To examine the sensitivity of observed Depression x Task x Valence
interaction, the data was aggregated using no cutoffs for short or
long reaction times, 2 second rt cut-offs and medians rather than the
means of reaction times for each person in each condition. Similarly,
data may be aggregated using an inverse transformation to weaken the
effects of outliers (Ratcliff, 1993). Under each of these
transformations the Depression x Task x Valence interaction was
statistically significant, and t#tex2html_wrap_inline1484# was similar to that obtained for
the 150ms low, 5 second high cutoff condition, as shown in Table 3,
p. 76. Moreover, no higher order interactions with depression were
present in any other condition, mediating the three way
interaction. The inverse transformation condition did yield a
statistically significant Depression x ISI x Valence interaction, F(4,67)=2.88, p=0.029,t#tex2html_wrap_inline1486#=0.147, possibly due to the extra
weight given to very short rt's which may have been guesses, at lower
SD's. In no case were there statistically significant interactions
between gender and depression, though, for the median aggregation, a
Gender x Task x Valence interaction was statistically significant,
F(2,69)=3.53, p=0.035,t#tex2html_wrap_inline1488#=0.093.
Table:
Statistical Properties of the Depression x Task x Valence Interaction
Under Different Methods of Data Aggregation
