Is using ANOVA considered right for group to group in this case?

[ohaneya]

Hello.
I am really happy to find the right forum which can help me.

I had divided a number of 671 children into three groups according to the style of their parents’ treatment, which are (Democratic, permissive, authoritarian), I want to test if level of education of the parents affect the style of parents treatment of their children. But I don’t know the right tool which I should use.
I had used ANOVA and post hoc, but the results of post hoc is not going right with ANOVA.


Thank you
Haneya

[CamilleJosion]

Hi, you should use ANOVA and the corresponding tests only if your data meet the assumptions of ANOVA, the most important being that your residuals are normally distributed, with 0 mean and identical variance. You can check these assumptions using a normality test and a variance test. If you are afraid of not meeting the normality assumptions, you can use the Kruskal Wallis test that also allows post hoc test for multiple comparisons. For the KW test, I recommend StatXact or XLSTAT as they have exact and Monte Carlo based p-values. For the post hoc test R and XLSTAT are probably the most complete.
Camille

[ohaneya]

Dear Camille
Thank you for your replay.
After a little study I found that ANOVA don't works with grouped data set. Both of (parents’ treatment) and (level of education) are grouped variables.

[CowboyBear]

After a little study I found that ANOVA don't works with grouped data set. Both of (parents’ treatment) and (level of education) are grouped variables.

Yep, a chi-square test or maybe Fisher's exact test could be the way to go. If you wanted to incorporate additional control variables you could also look at multinomial logistic regression.

[ohaneya]

Yep, a chi-square test or maybe Fisher's exact test could be the way to go. If you wanted to incorporate additional control variables you could also look at multinomial logistic regression.

100%. I wanted to write it but you were faster than me .

chi-square test is the right test, or Fisher if one of the groups has a frequency less than 5.

Thank you