[From Rick Marken (940924.2050)]
Bill Powers (940924.0829 MDT)--
Rick Marken, Jeff Vancouver --
I'm beginning to get the feeling of watching trench warfare here.
If we're going to make any pronouncements about Bandura's
scientific findings, how about someone summarizing the data? Even
just one experiment.
I don't have any Bandura papers at home, but how about the data from
Kernan, M. C. and Lord, R. G. (1990) Effects of valence, expectancies and
goal-performance discrepancies in single and multiple goal
environments, J. Applied Psychology, 75, 194-203.
There is really only one experiment described in this article, but there
were tons of data collected. The basic experiment seems to be as
follows (the Method section is extraordinarily elaborate and not always
clear):
Subjects performed clerical tasks. Basically, the tasks involved going
through stacks of invoices or requisitions and totalling purchases or
computing stock balances.
The experiment was a 2x2 factoral, completely between subjects with 40
subjects per condition (a total of 160 subjects in the experiment). One
independent variable is called "Goal" and it has two levels; one level is
"single goal" because the subjects were asked to work on only one task
(invoice) on each of the three trials of the experiment; the other level is
"multiple goals" becuase the subjects worked on two tasks (in whatever
way they wished apparently) in the three trial periods. The second
independent variable is called "valence" and it also has two levels; one
level is "high valence" because subjects got a reward for completing
the assigned number of invoices (or requisitions); the other level is
"low valence" because the subjects got nothing for reaching the assigned
goal.
There were a whole collection of dependent variables measured in this
experiment. One of the main dependent variables is "discrepancy"; this
is the difference between the performance on a trial and the goal
set on that trial (the latter, I presume, being the number of invoices
that the subject said s/he would complete). Performance is subtracted
from goal so this number (I presume) is positive if the person's goal
exceeds actual performance and is negative otherwise. Another measure
was "committment"; this is the difference between self-set and
experimenter set goals.
There is more than the usual load of statistical analyses and tables. I
don't know which are considered important so I'll just put up what
appear to be some of the main summary results (abbreviated):
Table 3 Descriptive statistics for single goal condition (so this is
averageing over 40 subjects in each condition):
Valence
Lo High
Variable M Sd M SD
Trial 1 discrepancy 5.88 3.67 6.45 4.14
Trial 2 discrepancy 1.00 2.98 2.43 3.10
Committment -1.47 3.45 -.68 3.40
Trial 3
Committment -.5 2.94 -.18 3.24
I'm willing to bet that all differences between means are significant.
The statistical tests on this particular data were not included in
the article (halleluja -- just about every other possible statistical
analysis was), but with 79 df for each comparison, how can you miss?
The researchers seem to think that the change in the size and variance
of the discrepency and committment scores over trials is important.
The real fireworks seem to be saved for the analysis of the "multiple goal"
condition of the experiment. There is still a "valence" factor to look at
(with 40 subjects in each valence condition). They did a major analysis
with "discrepencies" as an independent variable. It has 2 df and I see
that they split the subjects in the "multiple goal condition" into three
groups depending on whether the discrepency score for one task (the
requisitions) was higher (1), the same (0) or less than (-1)the other task.
The results of an ANOVA with "goal priority" as the dependent variable
("goal priority" being the difference between the subjects goals for the
requisition minus their goal for the other task) is as follows (this is
just for trial 2 but the results are the same for all trials):
Factor df F eta^2
A (Valence) 1 .63
B(Discrepancies) 2 7.88 .17
AxB 2 4.57 .10
The big finding is that factor B and the interaction (AxB) are significant
(p<.05). The graph on the next page shows what the interation is: the
discrepency variable has no effect on goal priority in the low valence
condition but a marked effect on the high valence condition.
Eta^2, by the way, is a measure of the proportion of variance in the
DV accounted for by the IV; it's equivalent to r^2 for ordinal data.
So factor B and the interaction pick up a whopping 27% of the variance
(across subjects, of course) in the goal priority variable.
That's it for now; I've got a headache. I think I'll read a couple
chapters of "Mind Readings" to recover.
Best
Rick