I am looking for exti poll data broken down by state and demographic sub groups for the 2016 election.
It’s a bit strange to ask this on IPUMS data site. Election polling isn’t conducted by the official statistics providers. This is mostly a media exercise, see http://www.edisonresearch.com/electio… – and contact them to see if you can buy the data from them (I would not be surprised if they ask for something ridiculous like $50K, as this is the data set that is not available from anybody else).
Having said that, I think that the methodology of data collection is fairly weak, even though arguably this could be the best you could do with an instantaneous population of voters that showed up for the election. It’s a clustered design with relatively few PSUs; subpopulation sizes (e.g., distribution of Jewish votes, or distribution of votes cast by the residents of Florida) would have may be 5 effective degrees of freedom. There are virtually no nonresponse and calibration adjustments that can be applied to these data given the nature of the population. I would not view this as a definitive picture of the electorate; this is more like a somewhat out-of-focus, somewhat overexposed photograph of a crowd.
A better methodology is available with American National Election Survey, which is a face-to-face survey with reasonably high response rates, but much smaller sample sizes (about 2,000). But the microdata for 2016 may not be available until maybe a year after the election.