Table 3.
Selected first stage results, tests of reliability and validity of instrumental variables
| School attendance at 12 months follow-up | Criminal activity at 12 months follow-up | Employment at 12 months follow-up | Substance use problem severity at 12 months follow-up | Abstinence at 12 months follow-up | |
|---|---|---|---|---|---|
| Method | Bivariate probit | Bivariate probit | Bivariate probit | Two-stage least-squares | Bivariate probit |
| Instrumental variables | Distance to treatment facility | Distance to treatment facility; very difficult traveling to treatment | Distance to treatment facility | Distance to treatment facility; very difficult traveling to treatment | Distance to treatment facility; very difficult traveling to treatment |
| First stage dependent variablea | Treatment initiation | Treatment initiation | Treatment initiation | Treatment initiation | Treatment initiation |
| Distance to treatment Facility | −0.013* (0.006) | −0.011 (0.008) | −0.013* (0.006) | −0.003 (0.003) | −0.013 (0.009) |
| Very difficult traveling to treatment | – | −0.792** (0.273) | – | −0.277** (0.102) | −0.734* (0.293) |
| Tests of reliability and validity of instrumental variables | |||||
| Joint significance of instrumental variablesb | 4.02* (0.045) | 11.31** (0.004) | 5.79* (0.016) | 4.91** (0.008) | 10.98** (0.004) |
| Test of exclusion restrictionsb | n/a | 0.45 (0.504) | n/a | 0.6781 | 0.25 (0.616) |
Statistically significant at p<0.05
Statistically significant at p<0.01
Coefficients and robust standard errors (in parentheses) are reported. Regressions control for site, percentage in community with less than high school degree, percentage in community unemployed, median household income in community, gender, age, race, participation in residential program in 6 months before baseline, measure of substance abuse severity, parent’s education level, familial alcohol/drug problem, and referral from the legal system.
Wald test employed for bivariate probit models. Chi-square values and p values (in parentheses) are reported. F statistics and p values (in parentheses) are reported for two-stage least-squares models. When only one instrument is used, the joint significance test is the same as the individual test for the single variable.
Chi-square values and p values (in parentheses) are reported for bivariate probit models, and Hansen J statistic is reported for two-stage least-squares models.