The assertion that there are factors skewing the results both requires and is shown by the discounting of all plausible demographic factors. If we are to demand more transparency, we will need to comprehensively discount each possible demographic factor.
Is analyzing every possible demographic factor feasible? For that matter, is it even reasonable to think that it's possible to come up with an exhaustive list of the possible factors to analyze?
Here are a few that have come to mind recently. After someone posted a link to a Romney campaign guy bragging about the effectiveness of their GOTV efforts, I did a little searching for more details. Starting in 2007, the Romney campaign claims to have adopted radical new campaign tools based on data mining large data sets about demographics, voting records, etc., working with a company called Targetpoint Consulting. What if they're targeting precinct captains for the precincts where they predict they can get the biggest bang for the buck, i.e., big turnout? Or what if they identify voters to phone based on both geographic (precinct) data and demographic data. What if predicted precinct turnout is one of the variables that Targetpoint uses to skew the spending of campaign money? There are lots of ways that a correlation could be introduced, but because it's a campaign strategy secret it's hard to prove or disprove the degree to which this happens.
Other campaign strategies may introduce correlations as well. Suppose that rather than using data mining to target specific precincts, the Santorum and Paul campaigns have adopted more of a "make sure we reach every corner" approach. Precincts that Romney ignored could then be precincts that other campaigns targeted, tending to produce an inverse effect.
Don't forget that we know that ordering the precincts by turnout can result in geographical clustering, another factor that could introduce correlations. I only know that happened for sure in the Va Beach data, but has anyone looked at other precincts to see if there are geographical correlations with the alleged "post-divide" precincts? In Va Beach, for example, if Romney targeted that northeast neighborhood then because of the geographical clustering, he targeted 60% of the "post-divide" precincts. If he targeted a second neighborhood to the southwest of that he'd have covered an even higher percentage of those precincts, without necessarily even having used a precinct-oriented strategy at all. And those two areas might already be within Romney's demographic sweet spot, making them more likely to be pro-Romney anyway and more likely gotv targets selected by data mining.
(Looking at that map, I wonder if what they did was divide the map into several large areas, then divide each area to have roughly equal voter population within that area. Or something along those lines that tends to make precincts of similar size within a small region. If that's a common way to do it, then this kind of geographic correlation could be found elsewhere.)
Here's another one. Someone over on dp was arguing that they could prove the Super Brochure was effective by looking at the percent saturation within a precinct and the pecent vote for Ron Paul. I think they were looking at South Carolina there, but they said they had plenty of other data to show this happening in other areas. I'm not a big fan of the SB, so I haven't paid a lot of attention to it, but from the web site it looks like you "buy" a precinct, and the more "super voters" (not sure what that is) there are the more it costs. So I'm guessing that the degree of SB saturation correlates pretty well with precinct size. If that's true, then here's another variable that introduces a correlation with precinct size, one favoring Ron Paul (if you believe the SB is effective) in the smaller precincts that have higher saturation.
I'm *not* arguing that SB saturation explains the graphs. If I'm right about SB saturation being inversely correlated with precinct size, then I think the known correlation between precinct size and %vote for Romney makes this a weak argument for the effectiveness of the Super Brochure, but maybe an SB fan would argue that not buying enough SB mailings to blanket the largest precincts is the reason that Romney tended to take those precincts away from Ron Paul. Either way it's another example of linkage between precinct size and campaign focus.
And ruling out demographic effects one at a time isn't sufficient. It could be the combination of Romney using data mining to get bang for the buck from his mailing and phone and gotv efforts, plus some degree of geographic clustering of precincts with similar sizes, plus Paul and/or Santorum doing the opposite and trying to have some presence in every little corner of a state; and all of that is before we start to consider how geographic clustering plus demographic factors could skew results. These are, by the way, kinds of effects that could explain Iowa, with its public manual counting reported in real-time, just as easily as a machine-counted state.
And then of course are the factors that nobody has thought of yet that could introduce correlation. Those are even harder to rule out.
