Off the Kuff Rotating Header Image

Three different views of how competitive Texas Congressional districts are

You are by now familiar with The Crosstab and its 2018 Congressional forecast, as I’ve referred to it multiple times in recent months. G. Elliott Morris, the proprietor of The Crosstab, now works at The Economist, and they have their own midterm forecast, based on a different model. And of course you are familiar with FiveThirtyEight, which (guess what) also has a 2018 Congressional forecast. What do we like to do when we have multiple data sources? We compare them, that’s what. Here are the forecasts for the Texas Congressional districts of interest:


Dist   XTab   Econ    538
=========================
CD02   15.1   50.0    9.3
CD03    6.4   29.0    0.9
CD06   16.5   40.0    7.3
CD07   50.3   27.0   46.6
CD10   21.0   13.0    2.6
CD14    6.4    1.0    1.3
CD17    4.4    1.0    0.5
CD21   20.3   50.0   20.3
CD22   20.5    8.0   14.8
CD23   66.3   83.0   49.4
CD24   29.0    9.0    4.5
CD25   11.3    0.0    7.0
CD27    6.5   20.0    0.3
CD31   11.2    4.0   20.1
CD32   47.4    1.0   17.4

Left to right, those are the projected percentage chances of a Democratic win in the given district from the Crosstab, the Economist, and 538. I have a hard time taking the Economist’s model seriously so I’m not going to say anything more about it. The other two provide an additional piece of data that’s worth looking at, which is a projection of the margin between the candidates in each district. Let’s look at that as well:


Dist   XTab    538   2012    2014   2016
========================================
CD02  - 9.1  -10.6  -26.9   -33.7  -20.3
CD03  -12.9  -20.6  -30.8   -37.1  -25.7
CD06  - 7.1  - 9.8  -16.7   -21.3  -15.2
CD07    0.1  - 0.6  -19.9   -31.4  -14.0
CD10  - 6.0  -15.9  -18.7   -22.6  -16.5
CD14  -12.4  -19.0  -15.2   -22.8  -21.3
CD17  -13.2  -22.5  -19.8   -28.9  -22.7
CD21  - 5.6  - 6.3  -22.0   -26.0  -18.7
CD22  - 6.2  - 8.5  -25.1   -33.3  -16.3
CD23    3.2  - 0.2  - 3.1   -15.5  - 1.2
CD24  - 4.0  -13.4  -23.3   -30.9  -15.9
CD25  - 9.0  -11.7  -20.0   -22.5  -21.5
CD27  -13.4  -25.4  -19.1   -30.3  -24.4
CD31  - 9.2  - 5.9  -21.0   -27.7  -19.3
CD32  - 0.4  - 6.6  -15.4   -23.7  -12.2

These numbers are all as of the September 23 update. They may have drifted a bit since then, one way or the other. I can tell you that in the few days that it took me to get my act together and finish writing this post, Democratic odds dropped a bit across the board, with the exception of an uptick in the 538 model in CD32. I have no idea if the loss in SD19 had anything to do with that, or if it was just reflecting whatever ebbs and flows their equations pick up on.

Note that 538 provides a range of possible vote shares for each candidate. In CD07, for example, they forecast a high of 55.1% and a low of 45.5% for John Culberson, with a high of 54.5% and a low of 44.9% for Lizzie Fletcher. The margin is the difference between the average forecast, which as of September 23 was 50.3% for Culberson and 49.7% for Fletcher. They take third party candidates into account as well. The Crosstab isn’t quite that granular, they just provide a forecast for the margin. Again, these numbers will drift around some.

The two points of interest here are where the forecasts differ – 538 is more bearish overall for Dems, though more optimistic in CD31 and considerably more pessimistic in CD32 – and how the current margins compare to previous years, which I show in the last three columns. As is my custom, I’m using judicial numbers for those margins, and for this I used one of the statewide judicial races. I don’t remember which one I picked for each year – did I mention that it took me a few days to write this post? – but it really doesn’t matter that much, they’re all within a point or so of each other. Look at the comparisons in the Harris County CDs – 02 and 07, which are entirely within the county, and 10 and 22, which are partially so. Everywhere you look, the indicators are for a rough year for the Harris County GOP. Never take anything for granted, of course, but dismiss the data at your peril as well. The postmortem here is going to be something.

Related Posts:

Comments are closed.