1.5. Statistical data analysis
All analyses were conducted with the SPSS 20.0 software
package (IBM, Armonk, New York). The training condition
variable was dummy-coded as 1 = mindfulness training;
0 = control training. Preliminary analyses evaluated success
of randomization (using chi-square and independent-samples
t-tests), tested for baseline to peak stress increases in the total
sample (using paired-sample t-tests), and tested for associa-
tions between treatment expectancies, psychological stress
perceptions, and biological stress reactivity (using Pearson’s
correlations). Preliminary analyses also evaluated whether the
control training program was effective at controlling for atten-
tion and expectancies across the three training days, using
mixed effect linear models (MLMs) and independent samples ttests.
MLMs, which are robust to missing data (relative to
listwise deletion of subjects in repeated measures ANOVAs),
were used for study analyses that included a within-subjects
variable (i.e., time). Participants who had partial missing data
over time were still included in MLM analyses, following intent-
to-treat principles. In all MLMs, variables of interest were
modeled as fixed effects using maximum likelihood
estimation. All MLMs modeled the repeated measures variable