Colleagues:
I'm writing to solicit opinions on a design problem I have inherited for a
piece of true field research.
Our section is doing a study of opinions about organizational climate on
peace-keeping operations in Bosnia-Herzegovina (BH)-our interest is the
conditions under which service personnel deploy, and two types of
deployment:are of interest:
a. an uninterrupted six-month deployment-under this condition, all
individuals arrive at the same time, spend six months in BH together, and
depart from the theatre of operations at the same time; and
b. an interrupted six-month deployment that is completed in a number of
"sessions?" in which people arrive in and depart from BH at various times.
A single opinion survey is administered at three fixed times while in
theatre. Respondents do not identify themselves to preserve confidentiality
of their responses.
Thus, I have something akin to but not identical to a 2 (between
groups--deployment type) X 3 (within groups-questionnaire administration)
split-block design.
Design problem
1. It is by no means a split-block experiment for the following
reasons:
a. We cannot partial out variance due to respondents
because we cannot match questionnaires with respondents;
b. In the interrupted condition, because people move in
and out of theatre on an unpredictable basis, there is no guarantee that the
sample completing the questionnaire at Time 1 is identical to that
completing at Time 2, and that both samples are identical to that completing
at Time 3. In fact, it is virtually certain they are not.
Consequently, while the questionnaire responses (T1, T2, and T3) are
undoubtedly correlated in the uninterrupted deployment condition, I cannot
quantify the correlation because I cannot match up the questionnaires.
And while the degree of correlation in the interrupted condition is
undoubtedly high, it will not be as high as the other sample.
2. Because I cannot match questionnaires, I have no choice other than
to treat this as a 2 X 3 randomized block design when it clearly not one.
And yet I'm aware this is not sound statistical/experimental design
practice.
I'd greatly appreciate comments/recommendations.
We are taking measurements at three times du
LCol Stephen A.T. Eyres, Ph.D
Director of Human Resources Research and Evaluation 4
Operational Effectiveness
Voice: 613 996 7408
FAX: 613 994 2701
[log in to unmask]
|