Hi
To find general information about SPSS or post a query, you can
subscribe to a newsgroup called "comp.soft-sys.stat.spss"
assuming that you have access to a news feed. I have a bunch of
statistics links on my homepage, pointing to some good SPSS
sources. Try "http://www.uwinnipeg.ca/~clark/stats.html".
On Tue, 9 Mar 1999, Stephen Black wrote:
> I'm supervising a student who is collecting sequential data over
> days. There is a baseline period, followed by the application of a
> particular experimental condition. She would like to compare the data
> collected during baseline with the data collected during treatment
> (for example, seven days of baseline followed by seven days of
> treatment). Initially she would like to examine the data one subject
> at a time.
>
> She could do this, I suppose, with an independent t-test for baseline
> data compared with treatment data. But the data are in the form of a
> time series, and I recall that something called trend analysis may be
> appropriate.
She could do a _paired-difference_ (i.e., _dependent_) t-test
comparing the mean performance over the baseline 7 days with the
mean performance over the 7 treatment days. This would collapse
over the days variable. To include both the days and condition
variable, she could do a repeated measures ANOVA. In SPSS
syntax, this would be:
manova b1 b2 b3 b4 b5 b6 b7 t1 t2 t3 t4 t5 t6 t7
/wsf cond(2) days(7)
The manova output would include a main effect for Condition, a
main effect for Days, and an Interaction term. Having 6 degrees
of freedom each, the Days effect and the Interaction could be
partitioned into various components depending on the pattern
expected for days (e.g., polynomial effects).
At the individual subject level, I think this is properly a time
series analysis, for which there are special procedures
(including some in the Advanced SPSS). I'm not that familiar
with them. For a student, I would suggest perhaps just plotting
the individual data across the 14 days with a break between the
baseline and treatment weeks. Plot best-fit lines separately to
the baseline and treatment and see to what extent their
intercepts and slopes differ. In essence, look for a
discontinuity that suggests a treatment effect. Campbell did
something like this in his old (but still excellent) article on
quasi-experimental designs for evaluation of real-world
manipulations.
Best wishes
Jim
============================================================================
James M. Clark (204) 786-9313
Department of Psychology (204) 774-4134 Fax
University of Winnipeg 4L02A
Winnipeg, Manitoba R3B 2E9 [EMAIL PROTECTED]
CANADA http://www.uwinnipeg.ca/~clark
============================================================================
