How We Rate Partisan Bias
This note describes how we rate partisan bias which is a more formal and more complicated notion of fairness than simple proportionality.¹
Note: We originally added this as a note in the Proportionality section on the Analyze tab. We have moved it to the Bias section of the Advanced tab.
Votes Bias and Seats Bias
The Bias section of the Advanced tab shows many measures of partisan bias.² Because larger values mean more bias, smaller raw values of partisan bias are better.
The votes bias and seats bias measures of partisan bias have straightforward graphical interpretations on the seats-votes curve.³ These are the fraction of votes more than (or less than) half that Democrats need to win half the seats and the fraction of seats less than (or greater than) half that Democrats win with half the votes, respectively. These correspond to the black diamond and black square on the sample shown above.
Normalizing Partisan Bias
As with the metrics on the Analyze tab where it’s hard to know how good or bad raw values are, we normalize raw votes and seats bias estimates to the range [0–100] where bigger is better.
We chose the following function to normalize raw values of bias (B) into ratings (R):
using 2% and 6% for h for votes and seats bias, respectively.⁴
This function inverts bias so that when bias is zero the rating is 100, makes the rating 50 for votes bias and seats bias values of 2% and 6%, and increasingly penalizes (lowers) the rating the more bias there is.
We combine these into a single rating for partisan bias, by averaging the two individual ratings.
Footnotes
- See Two Definitions of “Fair”.
- See Advanced Measures of Bias & Responsiveness.
- See Seats–Votes Curve.
- This is the formula in Microsoft Excel:
=ROUND(100*EXP(-ABS(B/b)),0)and the Normalizing Partisan Bias spreadsheet implements the rating formula and applies it to many states, where BS_50 and BV_50 correspond to seats bias and votes bias, respectively. See the Assumptions tab for how canonical values are rated.
