I think this would make a better essay topic. Tokenadult is right, that the revealed preferences by themselves are not really news. And yet to read any-more int

I think this would make a better essay topic. Tokenadult is right, that the revealed preferences by themselves are not really news. And yet to read any-more into the data is unsound.[1] To the extent there is a benefit from your insight, it should be flushed out in detail (ie, to convey & understand the micro-analytics). You may (and probably, do) have this all in your head already. But it does not come out either in the posted article or in your comment. Relevant > http://en.wikipedia.org/wiki/Multicollinearity* http://en.wikipedia.org/wiki/Multicollinearity* ___________ [1] A more detailed explantion is as follows: Notwithstanding the substance (ie, the politics), we see an argument of the form: Binary (0,1)=F(X1...Xn) | (challenger, incumbent) ie Vote = Function of (...) So this equation is approximated: (0,1) = f X1(party affiliation)+X2(vote in last election)...XN-1(Growth)...Xn(whatever else). At first blush, its not clear ex-ante why this wouldn't be dominiated by the first two explanatory variables. At second blush, it seems obvious that any beliefs further out in the vector will be likely correlated with the general set of beliefs captured in the first two revealed preferences (ie, X1, X2). Not only are thes Xn variables (eg, growth) a series of beliefs* (not actual observed data), those beliefs will be (predictably) co-lineated with other variable that are (exAnte) more likely to have greater predictive power. So, you can't safely say Xn-1 is either independent or by itself stonger than X1 or X2. [#,##] And if that's the case, there really is not much to this. A poll of this type is the wrong form of argument to support your conclusion (as it is not strong enough). There are other levels of logical and orthogonal attack on this general argument (will spare the ink), but in any event these too would be better avoided/adressed in a written essay on the subject. You may also have other levels of logical and orthogonal attack that support your general approach, which would be interesting to read. _______________________________ [#] For the technical statistcal reference: http://en.wikipedia.org/wiki/Multicollinearity http://en.wikipedia.org/wiki/Multicollinearity <Consequences of multicollinearity> As mentioned above, one consequence of a high degree of multicollinearity is that, even if the matrix XTX is invertible, a computer algorithm may be unsuccessful in obtaining an approximate inverse, and if it does obtain one it may be numerically inaccurate. But even in the presence of an accurate XTX matrix, the following consequences arise: In the presence of multicollinearity, the estimate of one variable's impact on y while controlling for the others tends to be less precise than if predictors were uncorrelated with one another. The usual interpretation of a regression coefficient is that it provides an estimate of the effect of a one unit change in an independent variable, , holding the other variables constant. If is highly correlated with another independent variable, , in the given data set, then we only have observations for which and have a particular relationship (either positive or negative). We don't have observations for which changes independently of , so we have an imprecise estimate of the effect of independent changes in. [##] With respect to the facts/data, the sub-population reported will predictavly over-represent in near proportion to reported data merely on pre-existing party affiliation (or: revealed preference). And historically, this sub-population will have already historically over-represented for the incumbent. So ex-ante, we expect nested logic and colinearity in Xbar. So then comes the element of differentiation: 1) In general english, if you support a candidate you most likely support his position on Xn. 2) In general english, if you support a party you most likely support the Party Candidate. 3) In general english, if you support a party you most likely support the party position on Xn. 4) In general english, if a party supports a candidate, it most likely support the candidates position on Xn. 5) In general english, if you previously supported a candidate you most likely support his position on Xn. Etc. While these may or may not be true, the issue is one needs a deeper analysis. The problem is one of differentiation and granularity from the data. If the sub-sample was neutral in ex-ante preference, this would help, but we know its not (admittedly). So its all muddled.