Section V: OUTCOME EVALUATION
Overview
I. Process vs. Outcome Distinction
A. Process – describing events within sessions associated with change
B. Outcome --effectiveness of therapy (TX ): A pre- post-change
II. Measuring Process Variables
A. Measures of change in--
-
Aspects of client verbalizations
-
Depth of emotion
-
Therapist interventions and consequences therefrom
B. Compliance with model
-
Did therapist do X, Y, or Z as stated in the treatment manual?
-
Someone rates session for therapist behaviors
III. Measuring Outcomes
A. Pre-, Post-, Follow-up
-
Controlling for repeated measures?
-
Intervening "treatments"?
-
Deterioration effect (does anyone get worse?)
B. Distinguishing between what is taught and what is therapeutic
People can learn what you taught them, but is that
therapeutic ?
I can learn to spin around but what condition does
that improve?
C. Ecological or External validity
Who's responsibility is it to show that the skills
taught in treatment
ARE effective (i.e., the right ones) for bringing
about a better
relationship ?
D. Sources of information
From whom do we get reports of improvement?
Clients, Therapist (?), Outsiders
IV. Threats to Internal Validity ("How shall I count the ways?")
Review your "Consumer’s
Guide to Evaluating Therapy Studies"
Non-random assignment to treatments
Differential therapist experience
Small n's per treatment
Inconsistent application of treatment elements
Failure to consider maturation effects
Attrition rates in treated group
Etc. etc. etc.
V. What IS the marital outcome variable?
-
First lecture of course, again!
-
Who decides? Who can tell?
-
Review assessment weapons stockpile!
-
What assessment devices do we have to index outcomes?
VI. Assessing Outcome Reliability
A. Box Score Approach
Count the number of studies that show
any improvement
B. Meta-analysis: what it is and isn't
Meta-analysis techniques review the literature
for studies that compare
treatment to a no treatment (a control condition).
The studies are rated
as you did in HW-1; that way quality is controlled.
Instead of counting
studies which show improvement, the idea is
to express numerically
how much improvement there was (relative to
a control standard) using
a comparison statistic that allows us to compare
studies of different
measures and scales. (That way the units are
made equivalent.)
Express change relative to a control group in
standard units:
Effect Size = TX mean – Control mean
S.D. Control
Note: This is the familiar z-score, where we subtract
two mean scores
and divide by a standard deviation. The larger
the result the more
the TX mean differs from the Control mean in terms
of distance from
the control mean. (Also a measure of effect size
ES.)
C. Calculating Effect Size
Based on area under the normal probability
curve, use the formula:
(.50 + Z area ) * 100 = % Better Than
EG. Some Sample Values of ESs (based
on values from a table of the normal
Probability curve):
If ES = Then % Better than=
.40 66%
.50 69%
.60 72%
.70 76%
.80 79%
.90 82%
1.00 84%
1.10 86%
Pre- Post- Comparisons:
Technical Issues
I. Making Pre- Post- Comparisons
A. Therapy effectiveness means change from Pre to Post on some MEASURE
B. Real Change vs. Noise
Depends upon the Reliability of the Measure
C.When is a Measure reliable?
-
When repeated testings yield same result
-
When test is long enough to sample the domain adequately
E.g., 3 items to measure satisfaction
vs 50 items
-
When the Standard Error of measurement is small
D. Standard Deviations (SD's) and Standard Errors (SE's)
-
SD = variability around the mean for a single sample
-
SE = variability of sample of MEANS around the "true" mean;
the SE sets the boundaries for
any given score.
The larger the SE
the more likely the real score falls within a
very wide window. SE is
given by the formula--
SE =
Thus, when N is large SE
gets smaller!
E. All of this is to get to this point: we want to express outcome
success of
individuals. How many people in the treatment
group showed significant
improvement? Improvement over what? Where they started
from;
how much is significant? If their change score is
divided by the SE of the
measure (e.g., DAS) we can determine whether
they changed significantly.
Reliable Change Index = RC = Post - Pre
SE
Note: SE in
this case is the standard error of measurement, which is
estimated by the equation
where r = the reliability of the
test, and SD = the sigma of test mean.
xx
The change is relative to error units! We can now count the percentage
of people who improved
significantly.
F. Clinical vs. Statistical Significance
It is possible to show statistically
significant change based on the R-C score,
but this may be different from
clinically significant change. A person (couple)
may move from a low DAS score
of 75 to 85, pre- to post-therapy. But 85 on
the DAS is still within the distressed
range for that test!
G. Also note that we have the option of reporting individual
spouse scores
or couple scores.
Couple scores would be the avergae of each person's
score. But what to do when
one person is 85 and the other is 110? Is the
average meaningful in this
case since it yields a score of 97.5, which is
almost "normal" although
one person is clearly distressed.
Use of Control Groups and Ethical
Issues
I. Alternatives to Control Groups
A. Treatment on demand (TOD): control over control group
B. Minimal contact agreements
C. Non-specific controls
II. Ethical Issues in Marital Therapy
A. Who is the client?
B. Issues of confidentiality
C. Therapist values vs. client
values
D. Therapist quality assurance
issues