A negative feedback-loop model page 10

ruminative thought, or behavior. However, the correlations between the latent variables

indicated by these items were all above .6. Models that examined paths to these latent

variables separately could not, as one might expect given the degree of collinearity, be

reliably estimated.

Conceptually, this high latent correlation suggested to us that the underlying

construct of aggressiveness was more meaningful in this age population than discrete

measures of cognitions, values, and behavior. For example, behavior alone may not

measure aggressiveness with much precision in a young adolescent population, as

actually enacting physical aggression might be constrained by factors such as small

stature or socialization against such enactment among females. Conversely, some young

people may enact aggressive behaviors because of peer norms and pressures without

being in fact aggressive by inclination or preference. We therefore considered a higher-

order factor model, in which the three latent factors would in turn represent a higher-

order aggression factor. Such models, however, typically require at least three indicators

for each of the lower order factors, and only two per factor were available in this data set.

For each time lag, then, a single aggression construct was constructed that was composed

of the three indices as indicators, without including the individual items comprising the

indices directly in the model.

*Control variables*. Four control variables were included in the model. The first,

sensation-seeking, was a latent factor measured using three indicators that inquired about

engaging in dangerous activities without worrying about the consequences, each

measured on a five point scale (1 = *not at all*, 5 = *very often*). These items have been

found to correlate well with more traditional measures of sensation-seeking (citation