Preliminary Inspection of Study Variables
Prior to conducting main study analyses, preliminary inspection of variable distributions were conducted on the alcohol ladder and drug ladder scores separately, as well as on the combined score used in all subsequent analyses. provides the distribution of scores across the seven rungs of the alcohol ladder (N = 296), drug ladder (N = 335), and combined ladder (N = 394). The frequency distributions were similar across the three versions, with the largest number of scores placed on the highest rung (7: I have decided to quit drinking/using drugs and plan never to drink/use drugs again) for each of the three ladders. The alcohol ladder had a mean score of 4.32 (SD = 2.11), and the drug ladder had a mean score of 5.43 (SD = 1.59). Skewness and kurtosis were −.29 (SE = .14) and −1.21 (SE = .28) for the alcohol ladder and −.89 (SE = .13) and .15 (SE = .27) for the drug ladder, indicating that both variables were near normally distributed. As shown in , scores on the alcohol and drug ladders were positively correlated (r (237) = .42, p < .001). The combined AOD ladder had a mean score of 5.0 (SD = 1.79), skewness = −.61 (SE = .12), and kurtosis = −.56 (SE = .25), indicating a normal distribution.
Distribution of scores across rungs of the Alcohol, Drug, and Combined AOD Ladders
also contains information on zero-order correlations between the ladder scores and theoretically related variables used to provide evidence of convergent and concurrent validity in later analyses with covariates (Kazdin, 1992
). Of note are the intercorrelations among the four motivation-to-quit items: desire to quit, perceived difficulty of quitting, expectations of success at quitting, and importance of treatment. The small-to-moderate magnitudes of these correlations (Pearson’s r
range = .11 to .37) show that these items are non-redundant despite being susceptible to common source variance, common method variance, and social desirability biases that inflate correlations between related variables.
Thirteen baseline demographic and health characteristics (all those listed in Measures: Baseline Characteristics except number of prior treatment episodes and current treatment status) were examined for their strength of association with the combined AOD ladder using regression analyses. These variables were entered simultaneously into a regression equation predicting the combined ladder score. Analyses showed only four variables significant at the p < .10 level (this criterion was selected to liberally identify potential covariates for inclusion in subsequent validity analyses, thus providing more a rigorous test of ladder construct validity). Higher ladder scores were associated with more legal involvement, lesser employment history, older age, and ethnicity other than African American. In the analyses for convergent and concurrent validity that follow, these four variables were entered as a group in the first step of the regression equations to control for their effects on the combined ladder score.
Convergent and Concurrent Validity
Convergent validity of the combined ladder was examined for four conceptually related treatment motivation variables. Four separate hierarchical regressions were conducted to predict the ladder score. In step 1, four covariates (age, ethnicity, legal involvement, job history) were entered; in step 2, the given treatment motivation variable was entered. Note that β is the standardized partial regression coefficient for the given predictor; sr2
is the squared semipartial correlation, which represents the unique contribution of the given predictor to the total variance in the dependent variable accounted for by the set of predictors (Tabachnick & Fidell, 1989
). All motivation variables were significant in the expected direction, with higher motivation scores predicting higher ladder scores: desire to quit using alcohol/drugs (N = 320; β = .57, p
< .001, sr2
= .31), expectation of success at quitting (N = 295; β = .26, p
< .001, sr2
= .07), perceived difficulty of quitting (N = 295; β = .12, p
< .05, sr2
= .01), and importance of obtaining treatment for alcohol/drugs (N = 374; β = .12, p
< .05, sr2
Concurrent validity was examined for three AOD-related baseline variables expected to be associated with readiness to abstain. Three hierarchical regressions were conducted as described above. All variables predicted the combined ladder in the expected direction: number of prior AOD treatment episodes (β = .10, p < .10, sr2 = .01), being in drug-free treatment at baseline (β = .17, p < .01, sr2 = .03), and PDA in the prior month (β = .26, p < .001, sr2 = .07).
