Evaluating Partisan Performance

Published in

Dave’s Redistricting

4 min readSep 23, 2020

You can evaluate the partisan characteristics of a map three different ways in DRA 2020:

All of them depend on the quality of the election data that you use. Absent specific knowledge about a state and the specific elections — or specific goals for your analysis — the best choice is the default election composite.¹

Methodology

To estimate the partisan characteristics of a map, we use:

A composite of statewide elections, and
Fractional seat probabilities instead of all-or-nothing accounting, except as noted.

District Statistics

The first way to characterize a map from a partisan perspective is to simply count the number of first-past-the-post (FPTP) Democratic and Republican “wins,” using the “Partisan Lean” vote shares in district Statistics. For this sample map, it yields a statement like “This map is 10R-3D.”

A slight refinement on that is reflected in the note below the table which incorporates a simple notion of competitiveness. For this sample map, it says “Eight districts lean Republican, three lean Democratic, and two fall in the 45–55% competitive range.”

The colors in the table correspond to how districts are colored by partisan lean, except that here districts that fall in the 45–55% competitive range aren’t colored.

Analytics View

While that approach is useful for a basic understanding of a map, it has two significant limitations:

FPTP accounting is all or nothing — It doesn’t admit any nuance or chance that competitive districts might actually “flip” to the other party.
Just focusing on the D–R (or R–D) “split” ignores what a proportional delegation should look like — It doesn’t incorporate the statewide vote share.

The second way to evaluate the partisan characteristics of a map is using the deviation from proportionality in the Proportionality section of Analytics view which addresses both of those issues.

Proportionality Section of Analytics View

Disproportionality measures how much the likely number of Democratic (or Republican) wins would deviate from the integral number of districts closest to proportional based on the average statewide vote share, i.e., there is a normative baseline.

Moreover, in contrast to simplistic FPTP accounting, this metric uses a probability distribution to estimate how likely it is that Democrats (or Republicans) will win the districts. In other words, instead of an all-or-nothing 0 or 1, the seat probabilities are fractions [0–1]. The tails of this probability distribution approach zero at 40% and 60%. A vote share of below 40% or above 60% is a sure fire loss or win, respectively, while in between districts get more competitive as you approach 50%.

The raw disproportionality is normalized to the range [0–100] where bigger is better, to make it easier to interpret the values.

When you color districts by partisan lean in Map view, the coloring reflects these seat probabilities. In other words, coloring districts by partisan doesn’t just use a simple red-to-blue linear gradient based on vote share like coloring precincts by partisan lean.

Advanced View

While that proportionality (and deviation from it) is a simple, intuitive concept — especially with the normalized rating — the more academically & judicially accepted class of metrics measure partisan bias.²

The third way to assess the partisan characteristics of a map is Advanced view which includes:

The two diagrams are powerful ways to visualize the various aspects of the degree of partisan bias. Different political scientists, litigators, and judges prefer some metrics over others, but they are a superset of what various experts use. The first five measures have been shown to reliably measure partisan bias, even in “unbalanced” states where the statewide vote share split is not close to 50–50.³