Table 2.
Multivariate HGLM analyses of lifetime non-medical use of anabolic steroids (weighted estimates)
| 1993 (n = 14610) | 1997 (n = 13714) | 1999 (n = 13143) | 2001 (n = 10552) | |
|---|---|---|---|---|
| Student (Level 1) characteristics | AOR (95% CI)a | AOR (95% CI)a | AOR (95% CI)a | AOR (95% CI)a |
| Sex | ||||
| Female | – | – | – | – |
| Male | 5.34 (3.27, 8.72)** | 8.52 (4.99, 14.54)** | 6.42 (3.85, 10.72)** | 7.42 (3.70, 14.91)** |
| Marital status | ||||
| Never married | – | – | – | – |
| Married | 1.68 (0.95, 2.94) | 1.11 (0.55, 2.24) | 2.30 (1.08, 4.94)* | 3.02 (1.35, 6.78)** |
| Age | ||||
| Under 21 | – | – | – | – |
| 21–23 | 1.55 (0.93, 2.58) | 1.35 (0.84, 2.17) | 1.22 (0.75, 1.98) | 1.02 (0.57, 1.80) |
| Over 23 | 2.08 (1.02, 4.24)* | 3.28 (1.68, 6.92)** | 0.72 (0.38, 1.36) | 1.24 (0.52, 2.95) |
| Intercollegiate athletics | ||||
| Non-athlete | – | – | – | – |
| Athlete | 2.43 (1.56, 3.78)** | 1.80 (1.20, 2.70)** | 1.63 (1.04, 2.55)* | 1.82 (1.06, 3.12)* |
| College (Level 2) characteristics | ||||
| Commuter status | ||||
| Non-commuter school | – | – | – | – |
| Commuter school | 0.41 (0.16, 1.02) | 0.96 (0.49, 1.87) | 1.94 (1.19, 3.18)** | 2.31 (1.07, 4.97)* |
| Estimated variance of college interceptsb | 0.20 | 0.05 | 0.24* | 0.83** |
Symbol (–) denotes reference category.
p < 0.05.
p < 0.01.
Odds ratios are adjusted for all predictors in the models, which include the student-level characteristics of sex, race, marital status, age, living arrangement, fraternity/sorority membership, grade point average, and sports participation, and the college-level characteristics of urbanization, geographical region, commuter status, school enrollment size, NCAA athletic division, admissions criteria and public/private status. The results for variables not significantly associated with lifetime non-medical use of anabolic steroids in at least two years are not shown (p > 0.05).
The p-values indicated for the variance components are based on chi-square statistics computed by the HLM software for testing the null hypothesis that a given variance component is equal to zero (Raudenbush and Bryk, 2002).