If the homogeneity of regression slopes assumption for ANCOVA and levene’s test was violated (significance ), what is the alternative test for ANCOVA? I have post-test mean scores as dependent, pre-test scores as covariates, one independent variables with two treatment modes, and one moderator variables. Is it Welch’s test suitable for ANCOVA?
It is not an alternative actually but you could run a linear regression and check for assumptions of it.
Or you could make use of a mixed model allowing for non-homogenous variances.
Another one might be going on with ANCOVA despite violations.
It is not an alternative actually but you could run a linear regression and check for assumptions of it.
Or you could make use of a mixed model allowing for non-homogenous variances.
Another one might be going on with ANCOVA despite violations.
Hi,
you can add more "fixed factors" in your model. Think of gender, position, etc. Addition of such variables has an impact on homogeneity test results, since the number of groups to be analysed is growing. Moreover, it is always good to show that you controlled for such factors.
Alternatively you can try structural equation modelling. There are other factors which should be taken into account such as model fit, discriminant and convergent validity, and others. But it can help you to overcome the struggle with homogeneity of variance. If your predictor variables are nominal, then in SEM you can make a group comparison (e.g. Male vs Female). There are some tutorials on youtube.
Faithfully,
Eugene
Thank you Mehmet Sinan Iyisoy & Yevgen Bogodistov for your suggestions, appreciate . I choose Welch's ANOVA test as the LEVENE's test result does not meet the homogeneity of variances of ANCOVA. https://statistics.laerd.com/spss-tutorials/ancova-using-spss-statistics.php
i drop the covariate from the model so not violating the assumptions of ANCOVA and run a one-way ANOVA.
It might not be the ideal solution, but a simple approach could be to calculate adjusted values of the dependent variable (ie adjusted for the covariates) and then use the Welsh ANOVA to test the null hypothesis of equality of adjusted means. Not sure how statistically correct this two-step procedure would be, but intuitively seems a reasonable approximation.
I agree with Mehmet Sinan Iyisoy. You could use linear mixed anova and use a diagonal covariance matrix to allow for different variances between groups.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Probability
- Probability Theory