term here.

We would thus use last mile cost as an instrument to regress beliefs of tolerance. Specifically, I would link a dataset which included both internet use, beliefs, demographics, and location to a dataset which included the last-mile-cost for internet access in that area. If this information was not available, broadband access availability (provided by the FCC) would be my instrument.


There are several complications with the above strategy. First of all, we must assume that people are giving honest answers to questions about their beliefs. Practically, we must accept this. Secondly, since the internet is always changing, external validity would play an important role: the study would be done using data from early 2000, where broadband access was more variant than it is now, and the content/use of the internet has certainly changed since then. Lastly, the validity of my instruments is questionable (is last mile cost really uncorrelated from the error? Last mile cost is fairly related to rural versus urban, and this seems related) and is an issue that must be explored further.

The other larger issue is that there may be two types of people in general with characteristics that are endogenously determined-people who want to self segregate, and people who seek the “correct” information. In this case, the effect of internet use may be capturing the net result of the mix of these two people, and it is difficult to control for this.


You have a good start here. I particularly like the relation to Glaeser's work. You should probably also look at Mobius and Rosenblat's "Getting Closer or Drifting Apart" paper (it is probably available on Prof. Mobius's website) for more ideas about how the internet affects interactions and perhaps beliefs.

As we discussed before, I think the IV is ok. The trick is finding the data we need. How is that going?

At the very least, did you chack out the social capital benchmarks study to see what info on beliefs it has?

You are on the right track, but you need to start getting the data assembled quickly, so you can start getting preliminary results and figure out what will need to be modified.

I think your idea is great. However, I have qualms with your instrument. While I agree that the instrument is probably relevant as people with broadband probably do use the internet more, I think that instrument exogeneity is difficult to ascertain. In your proposal you mention that the "last mile cost" is largely geographical. While geography might seem random, I would argue that where you live (rural or urban) makes a huge difference in your beliefs. Therefore individuals who face high "last mile costs" are not necassarily random and the causation towards their beliefs comes not from their ability to access the internet but instead from their geographic preference. Additionally, individiduals who do decide to pay this high "last mile cost" and gain internet access could skew your results. These individuals are likely to want to have internet and have high internet usage but also be richer. Therefore their beliefs could be caused by their income and not by their access to the internet.

I think that the instrument you propose is one of the best that can be thought of and performed feasibly in this experiment. Still, I just wanted to make you aware of some of the possible exogeneity issues that you might face.
Varun is correct, although this problem is fairly easily dealt with by controlling for individual and community characteristics. Even better, although I don't think you'll be able to do it, would be to do this with community or county fixed effects. That is, you would exploit variation in hook-up costs with smaller regions where omitted variables related to location choice are less of an issue.
Post a Comment

Subscribe to Post Comments [Atom]

<< Home

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

Subscribe to Posts [Atom]