I have two cases control groups (clinical (42) and healthy (42)), where I am measuring several continuous variables in each group. I have two categorical confounding variables (family history (0,1), gender (0,1)). When I ran MANCOVA, Box's test of equality of covariance is not significant. However, in Levene's test, one variable is significant, rest are insignificant. Can I continue to use MANCOVA? or are there other tests? Also, please note that when I ran the test, I entered group in 'Fixed factors' section, continuous variables in 'dependent variables' section and confounding variables in 'covariates' section. I am also doing a binomial logistic regression analysis post this test to find predictor. Am I doing this right? This is for my Phd Thesis. Thank you in advance.
Hello Sahithya,
As your data set indicates, it is possible for the multivariate homogeneity check (Box's M) to yield a non-significant result whereas a univariate homogeneity check (here, Levene's) is significant (for at least one variable). This apparent discrepancy can occur because: (a) the two tests define comparability differently (e.g., Box's M looks at more information); and/or (b) the sample size is not very large (in general, multivariate tests tend to be less powerful than univariate tests, though they can be sensitive to more types of differences).
For me, I'd stick with the multivariate checks (for normality of residuals as well); if data are satisfactory with respect to these checks, then continue with the manova as planned.
Good luck with your work.
Gust use mancova . Don't use covariates no needed here .compare your means with Duncan , actually that if you want to know the interaction if not you can use Anova for every factor alone and compare with LSD0.05.
Hi Sahithya,
I'd suggest to define your confounding variables as fixed factors as these are categorical. (SPSS expects covariates as continuous variables.) By default you will then get a model with interactions between the three fixed factors. If you want a model without interactions, you would have to build it accordingly instead of using the default.
I do not understand the purpose of doing a binomial logistic regression analysis afterwards.
- Badges
- Science topic
- Similar topics
- Biological Science
- Histology
- Sectioning