Abstract
<p>Time-series cross-sectional surveys document a decades-long rise in American affective polarization. But the measure is defined only over partisans, and who counts as a partisan is itself dynamic. Using the three ANES panels spanning 1990--96 and 2016--24, I separate the measured rise into two channels: hardening, within-person change among those who keep their partisan label, and churn, turnover in who holds it. Churn accounts for almost a third of the within-panel rise, about 1.6 times movers' share of the sample. Those who change partisanship already lean toward their destination party, in affect and predicted partisanship, before they move: a prospective signature of sorting. Because movers are less polarized than stayers, the aggregate trend overstates how much partisan affect has changed and understates animosity among committed partisans---bearing on how we quantify partisan animosity, interpret its movement, and judge what reversal requires.</p>