Pace is a real but MODEST variance lever: across the observed NBA pace range (~92-104) the same matchup's upset prob shifts by only a few points -- fewer possessions favor the underdog (sqrt(N) scaling), but the effect is small because r...
Fewer possessions mean fewer chances for the better team to separate, so the underdog's chances grow as pace slows -- the sqrt(N) scaling of a possession-margin model. But because real NBA pace barely moves game to game, the mechanical effect on any single matchup is small: a couple of upset-probability points, not a coin flip. A slow, grind-it-out game genuinely helps the underdog, just not by much.
Caveats & confounds
COUNTERFACTUAL floor: per-possession scoring margin and variance are held pace-invariant, so this isolates the mechanical variance effect only, not behavioral changes (teams that push pace may also change shot quality). Margin is assumed Normal (CLT over ~100 possessions), weaker for short in-game windows.
