Cognitive Predictors of Alcohol Involvement and Alcohol Consumption–Related Consequences in a Sample of Drunk-Driving Offenders

Abstract

Motivational theories of alcohol involvement emphasize a wide range of cognitive factors as precursors to “heavy” or high-risk drinking. Central to this consideration has been expectancies, drinking urges, triggers, and situational cues, all of which can synergistically or independently stimulate drinking. Unfortunately, empirical studies have scrutinized low-level or moderate drinkers drawn from the general population, and less is known about the role of cognitive factors as precursors to high-risk drinking. The present study examines the unique contribution of several measures of cognitive motivation to harmful alcohol use in a sample of convicted drunk drivers. Confirmatory factor analysis indicated the psychometric soundness of a model positing four latent predictor constructs assessing drinking urges/triggers, situational cues, positive and negative expectancies and outcome constructs assessing harmful alcohol use and perceived consequences of harmful drinking. A structural equation model indicated that each motivational construct was associated uniquely with both drinking and perceived consequences, with the largest overall effect in both cases associated with situational cues. Results are discussed in terms of identifying prominent cognitive factors that may foster harmful drinking among high-risk populations and their implications for treatment.

Keywords:

• alcohol involvement
• cognitive motivators
• confirmatory modeling
• drinking urges/triggers
• DWI
• expectancies
• situational cues

Notes

¹It is worth noting that expectancies have received the lion's share of attention with respect to motivational factors, perhaps owing to the wealth of information linking them with adolescent (Chen, Grube, and Madden, 1994; Christiansen and Goldman, 1983; Scheier and Botvin, 1997), young adult (Christiansen, Smith, Roehling, and Goldman, 1989), and adult alcohol use (Conners, Maisto, and Derman, 1992; Conners, O'Farrell, Cutter, and Thompson, 1986; Fromme, Kivlahan, and Marlatt, 1986; Marlatt and Rohsenow, 1980). Additional findings also implicate expectancies as predictors of high-risk drinking and suggest they may even portend alcoholism (Christiansen, Goldman, and Brown, 1985; Mann, Chassin, and Sher, 1987; Smith, 1994).

^a Standard ethanol content (SEC) is computed using a steady pattern chart, which captures the average consumption patterns based on beverage type, amount, and duration over several episodes. The formula for computing SEC is # oz × % alcohol × 2. SECs were computed for a normal drinking week and then computed per drinking day for a year, 3-month period, and peak drinking episodes.

(* p ≤ .05). Numbers on diagonal represent estimates of internal consistency computed by the Werts, Linn, and Jöreskog (WLJ) (1974) method for structural composites. The WLJ computational method provides a more conservative estimate of the lower bound Cronbach's (1951) alpha and corrects the estimate of internal consistency for measurement error. A series of fully saturated latent variable models that specified first-order constructs were conducted to obtain the parameter estimates used in the WLJ formula for computing alpha.

²The sample size was sufficiently large enough to conduct tests of factorial invariance and structural equivalence based on gender. Formal tests of measurement invariance examine whether the obtained dimensional structure (factor loading) differs based on gender. Following recommended conventions for testing model invariance (Byrne, Shavelson, and Muthén, 1989), we first tested factor loading invariance and obtained a model with partial invariance between female and male participants. Only one indicator out of a possible 20 constraints varied between male and female participants, and this included sexual enhancement, an indicator loading on positive expectancies (λ = 0.746 females andλ = 0.642 males). A model with one relaxed constraint significantly improved on the base model with no relaxed constraints, Δ χ²(1) = 7.8, p ≤ .001. The next step examined equivalence for the structural component of the model (paths from motivational constructs to problem drinking and consequence outcomes). A model positing equivalence for males and females fit well, and post hoc empirical specification searches indicated no improvements in the model fit would be obtained by relaxing any of the constraints. The final step involved testing whether covariance structures among the four motivational constructs and two outcome constructs differed based on gender. The LM test indicated two constraints should be freed including the correlation between situational cues and alcohol use (r _S,A = 0.437 females and 0.523 males, p ≤ .001) and between alcohol use and consequences (r _A,C = 0.562 females and 0.468 males, p ≤ .001) [nested difference test: Δ χ²(2) = 11.3, p ≤ .001]. Overall, the combined tests hypothesizing factorial invariance, structural invariance, and equality of covariances indicated little practical difference between male and female participants.

³An empirical test of a second-order measurement model that specified a single unitary construct reflecting cognitive motivation did not fit well [χ²(73) = 1508.92, p ≤ .001, NFI = 0.871, CFI = 0.876, SRMR = 0.103]. All the loadings for the four primary (first-order) constructs were significant and included drinking urges/triggers (λ = 0.660, p ≤ .001), situational cues (λ = 0.748, p ≤ .001), negative expectancies (λ = 0.475, p ≤ .001), and positive expectancies (λ = 0.640, p ≤ .001). The disparity in the magnitude of the standardized loadings indicates some appreciable conceptual differences in what each motivational construct assesses and supports retaining the primary first-order structure.