We have all heard of the Bradley Effect — the polls suggest that there will be more votes cast for Senator Obama than actually will be: Senator Obama is Black, and since some poll respondents do not wish to be considered racist they conceal their true intentions. Although difficult if not impossible to quantify rigorously, there is probably something to the theory.
I have another theory to suggest, the bottom line of which is that the spate of fraudulent voter registrations, many courtesy of ACORN, may have a similar effect and, quite possibly, an unintended consequence. This theory is predicated on assumptions for the first of which I have significant empirical evidence and for the second of which I have only modest empirical evidence. I must also acknowledge an abysmal lack of sophistication in statistical methodology.
Assumption 1. The various opinion polls are weighted to reflect, among other things, percentages of registered voters in the sample area. The sample used by Zogby, for example, is drawn as follows:
Samples are randomly drawn from telephone cd's of national listed sample. Zogby International surveys employ sampling strategies in which selection probabilities are proportional to population size within area codes and exchanges. Up to six calls are made to reach a sampled phone number.
The responses are then weighted as follows:
Weighting by [region, party, age, race, religion, gender] is used to adjust for non-response. The margin of error is +/- 2.9 percentage points. Margins of error are higher in sub-groups. (emphasis added)
Hence, if the respondent sample is deficient in a particular race, party or other sub-group, answers of respondents from the deficient subgroup are given greater weight. Such weighting is appropriate for any number of reasons, including the exclusion of cell phones from the national listed sample which, absent weighting, would cause over sampling of respondents with land line telephone service. Obviously, people with no telephone service, cellular or land line, are also excluded. I have no problem in general with the weighting process, because it makes sense.
Assumption 2. Although I have seen no data which affirm or reject the notion, I assume that most of the recent ACORN registrations have reflected affiliation either as Democratic Party or (possibly) Independent. This does not suggest any evil intent on the part of ACORN; it is merely a probable outcome of the areas and population groups in which ACORN makes its efforts — "low and moderate income neighborhoods and communities of color" where Democrats (or possibly Independents) are many and Republicans are few.
To the extent that voter registration drives have been successful, new voters have been registered; that is the purpose of registration drives. According to the most recent Zogby poll,
Among those voters who said they have registered to vote in the last six months, Obama leads McCain by a 53% to 37% margin. Among those who have already voted – about seven percent of the sample – Obama leads by a 52% to 42% edge over McCain.
The sixteen point spread among new registrants is much greater than the ten percent spread among those who have already voted. Both spreads are significantly greater than the 6.2 percentage point spread overall. In attempting to understand these numbers, it would be useful to recall the sampling and weighting methodologies, referenced above, used in calculating the numbers.
Assumption 3 Some, but not all, exemplars of the attempted fraudulent voter registration process have got past whatever screening process registrars of voters use. It would be unrealistic to assume otherwise, if for no other reason that the volume has overwhelmed many voting registration offices. Some have been detected, others most likely have slipped through. In some cases, it has been impossible to remove from the voting rolls those not entitled to vote, because they are felons which may not under a State law be entitled to vote, or otherwise. It may be appealing to speculate about how many fraudulent registrations have been detected, but it is quite difficult to determine now how many have slipped through. Few of the many dead people who have been registered are likely to vote, and neither would Mickey Mouse nor Ned A. Job likely attempt to vote, or be allowed to do so, even assuming that their registrations slipped through. Still, it seems probable that some indeterminate number of fraudulently registered voters will actually vote. That is not the point of this article. The resulting mess following 4 November may well be worth writing about later, but not now.
Thus, I think one may reasonably conclude that:
1. To the (unknown) extent that fraudulent voter registrations have inflated the apparent number of registered Democrats (or, for that matter, Independents or Republicans), the results have been reflected in the public opinion polls due to the sample selection and weighting processes outlined above. Based on recent accounts of ACORN's activities, it seems reasonable (to me; others may well disagree) to assume that there have been more fraudulent registrations as Democrats than as Republicans. If so, the polls are to that extent distorted in favor of a win by Senator Obama, thereby exacerbating the Bradley effect.
2. An indeterminate number of registered voters are a bit lazy. If it is raining, or the car has a flat tire, they have the flu, or for any number of reasons, they are less likely to vote than their more energized brethren. It seems reasonable to assume that if they feel that their votes are not necessary for their candidate to win, they are even less likely to vote. Many of them recognize what they perceive to be realities, and if their candidate is substantially ahead in the polls, just won't bother. This effect is, obviously, likely to be counterbalanced to an indeterminate extent by those whose candidates are substantially behind in the polls. They just won't bother either.
3. Opinion polls, due to their sampling and weighting methodologies, are flawed. It is unwise to put much faith in them. Regardless of what they suggest, it would be foolish to abstain from voting because of them. Nevertheless, some will do so.