Election Season Math: In Review

Following from some earlier posts about number-crunching in election season, here’s a starter for what I saw as the Early Votes came in …

At the county level, there were multiple ways of looking at how the Early Vote crowd voted. The two different individualized scores showed the electorate at 45.45% Dem and 54.16% Dem at the pessimistic and optimistic ends of the spectrum. Accounting for the biases seen in the 2008 electorate, these two scores placed the outcome at about 48-49% Dem.

The DPI-by-Precinct scores showed the Obama results clocking in at 49.98% Dem – less than a half point behind the pace from 2008. And since that score is in a multi-candidate field, it would be enough to open up Election Night with a small lead.

The reality was an opening mark of 48.00% (behind Romney) for Obama and a closing mark of 49.38% (ahead of Romney) for Obama.

Below is what the results showed among some of the districts in Harris County with contested district races. Two obvious DPI scores I chose were ones that had an easy correlation to 2012: Barack Obama and Adrian Garcia. I didn’t expect either of them to do as well as they did in 2008, so I added CO Bradford’s results from his District Attorney race to see how a narrow loss in 2008 might translate. Later in the process, as the results started showing a very 50-50 county, I added two judicials: Ashish Mehendru and Josefina Rendon. Rendon was the narrowest win in 2008 and Mehendru’s showing was the low end of the range from that year.

Finally, to compile a somewhat pessimistic average DPI of my own, I folded each of these into an average DPI score. Traditionally, the DPI scores created by others are just that – an average. And on a good day, they’ll tell you what they’re averaging.

                                    DPI SCORES
                     -----------------------------------------------------
        | ACTUAL EV |   Avg   | Obama  Garcia  Bradford | Mehendru  Rendon
--------|-----------|---------|-------------------------|-----------------
HD134   |   42.6%   |  43.0%  | 46.1%   49.8%   39.9%   | 39.7%    39.5%
HD135   |   37.7%   |  39.5%  | 39.3%   45.2%   38.3%   | 36.6%    38.0%
HD137   |   63.6%   |  63.2%  | 62.9%   67.5%   62.0%   | 61.2%    62.6%
HD139   |   79.2%   |  77.5%  | 77.0%   81.0%   76.7%   | 75.9%    77.1%
HD141   |   89.4%   |  86.5%  | 86.1%   88.2%   86.4%   | 85.6%    86.4%
HD143   |   71.0%   |  68.5%  | 64.4%   74.2%   68.0%   | 66.4%    69.7%
HD144   |   46.6%   |  51.7%  | 47.4%   58.6%   51.4%   | 48.8%    52.3%
HD149   |   58.9%   |  55.0%  | 54.6%   58.9%   54.1%   | 53.1%    54.2%
HD150   |   28.6%   |  31.7%  | 31.2%   37.7%   30.9%   | 28.8%    30.1%
CD2     |   30.6%   |  36.1%  | 36.6%   42.5%   34.9%   | 33.4%    33.4%
CD7     |   34.7%   |  38.5%  | 40.1%   44.5%   36.6%   | 35.7%    35.7%
CD18    |   76.5%   |  77.6%  | 77.3%   81.1%   76.9%   | 76.3%    76.3%
Const 1 |   59.0%   |  61.3%  | 61.8%   67.3%   59.4%   | 59.1%    59.1%
--------|-----------|---------|-------------------------|-----------------
COUNTY  |   48.0%   |  50.3%  | 50.0%   55.5%   49.2%   | 47.9%    49.2%

I found each of these DPI scores instructive, so I let them stand on their own. For instance, Obama’s DPI isn’t necessarily instructive for many Hispanic districts – and I think this is visible in HD143. Obama tends to perform at the low end of the spectrum of scores and local Hispanic candidates typically perform better. For a district like that, looking at Adrian Garcia’s DPI score is more helpful. And averaging down that peak score is even more helpful.

What surprised me upon seeing the first Early Vote numbers, was how close the numbers were in HD137 and HD147. In the case of 137, there’s a small sample size of only 28 precincts. As mentioned before, I ignored the DPI number all through EV and chose to focus on the individual scores, which had our district between 57-59%. Hubert Vo’s HD149 is instructive for how crossover votes aren’t accounted for in a DPI model like this. That’s always going to be an X-factor unless you have a very well-developed ID program with the data fed into the voter database and tracked closely. Most State Rep campaigns won’t have the manpower for that, so you’re really left to play that by ear in the real world.

All in all, I continue to be surprised by how accurate this approach is. If you take a look at what Robert Miller posted on E-Day, you’ll see what has traditionally been offered. This is basically showing the count of “Hard/Soft Dems”, “Hard/Soft Republicans” and a big “unknown” category mistakenly identified as “Independent”. I’ve heard those counts offered year in and year out – usually as proof that Dems were going to win big right before they didn’t.

I’d suggest to everyone that there is a significantly better way to measure Early Voting without getting that 20-30% unknown factor. This is the second cycle that I’ve used this approach and it’s been fairly good. There’s certainly some artistic license that goes with it to determine what kind of DPI measurement is best for a given race. In 2010, we were staring at wildly optimistic DPI’s from vendors assuming that 2006 and 2008 were a good basis of comparison. The 2010 results I posted were based on the DPI scores provided by NCEC, which is the standard place a lot of campaigns go for that information. But I’d suggest that if you have a good enough research ability on your team, you can and should know your own turf better than a national political shop.

Anyways, that’s what I saw going on during Early Voting. I’d hope that results like this could be posted publicly at some point in the future. But campaigns are fairly selfish entities, so I’m not sure how likely that is.

Related Posts:

Leave a Reply

Your email address will not be published. Required fields are marked *