Stevens, Hynan, and Allen (2000) meta-analyzed 80 psychotherapy outcome studies; with each study containing a no-treatment control, a common factor control, and a full treatment group. It was hypothesized that the relative magnitude of common factor and specific factor ESs would be moderated by the outcome domains of the phase model of psychotherapy (Subjective Well-being, Symptoms, and Life Functioning). All ESs were calculated as r, which is usually .5 x d. Unless otherwise noted, all ES listed below are sig, p < .05
Overall Results:
Common factor ES = .11: Common factor control – No treatment
Specific factor ES = .19: Complete treatment – Common factor control
Complete treatment ES = .28: Complete treatment – No treatment
ESs were not moderated by
phase model outcome domain, reactivity of measure, common factor credibility,
or treatment length. ESs were moderated by severity
of disorder (Chronic/characterological vs.
Acute/circumscribed): Common factor/more
severe, ES = .05, ns: Common factor/less severe, ES = .18: Specific factor/more
severe, ES = .18; Specific factor/less severe, ES = .19.
Conclusions:
The relative magnitude of specific to common factor ESs is larger than that reported in previous meta-analyses, which usually show both components having relatively equal magnitude.
Limitations: There is great heterogeneity in the common factor controls, and some common factor controls may have been designed as week conditions.
Luborski et
al., 2002: The Dodo is Alive…
This lead article is
an examination of 17 meta-analyses of comparative psychotherapy outcome
studies. The mean ES was d=.20, ns. After correction for researcher allegiance,
the mean ES reduced to d=.12. In previous research, Luborski
et al (1999) found that their set of researcher allegiance measures correlated
r = .85 with outcome. Possible explanations for this lack of difference are
offered. The explanation preferred by the authors suggests that the
pervasiveness of common factors across treatments limits the differences found
when treatments are compared (Rosensweig, 1936).
Other possibilities include (a) researchers with different allegiances
canceling out the allegiance effect across many outcome studies and (b)
methodological problems such as the lack of statistical power in comparative
outcome studies.
Messer and Wampold,2002, commentary
They like what Luborski has to say. Wampold,
1997, reported a meta analysis of comparative outcome
studies and also found an ES of d = .20, small and non-significant. In fact
Messer and Wampold believe that d = .20 is an
overestimate because they view some comparative therapies as not being genuine bonafide treatments. So Messer and Wampold
firmly claim that Common Factors are more powerful than specific therapy
ingredients. They argue against the findings of Stevens et al (2000) who
reported ESs for specific factors about twice as
large as ESs for common factors. They base this
argument on the contention that many of the common factor controls examined by
Stevens et al. may not have been treatments of integrity (which is probably
true). To bolster their conclusion, Messer and Wampold
also present data comparing the proportion of variance in outcome accounted for
by different treatments (1%) with the proportion of variance in outcome
accounted for by the therapeutic alliance (5-7%) and the proportion of variance
in outcome accounted for by the person of the therapist (6-9%).
Stevens, Hynan, Allen, & Braun (submitted manuscript), “Is psychotherapy more than skills acquisition with a helpful teacher?”, was a meta analysis of 33 studies, each containing at least one “more complex” complete treatment and one “less complex” treatment. 15 of the 33 studies contained a no treatment group. “Less complex” treatments were either Progressive Muscle Relaxation or Biofeedback. We argued that the less-complex treatments constituted bonafide controls for common factors.
Hypotheses: That “less complex” treatments, which contain common factors of skills acquisition in the context of a helpful relationship, would have a significantly larger common factor ES than that observed in Study 1. We also hypothesized that the more complex treatments would prove superior to common factor controls (aka “Is psychotherapy more beneficial than scuba lessons with Mr. Rogers?”)
Results:
Less complex ES = .19: less complex treatment – no treatment
More complex ES = .08: more complex treatment – less complex treatment
Complete treatment ES = .25: More complex treatment – no treatment
Even though it constitutes a small ES of .08, there is enough power in
this meta analysis to reveal ES = .08 significant, p
< .05. The more complex component constitutes 28% of the magnitude of the
complete treatment ES.
Limitations:
Many of the studies in Stevens et al. are limited to treatments of physiological and psychosomatic disorders.