I have 4 independent and 2 dependent variables and chose to conduct MANOVA, however, the dependent variables are not normally distributed and the DVs are highly co-related. Because of this, the test MANOVA cannot be undertaken.
Can I do two, 2 way ANOVA or is there any other statistical test that can be conducted?
My four independent variables are the ratio of the mix, pH, heating temperature and heating time
The dependent variable is the Emulsion activity index (EAI) and emulsion stability index (ESI).
I am trying to check the mixture of pea and whey protein isolates at different ratios and its effect on the EAI and ESI
Hello Anjali,
The best answer to the question: go univariate or go multivariate, depends on how you intend to approach and apply the results. Methods such as manova, discriminant analysis, canonical correlation, multivariate regression, and SEM models can all involve 1 or more IVs with 2 or more DVs, but people often decide it's easier to interpret univariate analogs. So the reality is that, while the multivariate perspective is more realistic (especially when you have theory or evidence supporting that DVs are not independent), folks often can't wait to abandon the multivariate universe and drop back to univariate models as "follow ups" to a significant multivariate effect.
If you have (and wish to evaluate the influence of) four IVs on a DV, I'm not clear as to how a two-way anova would be helpful (maybe the "2" was a typo?).
The normality assumption applies to the model residuals, not the raw scores on the DV. So, fit the model, save the residuals, and examine them.
Here are some options, if normality still proves to be problematic:
1. Robust regression
2. Bootstrap/resampling methods (either for the 4-way anova or the analogous regression model)
3. If it makes sense to do so, you could try transforming the DV scores, though your interpretation of parameter estimates (such as regression coefficients) have to be interpreted in terms of a new metric.
Good luck with your work.
Are the two dependent variables correlated?
If not using MANOVA does not require stringent requirements of normal distribution just check for skewness if it's between -1 and + 1 go ahead with MANOVA.
If correlated strongly, linear regression considers dependent variables separately on Independent factors.
- Badges
- Science topic
- Similar topics
- Social Science
- Media
- Journalism