To help interpret the significant 3-way interaction, treatment modality and retention were examined within each level of PSI scores (Figure 1). The weekly use of cocaine was higher among short-term than among long-term patients in the medium- (44.0% vs 20.5%; F1,719=43.69; P<.001; ES=0.54) and high- (36.7% vs 22.6%; F1,213=4.54; P<.03; ES=0.32) problem severity groups but not for those in the low-problem severity group (21.9% vs 16.8%; F1,655=2.51; P<.12; ES=0.13). Further examination showed that for medium-severity problems, differences between modalities for short-term patients were not significant (42%, 43%, and 47%, for LTR, ODF, and STI, respectively; F2,204=0.16; P<.86; ES=0.04); among long-term patients with medium-severity problems, however, cocaine relapse rates were significantly higher for those in STI programs (25%) than ODF programs (15%) (t=2.26; P<.03; ES=0.25). Results were similar for the high-problem severity group in that differences between modalities for long-term patients were significant (15% [13/89], 29% [11/38], and 38% [12/32] for LTR, ODF, and STI, respectively; F2,156=4.23; P<.02; ES=0.23) but not for short-term patients (38% [16/42], 27% [3/11], and 43% [3/7] for LTR, ODF, and STI, respectively; F2,57=0.27; P<.76; ES=0.10). Post hoc t tests showed that long-term, high-problem patients in LTR programs (15%) had significantly fewer weekly cocaine users than the long-term, high-problem patients in STI programs (38%) (t=2.69; P<.008; ES=0.56), and the differences between those in LTR and ODF programs (29%) approached significance (t=1.79; P<.07; ES=0.35).