Friday, October 25, 2013

Lessons in Statistics: The New MSU-Billings Poll

I saw on my Twitter feed that MSU-Billings has come out with it's latest Montana public opinion poll. You can read the article about the poll findings here and go look at the full report here. I take some issue with how this poll was reported and how some folks are talking about it in the Twitter-verse.

First, let's talk gay marriage. The poll results indicated that a plurality of Montanans favor gay marriage, with 46.6 in favor and 42 percent opposed. The problem with this statement is the margin of error in the poll is 5 percentage points. This means a majority of Montanans might actually support gay marriage (5+46.6 is 51.6 percent) OR less than a plurality support it (46.6-5 = 41.2). So, is not accurate to say that a plurality of Montanans support gay marriage. It would be better to say that opinion on gay marriage is mixed and that support for gay marriage is statistically tied with opposition to it.

Second, let's talk about job approval ratings. Montana Cowgirl, one of the Montana state politics blogs I follow, sent a Tweet saying that Steve Daines is the most unpopular politician in the state. Here are the job approval numbers in the poll:

Steve Bullock, Approval: 53 percent.
Jon Tester, Approval: 44.7 percent
Max Baucus, Approval: 44.1 percent
Steve Daines, Approval: 39 percent

Remember that margin of error? We can confidently say that Steve Bullock is more popular than Steve Daines because the Bullock's approval rating could be as low as 49 percent or as high as 58 percent. Bullock might be as popular as Tester and Baucus, and Steve Daines might be as popular as Tester and Baucus. Or Steve Daines might really be the most unpopular elected official (39-5 = 34). But we cannot confidently, based upon these results, say that Steve Daines is the most unpopular elected official with the large margin of error.

What is most interesting, however, the number of folks who are undecided about Daines and Bullock relative to the senators:

Baucus, undecided: 16.9
Tester, undecided: 15.9
Daines, undecided: 37
Bullock, undecided: 33

What this tells me, and I can stay this with statistical confidence, is that Steve Bullock and Steve Daines are the least KNOWN statewide elected officials. And, look at the disapproval numbers:

Bullock, dispprove: 14
Daines, disapprove: 24
Tester, disapprove: 39.4
Baucus, disapprove: 39

Looked at this way, I could say with confidence that more Montanans who can express an opinion on on Tester and Max Baucus neither clearly approve or disapprove of the job they are doing. I can also say, however, that among those who express an opinion on Daines' job that more approve of the job he's doing than disapprove (39-5 = 34, and 24+5=29). He and Bullock are the only ones according to these numbers who have a statistically discernible pro-job approval rating outside the margin of error.

Clearly, among those Montanans who express an opinion on Steve Bullock really like the job he's doing. I suspect that part of that has to do with a) Bullock's not associated with the mess and Washington and b) the legislature has gone home, so he's not associated with any divisive doings in Helena at the moment. Main lesson: It's good to be the Governor when Washington is blowing up.

All three of our federal officeholders likely have depressed job approval numbers because of the government shutdown and the debt crisis. What will be interesting to watch is whether Daines, in particular, can increase his approval ratings among those who are currently undecided about him.

3 comments:

David Parker said...

James Connor, who writes the great Flathead Memo Blog (http://www.flatheadmemo.com/) sent me this in an e-mail:

I generally agree with your BSP post, but not on the meaning of a statistical tie. See:

http://www.motherjones.com/kevin-drum/2012/10/one-last-encore-great-statistical-tie-fallacy

I'll respond with some other thoughts in a post soon, but James has a good point.

Anonymous said...

I've tried contacting both Scott Rickard and Craig Wilson, the two professors who headed up this study, to get actual demographic numbers, since so far they haven't included them. The other issue with polls is the sampling. If the sampling was skewed in any direction (age/gender/income/party) then that would also bias the polling. They should be releasing this stuff, as its standard with polls so it can be looked at for reliability and duplication by statisticians and other researchers.

David Parker said...

These numbers are essential to understanding these findings. Their methodology appears sound, but we should have the demographic data so see if there's any skew you describe.