In theory a random poll should find the statistical identification. But they aren’t truly random. Too many lifestyle issues interfering, IMHO. R’s either don’t or won’t answer because of cultural proclivities...
Couldn’t find the full link, but n’ertheless Gallup (and other polling institutions) may be truthful when they say they do not weigh their samples after party affiliation. However, if there is reason to believe that the party split is 35/35/30 then a result showing a 30/40/30 split should raise questionmarks? However, they never seem to question their onw polls that way.
Recently, as has been discussed on a number of threads, the weighting that is carried out, namely according to race and gender, has skewed the polls towards Obama (surprise, surprise) by over-weigthing for women and reducing the number of white voters. Of course this will show up also as a smaller number of GOP voters.
If you’re not weighting by party affiliation it doesn’t matter when or if you ask the question of party preference.
Now that I know Gallup doesn’t weight by party affiliation I understand their poll a lot better.
Bottom line is that I think the likely result is somewhere between how people self-identify and what actually occurred in 2008. Having said that, then Romney's lead is bounded on the lower end at 0 and on the high end at 7...a pretty good place to be two weeks out.
I actually think that is a valid methodology so long as the screen for likely voters is valid. What did he say about their likely voter screen methodology?
They know some groups are more likely to vote than be contacted by and answer a poll.
But weighting to “fix” this can just make it less accurate.
Some polls weight by party assuming the ratio will be same as 2008. Not likely.
i think it is a mistake not to try to weight by party affiliation. It’s a guess, but so is the screen for likely voter. It also makes sense to reweight by gender and race, if voting behavior varies a lot by gender and racial lines.
The exit polls don’t have any issue with identifying “likely voters”, gender and race, but still can get it wrong due to high rates of non-response by specific voter groups, usually republicans. Reweighting by party as well as the other factors can control partly for that.
If you get all the other demographics profile correct then you don’t really need to weigh party affiliation, because that is something that can change with every political season.
The key here is getting all the known demographics profile correct.
They do admit to weighting polls by race though, and they are undersampling whites by about 10 percent.
[ We only ask the person after we have polled them who they currently self-identify with at the end of the poll for informational purposes. ]
WEll DUUUUUGH... talk about a “TELL”... i.e. poker