The best-fit model predicted risk for DA in S2 as a function of the log of kilometres between S1 and S2 with parameter estimates (±95% confidence intervals) of 0.94 (0.92; 0.95). Prediction of DA included effects of cohabitation and an interaction of proximity and time since S1 registration with stronger effects of proximity early in the follow-up period. Proximity effects were stronger for smaller S1–S2 age differences and for same- v. opposite-sex pairs. Sibship analyses confirmed sibling-pair results.