Basic HTML Version
518
KIRSCH
Tabl e 1
Mean Weight Loss (in Pounds) and Effect Sizes as a Function of
Hypnotic and Nonhypnotic Treatment
sample
size
Study
Bernstein & Devine (1980)
Deyoub&Wilkie(1980)
Wadden & Flaxman
(1981)
Bolocofskyetal. (1985)
Barabasz &Spiegel ( 1989):
Standard
Control"
Barabasz & Spiegel (1989):
Individualized
Weighted mean
NH
9
18
10
52
14
H
9
17
10
57
16
15
All assessments
Mean
pooled
SD
8.67
6.34
5.74
9.84
6.79
8.07
Weight loss
NH
7.18
4.65
7.47
6.83
2.87
6.00
H
10.03
3.65
5.73
16.24
7.50
14.1
11.83
Effect
size
0.33
-0.16
-0.30
0.95
0.68
1.39
0.66
Final assessment
Weight loss
NH
7.68
5.30
6.10
6.83
2.87
6.03
H
13.50
6.00
4.60
21.83
7.50
14.1
14.88
Effect
size
0.67
0.11
-0.28 "
1.52
0.68
1.39
0.98
Note.
1Ib. = 0.4536 kg. NH = no hypnosis;H = hypnosis.
" The pooled
SD
for this effect was 5.38 .
b
Control group data corresponding to both standard and indi-
vidualized analyses of Barabasz and Spiegel (1989).
effect sizes.
3
Posttreatment and follow-up standard deviations were re-
ported by Wadden and Flaxman (1981). Because standard deviations
were not reported by Barabasz and Spiegel (1989), I contacted M.Bar -
abasz, who supplied exact standard deviations for each group. Use of
these standard deviations indicated that Kirsch et al. (1995) and Allison
and Faith (1996) had underestimated the effect sizes for this study.
Estimated Standard Deviations
Exact standard deviations were unavailable for two studies
(Bolocofsky, Spinier, & Coulthard-Morris, 1985; Bornstein & Devine,
1980), both of which had reported the results of two-factor analyses of
variance (ANOVAs), with one between-participants factor (treatment)
and one within-subjects factor ( t ime). I obtained exact means and the
original ANOVA tables for the Bornstein and Devine study from P. H.
Bornstein. The mean square error for the within-subjectseffects in the
Bolocofsky et al. study was calculated using the methods described in
Winer, Brown, and Michels ( 1 9 9 1 ). I then estimated the pooled stan-
dard deviation for each of these studies as (MS
error
/l —r
2
) '
7 2
; Smith
et al., 1980), using Allison and Faith's (1996) estimate of the relation
between measures
(r
2
=
0.45).
As can be seen in Table 1, the standard deviations estimated in this
manner are larger than any of the known standard deviations, suggest-
ing that the resultingeffect sizes are conservative estimates. Allison and
Faith's (1996) effect sizes of 0.07 for the Bornstein and Devine study
and 0.2 7 for the Bolocofsky et al. study imply standard deviations of
41 and 35 Ibs. (18.60 and 15.88 kg) , respectively. As these are more
than four times the largest of the known standard deviations, the effect
sizes they reported for these studies are likely to be substantial
underestimates.
Results
Fro m thes e data , I calculate d mea n effect size s weighte d by
sampl e size , the variance s of the populatio n effect sizes , and
confidenc e intervals , as describe d by Hunte r & Schmid t (1990 ,
pp . 285-28 7 an d 437-438) . Th e mea n weighte d effect average d
acros s assessment s wa s 0.66
SD
(varianc e = .20 ,
p <
.01 ) an d
tha t fo r the las t assessmen t perio d of eac h stud y wa s 0.98
(varianc e = .36 ,
p <
.001) .
The varianc e in populatio n effect size s indicate s the presenc e
of a moderator . On e importan t difference betwee n studie s wa s
the lengt h of tim e betwee n the end of treatmen t and the fina l
assessment , rangin g from 2 month s in the Deyou b and Wilki e
stud y to 2 year s in the Bolocofsk y et al . study . Th e origina l
meta-analysi s reporte d a significant correlatio n betwee n effect
size and tim e of assessment . A recalculatio n of tha t relatio n us -
ing differences in mea n weigh t loss in plac e of effect size indi -
cate s tha t it is highly robus t
(r =
.74 ,
p <
.01) . Althoug h thi s
account s for muc h of the variabilit y in effect sizes ,within-stud y
differences (e.g. , Barabas z & Spiegel , 1989 ) sugges t tha t proce -
dura l difference s in the hypnoti c componen t als o contribut e to
differential effectiveness.
Discussion
Kirsc h et al . note d tha t the exac t magnitud e of the effect of
addin g hypnosi s to weigh t reductio n treatment s "i s uncertai n
becaus e of the failure to repor t standar d deviation s in mos t of
the weigh t reductio n studies " (1995 , p. 218) . In thi s thir d meta -
analysi s of thes e data , I reduce d the uncertaint y by obtainin g
actua l standar d deviation s for two of the si x comparison s
(Barabas z & Spiegel , 1989) . Thes e additiona l data , the us e of
different calculatio n methods , an d th e exclusio n of a stud y
deeme d "questionable " by Alliso n and Fait h (1996 ) resulte d in
3
Effect sizes can be calculated as the difference in means divided by
either the standard deviation of the control group (Glass, 1977) or the
pooled standard deviation (Hunter & Schmidt, 1990). Where actual
standard deviations were reported, those of the control group were used
by Kirsch et al. (1995) to calculate effect sizes. To facilitate comparison
with the data as reported by Allison and Faith (1996), I used pooled
standard deviations in thisreanalysis.