JRP1994
Sr. Member
Posts: 2,036
|
|
« on: July 14, 2016, 06:43:42 PM » |
|
Some responses, including mine, aren't publishing on the first page of this thread. I'll try this again for the third time. If the responses on the prior page start showing up, please forgive me:
These Morning Consult state "polls" aren't polls in the traditional sense. As I understand it, Morning Consult took subsamples of their national January-June nationwide poll responses and used multiple regression to reweight them for state-specific demographic factors, like age and income, and economic data, like unemployment rate, to come up with a figure for the Trump, Clinton and Other percentage in each of the 50 states. This MRP method is allegedly more accurate than actual state polls according to one study they cite - but is not a traditional poll. Perhaps this is the wave of the future, but again, it's not a traditional poll.
Back in April, they not only did MRP regression, but also took subsamples of respondents from January to April in each state, weighed them for demographic factors, and put out statewide special-sauce poll numbers for those states. The results of these special-sauce state polls differed from the MRP results in some states. It's not clear that Morning Consult did that this time.
Given that, I don't think these "polls" should be entered into the database - but that decision is not made by me.
It can get you a reasonably accurate national horserace number but I agree that these polls shouldn't be in the database because they're not actually polls, but mathematical projections based on subsamples with high margins of error
Agreed.
|