ANOVA is very robust test to study variance. If data is unbalanced, when equal number of observations are there, one need to look for type II and III sub of squares rather than Type I. I dont know which software you are using, but most of the statistical software  give this automatically. For example, in SPSS, if data is unbalanced then ANOVA uses Welch's approximation and hence ANOVA is knows as welch's ANOVA.

About the post-hoc test, different tests are selected based on the whether equal variance is assumed or not assumed.  For example Tukey Kramer test is applied when variances are assumed equal whereas in the case of assumption that variances are unequal, Dunnets's test is selected.

I think you are confused with two tests, Bonferroni correction and Tukey-kramer . Tuckey- Kramer is more poweful test for post hoc multicompariosn than Bonferroni.

For more on that you please follow this link

http://udel.edu/~mcdonald/statmultcomp.html

ANOVA is very robust test to study variance. If data is unbalanced, when equal number of observations are there, one need to look for type II and III sub of squares rather than Type I. I dont know which software you are using, but most of the statistical software  give this automatically. For example, in SPSS, if data is unbalanced then ANOVA uses Welch's approximation and hence ANOVA is knows as welch's ANOVA.

About the post-hoc test, different tests are selected based on the whether equal variance is assumed or not assumed.  For example Tukey Kramer test is applied when variances are assumed equal whereas in the case of assumption that variances are unequal, Dunnets's test is selected.

I think you are confused with two tests, Bonferroni correction and Tukey-kramer . Tuckey- Kramer is more poweful test for post hoc multicompariosn than Bonferroni.

For more on that you please follow this link

http://udel.edu/~mcdonald/statmultcomp.html

Thank you  Mukesh,

1)  I found that tukey kramer is the correction of Tuchey for unbalanced data! i got confused now!

2) I am using Matlab statistics toolbox.

3) no I was asking if i should correct  alpha with bonferroni when analyzing the result from Tukey test or just keep it with 0.05 as the Tuckey test it self  takes care of it?

Again, In matlab i found that ANOVA is adapted for unbalanced data with unequal no. of samples. so i used it to study the global siginficant among groups , then i applied the Bonferroni corrections for its p value.

then to analyze the pairs significant , i used the multicomparison post hoc Tuckey - Kramer as mentioned in Matlab help that this test is corrected for unbalanced data.

so, what is your opinion?

thanks,

Enas

Ms. Enas Ismail

You are doing right.

Dont get confused. Tukey test  is for multicomparison and so is Tukey-Kramer. Only difference is that the former is used for data sets of equal group size whereas the latter one is used for unequal data sets, like yours. Apart from that you are doing good.

And as far as alpha value or significance values are concerned you can change them according to your need.  You are confusing in two different tests done for similar output. Both are multicomparison test but tukey tests are considered more powerful compared to Bonferroni test.

I think there is some sort of misunderstanding about what is written in MATLAB. If ou could share the pdf with marked portion where you have read that, i can try t help you out.

To understand these thing fully, i recommend you to read some statistics book rather than manuals. manuals can tell you how to do things, but not the theoretical part of reason behind why to do that in detail.

You can start with Zar- Biostastistical analysis book

http://www.amazon.com/Biostatistical-Analysis-Books-Carte-Edition/dp/0321656865

Mukesh

Thank you very very much! 

Your reply is very valuable 

enas

Good Luck in your research

The "Bonferroni" terminology can get confusing as there is a Bonferroni test applicable to analyzing differences when making multiple comparisons (i.e., trying to see where a significant difference exists between groups after an ANOVA suggests a difference is present).

What you seem to be describing here is a Bonferroni correction - a method of adjusting the alpha to reduce the chance at statistical error from running multiple inferential statistics. For example, if you were running two separate independent t-tests and wanted to ensure alpha = 0.05, then the Bonferroni correction should be applied such that the adjusted alpha = 0.025. Otherwise, running the multiple statistics increases the chance at an error.

As Mukesh noted, you seem to be doing this correctly here. In general, Tukey's test and its derivatives are preferential to Bonferroni.

With the unequal sample sizes, be sure to check that your data meet the basic assumptions for an ANOVA - namely that the different groups do have equal variance. In my experience, unequal sample sizes sometimes lead to unequal variance. There are different tests for equal variance. Levene's test is a common first choice. Another option would be the Brown-Forsythe test. If neither of those are available, Hartley's test is very easy to compute, but tends to be less robust.

I hope this helps!

dear Jay,

 thanks for your answer!

 I already select corrected ANOVA version for unequal samples number .... it may be Welch method.

Hi,

But I think it is still not clear if bonferroni correction is needed on the results of Tukey Kramer. As I am quite confused over it as well, please share your thoughts please.

Thank you!