Table 5.
Standardized Regression Coefficients (βs) Predicting 1990 Antisocial Behavior from Four 1988 Disciplinary Tactics and Psychotherapy by Covariate for Pre-Existing Behavior Problems
| Covariate for Initial Behavior Problems | Spank | Grounding | Remove Privileges | Send to Room | Psychotherapy | |
|---|---|---|---|---|---|---|
| None | ||||||
| Zero-order r | .27*** a | .25*** | .14*** | .13*** | .12** | |
| Controlling for other predictorsb | .18*** | .20*** | .11 | .14** | .39** | |
| Antisocial Behavior 1988b | ||||||
| Dichotomous | .13** | .17** | .11 | .10* | .28* | |
| Trichotomous | .10* c | .12* | .10 | .09f | .24* | |
| Continuous | .10 | .14f | .09 | .07 | .23 | |
| Externalizing Behavior 1988b | ||||||
| Continuous (measured variable) | .09 | .10 | .07 | .07 | .19 | |
| Continuous (latent variable)d | .04 | .04 | .04 | -.03 | .02 | |
| Gain in latent externalizinge | -.04 | -.07 | -.04 | -.05 | .01 | |
Note. N = 785 for the disciplinary tactics; 782 for psychotherapy. A positive β indicates that those receiving the disciplinary tactic in 1988 averaged higher 1990 antisocial behavior than those not receiving that disciplinary tactic, after controlling for the variables indicated. The βs are based on linear associations for the first row and the last 2 rows. The other βs are based on the effect size contrasting the weighted marginal mean antisocial behavior for one or more occurrences of the disciplinary actions (or seeing a psychotherapist) vs. the mean antisocial behavior associated with no occurrences of the disciplinary action or psychotherapy.
aThis equals the effect size estimated from Straus et al. [1] in Gershoff's meta-analysis,[5] since she based effect sizes on zero-order correlations. (r = .27 is equivalent to d = .56, using the DSTAT program used by Gershoff)[61]
bThe "other predictors" were those in the ANCOVA in Straus et al.[1]: gender, ethnicity, SES, cognitive stimulation, emotional support, and the 2-way interactions of the disciplinary tactic with each of those predictors. The 2-way interactions were not included in the structural equation modeling.
cOur best re-analysis of Straus et al. [1] Our estimate of β from their graph is .23, but the estimated βs from our 10 samples ranged from .10 to .13, before applying the natural log transformation to 1990 antisocial. In contrast, the F values for spanking in our 10 samples ranged from 3.0 to 5.3, which included the original F value (4.4) within its range.
dThe latent structural equation modeling analyses estimate linear associations between the frequency of each corrective action and Antisocial behavior in 1990, controlling for the other predictors in the model, including latent Externalizing problems, which has three indicators (Antisocial, Hyperactivity, and Headstrong), with correlated residuals between the latter two subscales. Fit indices: χ2s (13, N = 785) from 35.05 to 40.50, ps from .0001 to .0008; CFIs from .971 to .976; RMSEAs from .046 to .052.
eThis row summarizes the only analyses predicting the 16-item Externalizing score in 1990, in this case in terms of gain scores in that latent score from 1988 to 1990 (see Figure 2). Fit indices: χ2s (12, N = 785) from 7.65 to 13.06, ps from .36 to .81; CFIs from .996 to 1.000; RMSEAs from .000 to .011.
fP < .10.
*P < .05. **P < .01. ***P < .001.