temporal change. These examples indicate that a comprehensive
meta-analysis covering other mental health disorders may yield
quite different results.
The BDI has undergone some modifications during its 40-year
existence. The original BDI was revised and made more user
friendly in 1988, and given the acronym, BDI-Ia (Beck et al.,
1988). The latest version, the BDI-II, has incorporated an item
measuring hypochondriasis, changed the timeframe of symptoms
from 1 week to 2 weeks, and put more emphasis on measuring all
diagnostic criteria related to depression. Still, the forms are very
similar to each other (Beck et al., 1996). Despite these differences,
the treatment and control groups responded to the equivalent forms
at any point in time. Thus, these considerations should not pose
major threats to the validity of the current conclusions.
Very few studies (k 5) included correlations between the BDI
pre- and postintervention scores, requiring us to impute this value
for the remaining 65 studies. However, the potential for this value
to exert undue influence on the results does seem small for two
reasons. First, the variations in correlations need to be quite high
in order to change the ESs substantially. Second, and most importantly,
we have no reason to expect that the prepost BDI correlations
should change considerably over time. Although a shift in
therapy effect over the years changed the mean of the post intervention
BDI scores, the relative position between the pre- and
postscores should not have changed by much.
Recovery rates were calculated according to somewhat varying
criteria across the studies included in this analysis. The
most stringent criterion was a cut-off score for clinical depression
of 7 on the BDI, while the most liberal was 10. Although
this difference might not seem substantial, it could have a
confounding effect on the calculated total percentage of recovered
patients, and the correlation between recovery rates and
year of intervention.
A minor possible caveat relates to the time moderator. As all of
the studies‘ years were coded based on their publication dates, it is
conceivable that this date could vary somewhat from the actual
year of the intervention. However, it is reasonable to assume that
this discrepancy is similar to contemporary and older studies, and
that the difference between the publication and actual year of
intervention is not very large.