2008 jewish exit polls are false (Jews aren't that libreal)

[ag]

Oh the passion of prophet scorned!

Instead of screaming that this or that data is "fake" or whatever, one could make a fairly obvious point that it is dangerous to use data collected for one purpose to make other claims, especially about small subpopulations. The national exit polls are NOT designed to figure out how the Jews (or, for that matter, the Pastafarians) voted: whatever information of that sort is obtained is incidental. If one were trying to do Jewish exit polls, naturally, one would be weighting the sample to match the Jewish population distribution. The village of Kiryat Joel is negligible, if the goal is to figure out whether Obama or McCain will be president, but it is quite a bit more notable when viewed as a Jewish population center. As the data wasn't collected to figure out how the Jews voted, naturally it will be underepresented in the national subsample that happens to be Jewish.

There is nothing special about the Jews here - it would be the same if we tried to figure out how "the rich and famous 1%" voted: I strongly suspect that the national sample doesn't get many people from the Hamptons or the relevant Fifth Avenue blocks.  There are statistical techniques that could be used to try to remedy that, but I doubt they've been used in this case - at least nothingis mentioned about it. In any case, given that it is extremely unlikely that more than a couple Kiryat Joel residents were polled, that wouldn't be easy to do.

To sum up, the point that NY Jew makes is, actually, sensible. It may well be extremely misleading to make claims based simply on the proportion of the Jews in the national exit poll that claimed to vote D or R. There are simply too few Jews and they are simply too unevenly distributed and heterogenous.  One would definitely need to try to reweight the sample, at the very least - and it is not at all obvious that this has been done, nor that enough data has been collected from the ultra-Orthodox neighborhoods for this to be doable. At the very least, one would need to know more about the methodology here involved. While I would not go as far as saying the claims of that paper do not match reality, I would definitely support taking them with a big grain of salt until I see more data.

All that is reasonable enough. But NY Jew presents his - actually, quite reasonable - theory in such a way that one is inclined to dismiss it without even considering.

[ag]

my basic point is since most Jews who live in Jewish majority area vote very conservative and most jews who live in non Jewish majority areas vote liberal.  since random sampling is random and majority of all areas with jews are a non jewish majority random sampling of the country at large is worthless.  I don't think it's possible to make a accurate poll of the jewish vote that doesn't have a decent sample size from NY (enough to publish by the networks).

A perfectly valid - or, at least, plausible (to the extent that we don't have enough data to either confirm or refute it) - point. If only you made it without screaming, I would have bothered to read what you are saying a lot earlier.

[ag]

What?  That's not how random sampling works

Random sampling of what? This polls wasn't designed to give a good estimate of the Jewish vote - it's prime objective is to figure out who won. A "Jewish" poll would have to sample New York a lot heavier than Texas. A national poll would need to oversample NY to give reliable data on NY Jewry - and the fact that a separate number on NY is not reported suggests this is not done (sensibly enough). NY ultra-Orthodox are pretty negligible as far as the national exit poll is concerned, but much more important to figure out the Jewish vote.  Given how concentrated the ultra-Orthodox vote is, it is unlikely to have been sampled much. Perhaps some statistical adjustment has been made to take account of this: but it is not obvious. And, in any case, it is very unlikely that substantial enough numbers from those insular communities have been sampled in the first place to make such adjustment doable. Not at all a problem for the national exit poll. But a big problem if you are after figuring out how "the Jews" voted.

[ag]

And now let me make an insidious point

Who cares how the "average" Jews vote? It's the average body temperature in the hospital (including the morgue). Jews, at this point, are not a unified community. There are, properly speaking, the assimilationist American Jews  - "American citizens of the Mosaic faith" to use an old turn of words ). These are, of course, extremely liberal - probably, even more liberal than the polls suggest. And there are the ultra-Orthodox - the traditionalist isolated community, that continues its autonomous existence in the same way it had existed for millenia in Europe. These, of course, vote differently. If polls asked people to identify, what sort of Jews they are, most likely there would not have been a sufficient sample to make any claims about the latter, but we may reasonably conjecture they didn't vote in such large numbers for Obama ) These are two very distinct communities, with very little in common between them - why average them out?

[ag]

Yes, the argument is reasonable. Think of it that way. We do know that ultra-Orthodox live in some compact isolated communities in NY state. These ultra-Orthodox form a substantial part of the Jewish population not merely of the state, but of the entire US. Any poll that would have been designed to find out how the "average Jew" voted would take care of actually sampling from these communities.

To the best of our knowledge, no such poll has ever been conducted. Rather, national exit polls were used. As far as the behavior of an "average American" these isolated communities are not particularly important and could well have been missed. Neither, it seems, there has been an attempt to match the weights so that the ultra-Orthodox are properly represented in the sample (in fact, the data on the number of the ultra-Orthodox is not reported and, likely, wasn't even collected). Ultra-Orthodox are entirely inconsequential for the purposes of the national exit poll. But they ARE important for the answer to the question at hand. This MAY be a problem - finding out whether it is, would require looking at the data closely.

This is, I  believe, a well-known statistical problem of dealing w/ very small subsamples. It rears its head, for instance, whenever they try to estimate things about the ultra-rich from the general census surveys - there are just not enough ultra-rich in those samples, and the ones there are may very well be not very representative.

[ag]

But that argues that New York's Margin of Error is large, not that New York is somehow "underestimated"

Yeah, the point is that NY numbers may be wildly off - and that's the reason they are not reported separately. But the problem is, a good poll of Jewish voters would, probably, not have that problem

I don't know what Gallup does about Jewish polling. Have they actually polled Jews specifically?

Anyway, my point wasn't that NY Jew was right - it was that he was raising a valid issue. His choice of language was very unfortunate: so unfortunate, in fact, that I only started to read his posts in this thread to be able to claim he is talking nonsense. He isn't ))