Predictive Validity: Main Findings
Predictive validity of the combined ladder was examined with a series of Generalized Estimating Equation (GEE) models, an extension of the General Linear Model that permits a within-subject repeated measures examination of change over time as well as correction of variance estimates for longitudinal correlated data within subject (Zeger & Liang, 1986
; Zeger, Liang, & Albert, 1988
). GEE models tested whether baseline ladder score predicted substance use outcomes (number of problem days due to AOD use, PDA) and participation in AOD services (days in drug-free treatment, receipt of AOD services) over one year. To account for factors that may confound an association between the ladder score and long-term substance use, several covariates were tested as predictors of the substance use variables prior to adding the ladder score: gender, age, ethnicity, housing, education, legal involvement, psychiatric treatment utilization, program assignment, baseline abstinence, prior treatment episodes, and employment history. The following covariates had a marginal association (p
< .10) with a substance use variable and were retained in the final GEE model for that variable: prior AOD treatment episodes, for number of days in drug-free treatment; ethnicity and prior treatment episodes, for receipt of AOD services; gender, legal involvement, baseline abstinence, and prior treatment episodes, for problem days due to AOD use; and legal involvement and baseline abstinence, for PDA. Effect sizes were calculated for significant results using Cohen’s d
statistic; according to Cohen (1988)
= .20 is a small effect, .50 is medium, and .80 or greater is large.
A significant main effect of the combined ladder was found for PDA only, with higher ladder scores predicting greater abstinence over the 12-month follow-up period (B(SE) = 3.86 (.86), p < .001, d = 2.13). This indicates that a 1-point increase in ladder score was associated with a 3.86 unit difference in PDA combined across all 12 months. We also analyzed time by ladder interactions were found for PDA (p < .10), number of days in treatment (p < .10), and receipt of AOD services (p < .01). The more liberal criterion value p < .10 was used to prevent Type II error, given that testing for interactions using centered product terms is a highly conservative approach. To diagnose these interactions, hierarchical regression models (retaining identified covariates) were used to test ladder score as a predictor of outcome at each of the four timepoints separately: 1, 3, 6, and 12-month follow-up. For PDA, which was assessed monthly, the twelve timepoints were collapsed into four quarters for analyses of interaction effects: months 1–3, 4–6, 7–9, and 10–12. Higher ladder scores predicted greater receipt of AOD services (OR = 1.26; 95% CI = 1.08, 1.48; p < .01; d = .13) and more days in drug-free treatment (B(SE) = .49 (.29), p = .10, d = .19) at 1 month only. Higher scores predicted greater PDA at each timepoint, with the strength of association decreasing somewhat after the first quarter: months 1–3: B(SE) = 4.44 (.83), p < .001, d = 2.46; months 4–6: B(SE) = 3.08 (1.03), p < .01, d = 1.70; months 7–9: B(SE) = 3.73 (1.16), p < .01, d = 2.06; months 10–12: B(SE) = 3.40 (1.17), p < .01, d = 1.88.
Predictive Validity: Exploratory Analyses
Three sets of exploratory analyses were conducted to articulate the main predictive validity analyses. First, predictive validity analyses were re-conducted separately for the alcohol ladder (n = 296) and drug ladder (n = 335) to check for substance-specific properties of the ladder method. Results mirrored those found for the combined AOD ladder score. Main effects of both the alcohol ladder (B(SE) = 2.71 (.86), p < .01, d = 1.50) and drug ladder (B(SE) = 3.48 (1.05), p < .01, d = 1.92) were found for PDA, with higher ladder scores predicting greater percent days abstinence over time. Ladder by time interactions were also found for both the alcohol ladder (p = .08) and drug ladder (p = .003). Interactions were probed using the methods described above for the combined ladder score. As with the combined score, higher scores on both individual ladders predicted greater PDA at each timepoint, with the strength of association decreasing somewhat after the first quarter.
Second, we examined the incremental value of the combined AOD ladder over the 1-item question about motivation to quit using substances, “How strong is your desire to quit drinking/drug use at this time?” The desire-to-quit item was entered as a covariate in the GEE model described above, prior to entering the combined AOD ladder. The desire-to-quit item was not a significant predictor of PDA, whereas the combined ladder score remained significant (B(SE) = 2.52 (1.19), p < .05, d = 1.39), supporting its validity over the single motivation item.
Third, all predictive validity analyses were re-conducted for three subsamples: those in drug-free AOD treatment at baseline (n = 114), those in methadone treatment at baseline (n = 102), and those not enrolled in treatment (n = 177). A main effect of ladder predicting PDA was found for those in drug-free treatment (B(SE) = 4.15 (1.38), p < .01, d = 2.29), and for those in no treatment (B(SE) = 3.02 (1.43), p < .05, d = 1.67). No main effect for PDA was found for the methadone subsample, and no effects were found for any other outcome. The time by ladder interactions found for the full sample were not significant in any of the three treatment subsamples, which may be due to reduced power that resulted from splitting the sample.