Reply #65: The abstract is only 6 short paragraphs.

[eomer]

As one would expect, the abstract explains the thrust of the paper.

DU rules limit me to 4 paragraphs but 4 of the 6 are enough to give you the thrust:

E/M hypothesized that the discrepancy between their exit poll results and the reported vote was due to
different exit poll response rates by Kerry and Bush voters. However, US Count Votes' simulations
show that no plausible Kerry and Bush response rate distributions with constant mean can (with any
realistic chance) reproduce the distribution and values of the Edison/Mitofsky1 data for mean "within
precinct error" (WPE), median WPE, and overall response rates.²

US Count Votes has simulated a variety of exit poll response rate (Gaussian3) distributions for Bush and
Kerry voters and studied the resulting exit poll within precinct error distributions. The simulations thus
far suggest that possible ways to reproduce patterns of mean and median WPEs, and overall response
rates that resemble the distribution of the actual reported E-M exit poll data include:

voter exit poll response rate distributions with means that vary widely with the percentage of
Bush and Kerry votes cast in precincts and seem implausible.

very plausible exit poll response rate distributions by Kerry and Bush voters, accompanied by
vote shifts from Kerry to Bush.

I'll wait while you read the abstract ... .... ... ...

OK, now you know the thrust of the paper.

The thrust is that the USCV simulations were not able to reproduce the patterns of WPE means and medians and the overall response rate when they were based on constant mean response rates and no other assumed bias. The simulations were able to reproduce the patterns of WPE means and medians and the overall response rate with either of two sets of assumptions. One successful set of assumptions used response rates with means that varied according to partisanship with no other bias. The second successful set of assumptions used constant mean response rates combined with vote switching from Kerry to Bush.

The graph on page 10 shows the response rate means for the different levels of partisanship that were used to successfully simulate. The graphs on page 9 show the WPE means and medians that these response rates produced and show how similar to the actual means and medians they were.

That's the thrust. And, by the way, that is basically the same thing that Ron Baiman is quoted as saying in the OP of this thread.

As far as what you should comment on, I would think that USCV's conclusion that the simulations based on constant mean response rates do not reproduce the WPE means and medians and overall response rate would be a good place to start.