David M Rothschild on Posted on

Yesterday there were two polls released with quite the spread: Clinton +12 and Trump +1. I noted then that the key findings are actually not that different if you focus on Trump 41 and Trump 43; Trump is at a low ceiling, even on polls that favor him. But, let me take a little closer look at at something of interest: the pro-Trump IBD poll has 261 Democrats / 261 Republicans / 275 independent’s today and 254 Democrats / 254 Republicans / 259 independents yesterday (the aforementioned poll).

There is no public methodology for IBD poll, so this is my quick read at their cross-tabs.

Pro-Republican Bias in Raw Data: this is way too many Republicans relative Democrats as Democrats are somewhere between 5-10 pp more than Republicans in general population. But, this is also a lot of independents. If a poll pushes people to lean, between 80-85 percent of respondents will admit they are really aligned with a party. Without leaning, something between 65-70 is reasonable.

Weighting Does not Solve: IBD does weight the poll to 291 Democrats / 235 Republicans / 271 independents, which is perfectly reasonable without leaning.

The problem is that if the raw sample is oddly Republican, it is highly probable that the people within the sub-group are also oddly Republican. Probably even conditional on the other demographics IBD weight on, whatever they are. Thus, even after weighting, it is likely that their Democrats and independents are more Republican leaning than a random selection of Democrats and independents. Today’s poll, 41 to 41, has 80 percent of Democrats voting for Clinton (yesterday’s was 77 percent). That is very low, especially with 83 percent of Republicans voting for Trump.