Two Definitions of “Fair”
Note: I significantly revised this post 1/1/22.
We give you two ways to evaluate how fair a map is politically in DRA 2020:
- In the Proportionality section of the Analyze view, we report the likely disproportionality (meaning deviation from proportional representation).
- In the Bias Measures section of the Advanced view, we report many advanced measures of bias.
Paradoxically, there’s a difference between:
- Saying that a map deviates from proportional representation, and
- Claiming that a map is biased
In other words, there are two distinct definitions of what it means to say a map “biased” or “fair.”
The notion of proportional representation is pretty simple: the share of seats one party wins should reflect the share of votes it received. Proportional representation is what most people intuitively expect “fair” to mean — it is the obvious (little ‘d’) democratic ideal.
In contrast, “bias” is a more formal and complicated notion that political scientists and courts have used for decades. It is most commonly based on the concept of partisan symmetry, which means that it’s OK for the seat share that one party will likely win to be greater than the share of the votes they will likely receive, as long as that would also be true for the other party if the situation were reversed. In other words, “partisan symmetry is the principle that a particular share of the total votes received by a party should translate into a specific number of legislative seats, regardless of which party received that share of the total votes.”¹
Beyond being different, the two notions can sometimes be in conflict: a map that is considered biased according to some Advanced metrics can still likely yield low disproportionality. Moreover some Advanced metrics might indicate that the map favors one party while disproportionality suggests the opposite.
The reasons this is possible are complex.² The tl;dr is of all the measures we report only (dis)proportionality and the efficiency gap are reliable measures of partisan advantage. The other measures of bias either measure some other quantity (but not partisan advantage) or aren’t reliable across a wide range of statewide vote shares.
We originally chose to include disproportionality on the main Analyze tab instead of any of the metrics on the Advanced tab, for the simple reasons that:
- Proportionality is a basic little ‘d’ democratic principle, and
- People understand proportionality without explanation.
With our scale — almost 20,000 users! — and a such small team of volunteers, we simply don’t have the luxury of high-bandwidth communication with people. We had to do something simple & intuitive.
Only much much later did we discover that (dis)proportionality is a reliable measure of partisan advantage while most of the advanced metrics are not.
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.
Footnotes
- See the definition of partisan symmetry in Ballotopedia (link).
- See Estimating Seats–Votes Partisan Advantage.
