5

It doesn't matter whether the estimates were made in terms of percentages or proportions

It also doesn't matter whether the estimates resolved by a binary truth (e.g. "What's the probability David Cameron will still be UK Prime Minister on Jun 1 2014?") or whether the truth can range anywhere from 0% to 100% (e.g. "What percent of votes will Barack Obama get in the 2012 Presidential Election"?).

I also need to be able to find out the ground truth that resolves the estimation - e.g. I can go and look up what percent of votes Obama got.

There's no need for the questions to be political; they could be about anything.

user1205901 - Слава Україні
  • 335
  • 1
  • 6

1 Answers1

3

the health and retirement study asks what's the probability you think you'll be in a nursing home in the next few years.

http://www.asdfree.com/search/label/health%20and%20retirement%20study%20%28hrs%29

you would also be able to calculate how many of these probability guesses came true. hrs is a longitudinal survey, so they follow americans aged 50+ until they die - they also follow people if they move into a nursing home. you could, for example, you could use the "Probability to live 75+" question at the point in the panel when each respondent turns 65 to calculate what share made it vs what share said they would. the link i provided has usage examples. look at the RAND codebooks for "reported probability" and be careful about how you're using the survey weights.

Anthony Damico
  • 1,480
  • 10
  • 16
  • I realise I'd neglected to mention a point in the OP, which is that I need to know the ground truth that resolves the estimation - that is, I need to know whether the event ended up happening or not. Though it wasn't entirely clear from the survey design document at that link, I guess I wouldn't get to find out whether the responders actually do end up in nursing homes or not. – user1205901 - Слава Україні Aug 16 '15 at 08:27
  • 1
    hrs can do that. see edit. good luck! – Anthony Damico Aug 16 '15 at 11:58