The TT internet poll makes standard internet poll mistake

Karl-Thomas Musselman writes:

I've never seen 3rd Party candidates poll this high in Texas. Up to 15% of the vote in the Railroad Commissioner race is potentially tied up between the Libertarian and Green party.
There is zero chance that numbers like this happen on election day. Glass will not be at 8%, and no Libertarian will crack 10%, except where there is no Democrat on the ballot.

He posts these numbers from the YouGov poll:

Perry 50% White 40% Glass 08% Shafto 02%

Lt. Governor
Dewhurst 51% Chavez-Thompson 38% Libertarian 09% Green 02%

Attorney General
Abbott 55% Rodnofsky 35% Libertarian 11%

Combs 51% Libertarian 11% Green 09% Undecided 29%

Land Commissioner
Patterson 50% Uribe 37% Libertarian 12%

Ag Commissioner
Staples 50% Gilbert 37% Libertarian 12%

Railroad Commisioner
Porter 50% Weems 34% Libertarian 10% Green 05%

Those third (and fourth?) party numbers are just laughable, and they illustrate why I don't generally mention or comment on the internet polls. 15% of the vote in the Porter v Weems Railroad Commissioner battle goes Green or Libertarian? No way, no how. 12% of the vote going libertarian in the AG, Land Commish or Ag races? LOL, I don't think even most Libertarian activists are that optimistic.

It's a known problem with these internet polls that they tend to oversample certain types of voters, who are exactly the types who will vote for Libertarian or Green Party candidates. When you don't even attempt to control for that, it's just not a poll that you can take seriously.

Further, previously the TT has followed their internet polls with a series of articles breaking down their internet poll. The problem with that is that subsample results are even more statistically suspicious than the topline results. It really shouldn't be done.

Posted by Evan @ 10/26/10 05:04 AM


Previous Entry | Home | Next Entry


No comments yet

Add Comments

No flames or impolite behavior. HTML will be stripped. URLs will be transformed into hyperlinks.