Multivariate analysis of Wisconsin polling data

(This is cross-posted at Daily Kos)

A couple weeks ago, Kos/PPP polled all the Wisconsin Republicans up for recall and found some very interesting results. However, he did not poll the Dem races up for recall, as well as the statewide upcoming Supreme Court race. In an attempt to rectify this fault, although I’m no Poblano, I decided to try to use multivariate regression to try and model the Wisconsin polling data using information from each district.

Fortunately, the dynamics of the race were much simpler than Clinton vs. Obama 2008, and I found in the end that the polling results could be described by only two variables (which is very nice, as we only had eight data points, so the model shouldn’t be overfitted):

1. Obama: The percentage of the vote Obama received in 2008 (Courtesy of SSP)

2. Incumbency: The number of years the person has been in office (for instance, someone elected in 2008 would have an Incumbency of 2 years.) – Numbers from SSP above.

I also experimented with other variables (which I discarded in the end as being not statistically significant):

3. Barrett: The percentage of the vote Tom Barrett received in 2010. (Thanks to the Journal-Sentinel)

4. Scandal: A 1/0 value describing the unique circumstances of the aptly named Randy Hopper, and perhaps Mr. Prosser as well.

5. Kerry: Percentage of the vote Kerry received in 2004

I also decided on using percentages rather than margins as there was a better correlation between the two.

In the end, my 2-variable model describes very accurately (within +/- 1.5%) the percentage of voters who would commit to voting for a Democrat in a hypothetical election this year; the spreadsheet is included below.

(The main prediction is highlighted in red. There are other columns to the right which include the additional variables that did not turn out to be significant.)

In short, the vulnerability of each Senator is based mostly off of Obama’s performance in the state in 2008, along with a small bonus from incumbency (about 0.3 points per year in office.) Thus, Hopper is quite vulnerable simply from being a freshman (the scandal had not impacted his poll numbers at that point yet), while Alberta Darling has built up goodwill from being in office for 18 years.

Extrapolating this model for the three Democrats who are considered semi-vulnerable, we find they are mostly safe. The only one who’s really vulnerable is Mr. Holperin, who was first elected by 2.5% in 2008 and represents a seat Obama won by single digits. Note that I give the Democrats negative incumbency so it gives a bonus to the D #s (rather than a penalty), and since the model considers undecideds, anything 48% or up is probably leaning D.

Examining Justice Prosser, who gets elected by the State of Wisconsin as a whole, we find that the seat is probably somewhat leaning D at this point, but I would put the margin of error much higher on this estimate – the race is still developing, and a Supreme Court race is very different from a Senate one.