As I may or may not have said previously, I'm a huge fan of FiveThrityEight.com, and recommend it highly to the uninitiated, including and especially his most recent post on the integrity of Rasmussen Reports (or suspected lack thereof). He cites pro- and con-arguments, although (IMHO) the former are the stronger ones.
He misses one interesting point, however, specifically on the question of how Rasmussen models its samples for its surveys. It uses what it calls "likely voters," which seems reasonable enough until you realize also that it considers conservative or Republican voters to outnumber liberal or Democratic voters in this category. Of course, depending on the circumstances in any given year, the pool of "likely voters" may include many left-leaning and/or "swing" voters. But, by gradually adjusting its samples to include more and more of these latter voters as Election Day gets closer, Rasmussen is able to play a very two-faced game in its polling: use the early poll results in an attempt to "push" a conservative outcome, and then make adjustments in the samples of its later polls to gradually include more truly "likely voters." Ultimately, this allows them to produce accurate "final polls" that it can then cite as proof of its professional brilliance, while not denying themselves of an opportunity in the early stages to actively shape the outcome.
Far fetched? I really don't think so. Who knows what an enterprising liberal hacker would find if he or she could get into Rasmussen's e-mail? Maybe Pollgate would eclipse Climategate in importance among "likely voters." I'm not rooting for someone to break the law, but I wouldn't be surprised if someone did so and discovered I was right.
No comments:
Post a Comment