Howard Dean almost made history in 2004. While John Kerry just plain fumbled the opportunity handed to him, Dean’s fund-raising and grassroots registration and turnout drives were spectacularly successful. Over 58 million Americans voted for Kerry-Edwards in the presidential election, an amazing success for the campaign’s workers and planners. President Bush, however, claimed over 62 million votes, on the strength of a similarly epic turnout effort. The 2004 election turned on many points of decision, but in effective terms both parties knew they had to maximize their turnout.
Fast forward to today. Rasmussen Reports (whom I distrust to some degree, because they do not reveal all of their methodology and weighting rationale to the media, bloggers, or academics) released a poll showing Barack Obama leading John McCain by 2 points, 48% to 46%. The race is statistically tied, says Rasmussen. Looking more closely, however, I noticed how Rasmussen has weighted the respondent pool, with 39.7% weighted as Democrats, 32.1% as Republicans, and 28.2% as unaffiliated.
The significance of this weighting is that, as I have warned before, it is not based on any established statistical record or census report. Instead, Rasmussen does what it calls “dynamic weighting”. What they do, by their own admission, is to establish “baseline targets” for party affiliation by reviewing “survey interviews with a sample of adults nationwide completed during the preceding three months”. When translated into English, Rasmussen is admitting that they use the results from previous polls to weight their new polls, even though this is – by definition – circular logic and is invalid for any poll which is authentically using pure RDD methodology in its respondent contact procedure (that is, a pure random method of contacting people means that you have near-absolute certainty that you are not calling anyone from prior polls, and therefore there is no connection between your present and former respondent pool except that they meet the defined criteria, and therefore conditions of the former do not influence the conditions of the present). Rasmussen has no effective basis for claiming the party affiliation in its polls, as a result of its invalid system. This happens a lot with polling groups, but it is a critical factor in the results they report, and as such I must warn the readers regularly about this practice. While no proven scientific method exists to track and report party affiliation between elections, the reader should at least expect a consistent ratio to be used, and for polling groups to obtain the basis of party weighting using outside and objective sources for their party identification data.
– continued –
]]>< 