Sam Wang's methodology is similar in many aspects to that of Nate Silver. Like Silver, Wang uses a meta-analysis approach in his predictions by combining data from all available state polls. Wang cites that it is important to aggregate different polls as a way of accounting for outliers present in different polls. While Nate Silver uses an averaging method to account for outliers that may be present in the data, Wang instead uses median based statistics. Wang claims that a the median value is often more representative than an average value when outliers are present. Also like Silver, Wang looks for day-to-day changes in the polls and identifies "shocks" to the campaign that cause drastic fluctuations in the polls such as democratic or republican conventions, and presidential debates.
It can be argued that Sam Wang has more faith in the polls than does Nate Silver. Wang claims that the polls do well enough on their own that there is no need for the econometric factors that are so abundant in Silver's methodology. Also unlike Silver, Wang does not throw out any polls before conducting his statistical analysis. It is interesting to note how accurate both Wang and Silver are in their election predictions given their slightly different methodologies. Perhaps the key to accurately predicting elections relies more on the quantity of polls used rather than on the variation of statistical analysis that is conducted.
http://www.forbes.com/sites/alexknapp/2012/11/07/election-forecaster-sam-wang-on-the-future-of-polling-and-punditry/
http://election.princeton.edu/faq/
The median can be used to describe the skewness, when presented with the mean, of a graphed set of numbers. Wang represents median statistical analysis and Silver performs mean statistical analysis of poll data. Perhaps an additional meta analysis of both Wang's and Silver's meta analyses could account for the skewness of the nation and thus produce a more accurate prediction. The only fear of such an analysis, is that it seems as though the statistics become further and further from reality. Does anyone know if it is possible to go too far with statistics, and actually produce values that have no real meaning? I mean I know how to, but do you think that an analysis on an analysis would yield such results?
ReplyDelete