Monday, June 30, 2008

A social science attack on that wicked voter ID

In my days as an employed academic, I used to follow the social science literature quite closely. I think I can say without fear of contradiction that I knew the current findings better than almost all of my academic colleagues -- and I have the published critiques to show for that.

And keeping up with the scientific literature was particularly onerous for a conservative. One knew that the summary and conclusions of any given article would always be "spun" as supporting a Leftist viewpoint. So one had to go to the "Results" section of the article and plough through a lot of statistics in order to find out what really happened in the research concerned. That did of course take a lot of time but was often very instructive. I have seen results that could not have been more destructive of a Leftist theory presented as if they supported the Leftist theory. I offer a small appendix below in which I give an example of that.

After about 20 years of that, however, I gave up. There was so little wheat among the chaff that I just ceased to take the whole body of social science literature seriously. What was reported was usually very poorly done (Leftists corrupt anything they touch) and anything that was openly supportive of a conservative view would almost never get published anyway. So one was reading bigotry rather than science.

So it is only now that an article published last January has come to my attention. And even now I cannot justify a long look at it but I thought that I might make a few comments. The article claims that asking for ID from voters is a BAD THING. I reproduce a summary of it below and I will then go on to point to some of its weaknesses.
A new Brown University study reports that U.S. states that require voters to present identification before casting ballots have lower levels of political participation. The research also indicates that voter I.D. policies discourage legal immigrants from becoming citizens, particularly for blacks and Hispanics, reducing odds of naturalization by more than 15 percent.

Since 2000, and stimulated by new security concerns after 9/11, there has been an upsurge in state requirements for voter identification. By 2004, a total of 19 states required some form of documentation of a voter's identity, sometimes in the form of photo I.D. Proponents of such requirements believe identification is a necessary tool to prevent voting fraud, such as voting by noncitizens or people who are otherwise ineligible to register. Others argue that whatever its intention, I.D. policies have the effect of suppressing electoral participation, particularly among minorities.

The report, co-authored by S4 Director John Logan and graduate student Jennifer Darrah, concludes that voter I.D. is one of many factors that negatively influence civic participation in the United States. The report states, "At a time when many public officials express regret that immigrants seem to lag in their participation in mainstream society, even small suppressive effects on naturalization - the formal step to becoming an American citizen - work in the wrong direction and should be taken into account as people evaluate the benefits and costs of more stringent identification requirements."

The new study extends previous research on I.D. requirements by analyzing not only voter turnout, but also voter registration and - "the key prior step for immigrants" - the decision to become a citizen, across racial and ethnic groups. Key findings include:

* in states with a voter I.D. policy in 2000, the odds of naturalization for foreign-born residents of the United States were reduced by more than 5 percent, with the strongest impact on Hispanics;

* in election years from 1996-2004, the odds of being a registered voter among citizens aged 18 and older were higher for whites by about 15 percent in states with voter I.D. requirements. But this effect was more than counterbalanced by a reduction in white voter turnout. In 2004 alone the net effect was to reduce white turnout in these states by about 400,000 votes;

* in this same period, voter I.D. policies reduced Asians' registration and diminished voter turnout by blacks and Hispanics, by about 14 percent and 20 percent respectively. The net reduction in minority voting in these states in 2004 was more than 400,000 votes;

* the suppressive effect of voter I.D. disproportionately affected not only minorities, but also persons with less than a high school education and less than $15,000 income, tenants, and recent movers. While persons with these characteristics are substantially less likely to participate in civic affairs regardless of their state of residence, they experience an additional significant reduction in participation relative to others in voter I.D. states.

"It is incredibly clear how voter I.D. requirements disproportionately affect and suppress minorities," said Logan, professor of sociology. "This data shows that if voter I.D. policies had not been in place in 2004, voter turnout would have increased by more than 1.6 million. That is a strong argument in itself for change."


Those "incredibly clear" results are not so clear if one looks at them with the skeptical eye that is proper in science, however. For a start, how did they equate States with and without voter ID laws? As a broad generalization, I would expect that it would be the more conservative States that have such laws. So are observed differences between the States caused by the greater conservatism of those States or are they caused just by the voter ID laws? It could be either one of those -- and any attribution of the interstate differences to the voter ID laws is nothing more than speculation.

There are of course statistical means (analysis of covariance etc.) for holding one influence steady while examining the effect of the other influence but that requires a good measure of both influences. And how does one quantify the degree to which a State is conservative? Does one use percentage voting for the GOP in the previous Presidential election? Maybe. But as many conservatives will tell you with some vehemence right at this moment, even a GOP Presidential candidate may not be very conservative so a vote for him could be a long way from an expression of conservatism. So statistical control founders on such objections.

In essence, then, the research above is essentially epidemiological -- and therefore heir to the big limitation of all such research, the limitation that correlation is not proof of causation.

And there are in the results themselves indications that the guesses about causation are poor. How do we explain that voter ID allegedly increased white voter registration but reduced white voter turnout? The two effects seem contradictory. Surely registration should INCREASE turnout and surely ID requirements should REDUCE voter registration? Yet the opposite happened in both cases. One can of course come up with ad hoc explanations for both effects but once again we are forced into speculation rather than having clear evidence of anything.

And one should finally note that a reduction in voter turnout is precisely what the voter ID laws aimed at. If you prevent ineligible people from voting, that must (ceteris paribus) lead to a reduction in the numbers who vote. So if the research above proves anything, it proves that voter ID laws had the intended effect. The fact that the reduction seems to have been particularly marked among Hispanics (many of whom suffer from a sad lack of "documents") supports that interpretation.


An article on racism by Gough & Bradley (1993) is an example of how a respected author in the field concerned can reverse the plain implication of his research results. The article started out well. Gough & Bradley were unusual in that they used a properly constructed multi-item scale to measure rated racist behavior. They correlated it with a form of the California "F" scale (usually described as measuring authoritarianism but perhaps more informatively referred to as measuring a type of old-fashioned thinking). They found a correlation between the attitude and behavior measures of essentially zero (.08). A clearer disconfirmation of their theory would be hard to imagine.

So did they say: "We were wrong"? Far from it. They then decomposed their attiutude and behaviour indices into the individual items making up those indices and looked for correlations in the large matrix of correlations between the individual items. And there were some non-negligible correlations there. But there would be by chance alone! If you take 5% probability as your criterion for significance (which is conventional) and you have 100 correlations, 5% of them will (ceteris paribus) be identified as significant! What Gough and his friend did was then exactly what you are warned against doing in Statistics 101. And on the basis of that fraudulent procedure they claimed to have produced evidence in support of their theory

Reference: Gough, H. & Bradley, P. (1993) Personal attributes of people described by others as intolerant. In P.M. Sniderman, P.E. Tetlock & E.G. Carmines (Eds.) Prejudice, politics and the American dilemma (pp. 60-85) Stanford, CA: Stanford University Press.

No comments: