| |
 |
|
|
Science Forum Index » Statistics - Math Forum » survey analysis -- ordinal scales
Page 1 of 1
|
| Author |
Message |
| Ming |
 Posted: Thu Jan 25, 2007 11:08 am |
|
|
|
Guest
|
My friend asked for some statistical help on a finance department
analysis questionnaire. There are 3 propositions and several
sub-hypotheses. Your input is highly appreciated.
Proposition 1:
H1a:
Dependent variable: 14 scale questions (scale 1-7) for management
orientation.
Independent variable(s): CEO/Chair duality, a binary variable.
Statistical questions: Their sample size is quite small (~2 , and
initially those scales were ordinal scales, so I think the easy way is
to treat them as interval scales such that we can calculate means and
use non-parametric ways to compare them. Otherwise I have no idea how
to collapse these 14 scale questions statistically to turn them into a
single dependent variable regarding hypotheses H1a and H1b.
H1b:
Dependent variable: same as H1a.
Independent variable(s): 2 scale questions on insider/outsider.
Statistical questions: Same thins as in H1a, and I can then average the
2 independent variables, and it will turn out to be a linear
regression?
Proposition 2:
H2a: to test if there is significant difference in management
orientation between the whole cohort and individual sub-functions
(note: they are not subsets). Each of the sub- functions contains 2
categories: ACTUAL and IDEAL.
Statistical questions: Since I can average the scales of ACTUAL and
IDEAL, a Wilcoxon Rank-Sum Test would be appropriate.
H2b: to test if there is significant difference in management
orientation between every 2 sub-functions.
Statistical questions: Same thing as H2a.
Proposition 3:
H3a - H3d:
Dependent variable: a binary variable.
Independent variable: 14 scale questions as in H1a.
Statistical questions: My friend wanted to use factor analysis to
reduce the number of variables. If so, the obtained factors will then
be put into a multivariate logistic regression. However, if we start
from bivariate logistic regressions and use a cut-off p-value (say 0.2
or 0.3) to remove some initial variables, and run correlation analysis
at the same time to remove collinearity, and subsequently run a
multivariate logistic regression, we will get a
different result from what we get from the first way, because factor
analysis did not remove any variables. What do you think is a
statistically appropriate way for that?
Thanks very much. |
|
|
| Back to top |
|
| |
|
Page 1 of 1
All times are GMT - 5 Hours
The time now is Sun Jul 27, 2008 3:32 am
|
|
