Election Season Math Primer

As the Texas-based portion of the world awaits Early Voting, I thought I’d highlight something I started using in 2010 on the Bill White campaign. Since it was a new method in 2010 that I didn’t have any background use on, I didn’t want to play it up too much at the time. Basically, it’s a Democratic Performance Index (DPI) of a district, county, or state, based on precinct-level turnout and precinct-level DPI multiplied out.

The theory behind the tool is simple: assume that whatever precinct-level metric you use is a static number regardless of who votes; as voters turn out, they don’t always do so in the same proportion on each day of Early Voting or even from past elections; multiply the turnout from whatever day of Early Voting you have and multiply it by that precinct level metric; do this for all precincts and you’ll have a pretty decent benchmark for what an election is going to look like.

Here’s an example of what two different elections might show for a mythical district with four precincts:

Four Precincts, Two Elections, Different Turnout (TO)

                  |    Election 1    |   Election 2
------------------|------------------|----------------
                  |    TO  Dem  GOP  |  TO  Dem  GOP
------------------|------------------|----------------
1.  50% D - 50% R |   100   50   50  |  50   25   25
2.  75% D - 25% R |   100   75   25  |  60   45   15
3.  25% D - 75% R |   100   25   75  | 100   25   75
4.  90% D - 10% R |   100   90   10  |  50   45    5
------------------|------------------|---------------
     TOTAL        |   400  240  160  | 260  140  120
------------------|------------------|---------------
                           60%  40%  |      54%  46%

Seeing the impact of turnout and the relative similarity in precinct-level performance has demonstrated to me that the vast majority of what the 2010 election showed was the impact of turnout changes moreso than any group of voters changing their minds about which party they supported at the polls. Without a doubt, there is always some amount of mind-changing that takes place from election to election. But the turnout differentials we saw in 2010 were nearly enough to explain the full impact that we saw on election outcomes. All I do to arrive at this is replicate the above example out over a county or district.

What I saw in 2010 with this method is as follows …

                    DPI-prec    ACTUAL
---------------------------------------
47 .... Bolton ..... 52.05      46.18
48 .... Howard ..... 49.11      48.53 (Won)
50 .... Strama ..... 55.39      54.84
52 .... Maldonado .. 46.21      38.00
93 .... Pierson .... 50.80      47.58
96 .... Turner ..... 46.03      47.60
101 ... Miklos ..... 49.49      48.18
102 ... Kent ....... 43.46      45.36
105 ... Haldenwang . 46.79      44.89
106 ... England .... 48.97      48.49
107 ... Vaught ..... 47.47      46.48
132 ... Mintz ...... 36.40      31.69
133 ... Thibaut .... 46.95      42.43
134 ... Cohen ...... 47.97      49.31
137 ... Hochberg ... 56.00      58.71
138 ... Camarena ... 44.00      35.48
144 ... Molina ..... 42.30      38.27
149 ... Vo ......... 51.20      52.23

And here’s what I saw in the countywide totals compared to the ending Bill White 2010 vote share, which was admittedly at the high end of the spectrum for Dems that year.

County ...... DPIprec .. Actual (Bill White)
---------------------------------------
Bexar ....... 50.43% ...  48.82 (Won)
Collin ...... 33.92% ...  33.12
Dallas ...... 55.31% ...  55.23
Denton ...... 37.00% ...  32.84
El Paso ..... 64.10% ...  61.29
Ft. Bend .... 44.80% ...  47.04
Galveston ... 44.21% ...  41.09
Harris ...... 48.65% ...  50.23
Hidalgo ..... 70.46% ...  66.82
Jefferson ... 52.64% ...  48.15
Montgomery .. 24.46% ...  22.62
Nueces ...... 48.97% ...  45.12
Tarrant ..... 42.45% ...  40.97
Travis ...... 60.13% ...  59.77
Williamson .. 40.11% ...  36.89

Among the districts and counties with the biggest disparities (Maldonado & Denton, for instance), I think it’s worth pointing out how much disagreement that myself and others had with what we thought were too-generous DPIs since they appear to have been weighted too much by the pro-Dem, 2008 wave.

Redistricting has done a number on comparing precincts from 2 years ago and 4 years ago. It was easier to manage this for the tiny little State Rep district that I’m working this time. I also don’t seem to have a current DPI score for precincts. And since I never agreed with the DPI scores that I saw in 2008 and 2010, I figured I may as well create my own and use slightly more pessimistic datapoints to use as benchmarks. That seems to rid me of much of the problem of using the DPI numbers provided by others that aren’t explained fully and that I would likely take some amount of issue with.

When I took those DPI measurements and stacked them against 2004 turnout, I get HD137 as 57.4% Dem. Compare that with an average of the 2008 statewide results at 61.8%. And for whatever it’s worth, the GOP-friendly BIPAC group scores the district as being 56.98% Dem. For my purposes … close enough.

My quick & dirty methodology was designed because of frustration over hearing a lot of the bogus numbers that seem to get parroted every election. There are individual classifications of voters as being Hard Dems/GOP, Soft Dems/GOP, and (my favorite) Unknown. I remember hearing from people who rake in multitudes more consulting dollars than I ever will about the importance of these numbers. But they have serious flaws and leave everyone guessing what “Unknown” really is (or why it accounts for 20-30% of the electorate) and why Hard Dems + Soft Dems never adds up to 50. Invariably, you’ll hear a campaign with zero shot of winning point out that Early Voting shows them leading, say 38-30 with 32% Unknown based on this measurement. It sounds great until the real votes start being counted. In short, this is the worst metric I’ve ever seen used in campaigns and it still amazes me how often it gets used. It just needs to die off … or at least stop being taken seriously.

I think there’s a lot of room for improvement over the way I use a DPI-by-Precinct method. But, to date, I’ve only tried to rely on it as a decent benchmark that offers as much of a binary count without some big “unknown” variable that nobody knows anything about. The results were surprisingly good from what I saw in 2010 and if anyone is interested in learning how to apply this method this time around, feel free to drop me a line.

Related Posts:

Leave a Reply

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