Fixed Effect Estimates and Deviance Statistics from the 3-Model Sequence for Each Dependent Variable within Each Thought Condition.

Note.*p**p***pAll models include thought-content contrast codes as covariates. AGE is coded 0 for younger adults (n = 42) and 1 for older adults (n = 38). TIME (coded 0–3) is the linear effect of the four successive thought sequences of a given instructional condition (Suppression or Monitor) and TIME2 is the quadratic effect. Frequency and Duration are log-transformed to reduce positive skew, with respective transformed ranges of 0 to 2.9 and 3.8 to 11.6 (un-transformed ranges are 1–17 for Frequency and 46–104,890 ms for Duration). Difficulty ranges from 1 to 5.The AGE initial status: Younger parameter is the model-estimated value during the first thought sequence for the younger group (the model intercept because this group is coded 0). The AGE initial status: Older (difference) parameter is the older group's difference from this intercept value. In Models A and B, the AGE initial status: Older (difference) parameter can be interpreted as a level effect of age. In Model C, however, by including the age-by-time interaction terms, the AGE initial status: Older (difference) parameter now reflects the estimated older adult difference during the first thought suppression sequence. Analogously, in Model B, the TIME and TIME2 parameters index the linear and quadratic change trajectories for the full sample, collapsed across age. In Model C, however, the TIME and TIME2 parameters now index the linear and quadratic change trajectories for younger adults (the group coded 0), and the interaction terms, AGExTIME and AGExTIME2, estimate how the older adults’ trajectories differ from those of the younger adults.Deviance (−2LL) = −2*the sample log-likelihood, an index of the difference between the current model and a perfectly fitting saturated model (Singer & Willett, 2003). It follows a standard chi square distribution. LRT (Likelihood Ratio Test) = ?−2LL/?df from the previous model. In addition to the fixed effects shown here, all models also included estimates of the within- and between-person variance and covariance parameters that are characteristic of multilevel models. This is why, for example, the change in df from Model A to B is 7 rather than 2.Results of random level-2 effect estimates (Intercept, TIME and TIME2) are not shown. Intercepts varied significantly across individuals in all models, but effects of TIME and TIME2 did not vary significantly for five of the six dependent variable-thought instruction conditions. The exception was for perceived difficulty in the suppression condition, where significant variation was evident for both the TIME and TIME2 effects. That is, after accounting for the systematic effect of age, significant interindividual variability remained in participants’ patterns of perceived difficulty over time.