There are three groups from different professions. Data is not normal. Since ANOVA comes under the category of parametric statistics, this can not be applied. What to do now?
What kind of data do you have? How is it obtained?
How do the diagnostic residual plots look like?
.
well ... the "automatic pilot" answers "Kruskall-Wallis" but Jochen is right, the first question should be what kind of data, how severe is non-normality, is there a reasonable hope to cure it by transformations etc
.
Hello Rajeev,
The more you can tell us about the experiemnt and the data, the better the chances are that you will get solid advice here.
Just based on what you have said above, and assuming that a parametric method should NOT be used here, a nonparametric ANOVA based on ranks may be a good option for you to use, but equally useful could be a transformation of all data to normal quantiles.
What if a parametric method can be used though?
Do you have count or frequency data? Then start using generalised linear modelling (GLM) with appropriate link function (in the case of counts most likely poisson or binomial)
1.If the sample sizes of the 3 groups are same, then anova is robust for violation of normality assumptions
2. Alternately, transform the variable to make it normal.
or 3. use non-parametric Kruskall-Wallis test.
ANOVA, as I understand, may be possible in this case. In ANOVA, it is not necessary that the observations should follow normal distribution.
However, ANOVA is based on the assumption that the errors associated to the observations are independently and identically distributed random variables following normal distribution with zero expectation and common unknown variance.
you can do a Box-Cox transformation on the data set to attain normality, and proceed with ANOVA on the transformed data-set.
- Badges
- Science topic