Monday, March 14, 2022

Understanding District Partisan Lean in 2022: A Case for Chilling Out

 by Christopher Cooper

At long last, we know the details of North Carolina's State House, State Senate, and Congressional districts. Soon after they were enacted, think tanks, academics and journalists began to analyze how these districts leaned according to various partisan metrics. Sometimes these competing metrics can get a little confusing, so I'll review a few of them below (note: not an exhaustive list), describing how they are calculated. The bottom line, however, is that which metric you choose doesn't matter much in how you understand the partisan lean of North Carolina's districts--a point that I'll review in greater detail below. 


  • 2020 Two Party Presidential Vote Share: The "two-party" language simply means that the third party candidates are removed. So, the denominator only includes Democratic votes + Republican votes. This is a standard adjustment Political Scientists and data journalists make to make results more comparable from election to election.
  • 2016 Two Party Presidential Vote Share: Similar to above, but for 2016 instead of 2020. 
  • Bitzer's Big Three: Political Scientist and all-around good guy Michael Bitzer frequently uses this one in his analysis. It's the two party vote share from the the 2020 Presidential, gubernatorial, and U.S. Senate elections averaged together. The logic for this measure is that all three are high profile races with clear implications for partisan politics, but each has slightly different characteristics.
  • Civitas Partisan Index: Civitas has used a few different formulas over the years to compute their CPI (all reliable and accurate measures), but the current version uses the average two-party vote share over the 10 Council of state elections in 2020.
  • Competitive Council of State Elections: I often use this metric, which is a combination of the two most competitive council of state elections in the previous year (Secretary of Labor and Attorney General). Since they were both close elections, a Democrat won one and a Republican won the other, and neither election is very high profile, I tend to think it gives a good picture of baseline partisanship.
  • 2020 Composite: I've used this measure as well--it's an average of the two party vote share across every 2020 election in the NCGA's stat pack (if you want to play with the stat pack data yourself, I reformatted it into Excel. You can download it here). 
  • 2016-2020 Composite: The same as above, except it also includes the 2016 votes for President and Lieutenant Governor (the two used in the stat pack).

Usually this would be a time in a blog post when a Political Scientist would drone on about how measurement matters--about how small changes in the measure we use can produce dramatically different results. This isn't that kind of post. Instead, my message is simple (a rare moment of zen from me)--choose whichever measure you want; it won't make a dime's worth of difference in the conclusions you draw.

(In this case) Measurement Doesn't Matter 

The correlation coefficient is a simple, but fairly powerful statistic. It ranges from 0, indicating that two variables don't vary together at all (ex. number of days you wore Toughskins jeans when you were three years old and your intelligence) to 1, indicating a perfect correlation (ex. frequency with which you read Jeremy Markovich's NC Rabbit Hole newsletter and the amount of NC miscellany you know). Political Scientists almost never see correlation coefficients close to 1. If we get a correlation coefficient of .6, it's cause for celebration (warning: this is a one paragraph blog-post level explanation, don't use this to crib notes for your research methods midterm unless you want to get a C-).

The table below lists the correlation coefficients between a number of different ways to measure partisan lean of NC districts for each of the sets of new districts. The bottom line is that they are all almost perfectly correlated. Whether you use the 2020 Presidential election, the 2016 Presidential election, the 2020 Composite, the CCSC score, or the 2016-2020 composite you will get almost exactly the same answer. If there's a different set of election(s) you favor, I'll bet you dinner at Buxton Hall (or Plant for the Vegan readers) that it won't produce a substantively different answer, either.

Correlations Between Various Measures of Partisanship Under NCs New Legislative Districts



NC Senate

NC House

2020 Presidential: 2016 Presidential




2020 Presidential: 2020 Composite




2020 Presidential: 2020 CCSC




2016 Presidential: 2020 Composite




2016 Presidential: 2020 CCSC




2020 Composite: 2020 CCSC




2020 Big Three: 2020 Presidential




2020 Big Three: 2016 Presidential




2020 Big Three: 2020 Composite




2020 Big Three: 2020 CCSC




StatPack 2016/2020 Composite: 2016




StatPack 2016/2020 Composite: 2020 President




StatPack 2016/2020 Composite: 2020 Big Three




StatPack 2016/2020 Composite: 2020 Composite




StatPack 2016/2020 Composite: 2020 CCSC




Note: All measures are Republican two-party vote share (third party candidates were removed). CCSC is the Competitive Council of State Composite and includes results for the two closest Council of State Races. 2020 Composite is a composite of 10 statewide elections from 2020. 2020 Presidential and 2016 Presidential are simply the Republican two-party vote share. 2020 Big Three is the two party Republican vote share for the President, Governor and US Senate races in 2020. StatPack Composite is the composite score calculated from the Stat Pack, it includes all of the elections in the 2020 composite, plus the 2016 LG and Presidential Races. All data are from the stat pack on the General Assembly web site.


These incredibly high correlations are the result of two factors. First, most of these measures use at least some of the same elections, so it's not surprising that they're related. Second, modern voting behavior is stable and predictable. One election predicts another election with shocking frequency.


What Does R+4 mean?

While all of the vote share measures give you almost exactly the same answer, the scores begin to look somewhat different when translated into district lean--the familiar R+4, D+2, and the like. The idea behind translating vote share into a "district lean" is that you can see at a quick glance which party is favored in a given district and by how much.  


One reason for any apparent difference in district lean is that most measures begin with an assumption of a two-party vote share, but some do include third parties. While this does not matter in all elections, sometimes there is enough third party support to have a sizeable impact on the district lean reported. Remember Ross Perot in 1992? (If so, I just dated both of us)

The other difference is a result of what the party's vote share is compared to. Imagine a race for dog catcher where the Republican candidate wins 55%-45% overall, but 60%-40% in District 1. To compute the partisan lean for dog catcher in District 1, you'd have two options:

  • Absolute Lean: compare the party's vote share to the other party's vote share in that district. Using our dog catcher example, District 1 would be an R+20 district (60-40=R+20).

  • Relative Lean: compare the party's vote share in that district to the statewide vote share for the office. Using this measure, District 1 would be an R+5 (60-55=R+5).

Both the absolute and relative lean measures are defensible and reasonable. They just give slightly different answers for partisan lean. But, because they all come from almost perfectly correlated measures of partisan vote share, they will provide almost identical answers when considered in context.

What these Measures Do and Don't Tell us

These measures, of course, apply to the district, not the candidates. Incumbency, the campaign itself, and a variety of other factors can and do matter in the outcome. Nonetheless, these measures do a remarkably good job predicting the outcome of future elections.


In Conclusion: Relax

I spend a lot of time thinking about how better measurement can give us a more accurate picture of political reality. In this case, however, measurement is overrated. Given the polarized and predictable nature of today's political environment, any of the commonly accepted metrics of recent election results will give you almost exactly the same results. So, pick the one you like best and focus on the substance of what the measures tell you, rather than arguing about which measure is the "right" one.



Chris Cooper is the Madison Distinguished Professor of Political Science and Public Affairs at Western Carolina University. He tweets at @chriscooperwcu