Sunday, April 15, 2012

Study 'o the day: Walmart and hate groups

Goerz, Rupasingha and Loveridge (2012) Social Science Quarterly (gated) document a correlation between the number of Walmart stores in a county in 1998 (!) and the number of hate groups in a county as of 2007. Their paper has gotten written up at Atlantic Cities (who offer some skepticism but probably should have passed on mentioning it at all) and a site called Live Science (linked to by the consumerist) in that familiar sort of we're saying its not causal but really we want you to think it is sort of way common to journalistic accounts of academic papers.

The article itself encourages this, as the authors use the term "determinants" rather than causes but the authors offer up some causal stories; my favorite involves the way Walmart's "Save Money, Live Better" causes people to think like racist protestants, or something like that. That bit is on pages 10-11 if you want to read it; the folks at Atlantic Cities didn't like this bit either.

The other fun bit is the puffing up of the Walmart variable as being the "most statistically significant". Indeed, it has the second largest t-statistic in the final column in their Table 2.  What matters, though, to interpreting the results is its substantive significance, not its statistical significance. 

My explanation: Walmart is better at predicting where low income whites live - they are (many of) the folks who shop at Walmart so Walmart has some incentive to be good at locating them - and they are also likely the ones who form groups that end up on the Southern Poverty Law Center list - than are the authors, whose remaining covariates do not dispel my concerns about omitted variable bias.

1 comment:

rkeefe57 said...

There are several problems inherent in the Goetz & Company study from the get-go:

1. There is no legal definition of "hate group," which is why even the FBI does not, cannot, designate hate groups.

What exactly was Goetz studying then?

2. The SPLC uses the deliberately meaningless term "hate group" in its fundraising materials because it allows them to denigrate any group without accusing them of any actual crimes.

Without a legal definition, a "hate group" is anything the SPLC, a private fundraising organization, says it is.

That's simply not good enough.

3. For reasons known only to the authors, Goetz, et al, compared the number of "hate groups" on the SPLC's spurious 2007 map with the number of Wal-Mart stores in the US in 1998!

What's up with that?

If the authors had compared the number of "hate groups" claimed by the SPLC for 1998 there would have been almost half as many groups, which would completely skew their findings.

So would comparing 2007 "hate group" numbers with 2007 Wal-Mart numbers because there were more than 1,000 more Wal-Marts around than in 1998, a 40% increase.

As long as you're picking and choosing your data sets to fit your hypothesis in advance, why not compare 2007 "hate groups" with 1907 blacksmith shops and correlate the connection between "hate groups" and horses?

But wait! There's more!!

4. In 2008, the year the 2007 "hate group" data was collected, it turns out that 147 of the 926 alleged groups were not affiliated with any known city or town.

These homeless "hate groups" simply float about in the SPLC's fundraising literature, padding the numbers.

That's over 13% of the total. Is it possible that a 13% discrepancy is statistically insignificant?

By 2010 the number of phantom groups had doubled to 26% and still stands at 1 in 4 today.

5. Last October, the SPLC's public relations chief, Mark Potok, the man in charge of cooking up the SPLC's "hate group" numbers each year, admitted on camera that his numbers were "anecdotal," "a very rough estimate," and "an imperfect process."

In the past, Potok has admitted in the press that his reports include "groups" that are nothing more than post office boxes and that his "Hate Map" fundraising tool "does not include original reporting by SPLC staff.”

THIS is "hard data"??

Maybe there is actually a correlation between "hate" and Wal-Marts or maybe not, but either way, picking and choosing the data sets that produce the desired results or relying on SPLC fundraising propaganda is a dismally poor way to go about proving it.