Justin Wolfers and I are presently writing a paper on, inter alia, the forecast errors of various pollsters in predicting the two-party preferred vote share. Alas, two of the major pollsters don’t publish most of their old two-party preferred results. Does anyone have two-party preferred voteshare numbers for Newspoll and AGB McNair (now AC Nielsen) for the 1987-1998 Australian federal elections? Just to be a pain, I’m after all polls over those election cycles.
Appropriately adulatory footnotes will be written in your honour…..
Andrew
Mumble says that Newspoll only asks about preferences during campaigns:
http://www.mumble.com.au/federal/newspoll.htm
That’s one way of avoiding academics saying that your two-party preferred is wrong!
Hasn’t Jack (is it Wolfman?) at Stanford done this already??
Other than that, all I think of is ANU’s excellent library.
Andrew, I’d be moderately happy with data from the campaign only, but I’m pretty sure a lot of 2PP numbers are out there, judging from some tantilising snippets I’ve run across.
Sinclair, you may be thinking of the graph in Wolfers & Leigh (2002), where we calculated the absolute standard errors based on primary voteshare. But I think now that the allocation of preferences may be what sets a good pollster apart from a bad one.
From the Mumble site, it seems they allocated 2PP from last election preference shares. So while 2PP shares may have been published, there is no guarantee that they come from data collected at the time of the poll.
That’s ok — we’re looking at the predictive power of pollsters. So we want the ones with bad methodologies (using previous election to allocate preferences) as well as good ones (asking respondents for their second preference).