Well, not quite in those words. But it could’ve been written that way.
Thanks to today’s Wall Street Journal I know that a guy by the name of Seth Stephens-Davidowitz has–essentially–correlated the share of votes for Obama in 2008 with the frequency of google searches involving the n-word.* He finds that the greater the frequency of n-word use in online searches, the smaller Obama’s vote share, to the extent that “racial animus in the United States cost Obama three to five percentage points in the national popular vote in the 2008 election.”
Stated this way, the finding doesn’t seem like all that big a deal. After all, who cares that much (aside from Barack and Michelle) whether the guy won by 53-47 or 55-44? But there’s another way to interpret these findings that really does seem like news. Here’s what I have in mind:
Economists–which Mr. Stephens-Davidowitz will be upon his graduation from Harvard–are supposedly taught early on that orderings only are meaningful in terms of more or less, and not in terms of absolutes. In the case at hand, the point can be made very simply by granting Mr. S-D his claim that the frequency of n-word google searches constitutes a good proxy variable for the extent or intensity (or both) of “racial animus.” If that is true, then it is almost certainly true that the very same measure also constitutes a good inverse proxy for the extent of “support for affirmative action”.** So what, you may well ask?
Well, that means that Mr. S-D’s study could be equally well summarized like this: “support for affirmative action in the United States gave Obama three to five percentage points in the national popular vote in the 2008 election.” Now, I can’t say for sure that a google-search-intensity measure for “affirmative action is good” or something similar would yield exactly the same estimates at Mr. S-D’s proxy does, but then it isn’t my study, so I’m just going with the assumption that this n-word-search proxy is measuring racial attitudes exactly as well in either direction.
Of course, there’s one very big difference between the two otherwise-identical interpretations of the results that I’ve proposed. To say that “racial animus” cost Obama votes is to say that it did not affect the outcome of the election–it simply means that it cost him a landslide. But to say that “support for affirmative action” gave Obama up to five percentage points is to say that it may have meant the difference between a comfortable victory and a squeaker–or even, perhaps, a narrow defeat.
Those double-edged swords are dangerous, Mr. Stephens-Davidowitz. Be careful when you swing one.
*More accurately, he’s correlated votes for Obama in 2008 relative to votes for Kerry in the same geographic area in 2004 with the aforesaid nasty word.
**Mr. S-D considers something similar to this point on pages 19 and 20 of his study, but dismisses them. He argues, for example, that because google searches for “civil rights” topics are “only moderately,” and not significantly, correlated with n-word searches that this means the n-word-search variable is not negatively correlated with support for affirmative action. Even stranger, having justified his variable on the grounds that it’s a more reliable proxy for true racial attitudes than responses to surveys that try to measure those attitudes (which have no relation with 2008 voting patterns), he then uses survey data to argue that his proxy is not an inverse measure of “pro-black” attitudes. Since I don’t find his reasoning in this section persuasive, I don’t feel moved to address it here other than to mention its existence.