Tuesday, October 7, 2008

The Bradley Effect and Obama

Current McCain-Obama polls are increasingly widening in favor of Obama. An infrequently discussed topic concerning this polling is the contention that a potentially significant portion of those potential voters who say they’ll vote for Obama won’t. An important (unanswered) question is, “How large is this effect?”

A potential method for measuring this “Bradley effect” (also known as social acceptability bias) in the current Presidential polling of “undecided” voters would be to take a random subsample of those polled and administer the Marlowe-Crowne Social Desirability Scale (MCSDS). A statistical control measure for response bias, the MCSDS is a 33-item scale that describes highly desirable behaviors with a low probability of occurrence. For example, the MCSDS describes either desirable but uncommon behaviors (e.g., admitting mistakes) or undesirable but common behaviors (e.g., gossiping). The MCSDS may be reliable in measuring social desirability. Those in the subsample that engage in significant amounts of socially-desirable responding could then be discarded from statistical consideration; mid-range scorers on the scale of socially-desirable responding may or may not be included in statistical consideration at the pollster's discretion (or their answers could be recalibrated in proportion to their perceived degree of skew) in order to adjust the polls for the Bradley effect.

No comments: