The reviewer makes a valid point. Reporting the regression analysis instead of the ANOVA might be something to consider as well as it removes the issue of artificially dividing your continuous data into arbitrary groups.

Hi Vladimir,

I think that any research has to have a clear method line. Regression which is an extension of the correlation studies deals with continuous data, ANOVA deals with categorical independent variables. So, if we DO NOT have a clear idea which variable(s) is / are the cause of other variable(s) we normally use correlation, but if we have decided that the relationship between independent  and dependent variables is a cause -effect one I think we should use ANOVA or any of its variants. 

Good luck

Artificially dividing a variable is seldom a good thing. Similarly, conducting a linear regression when the relationship (and your hypothesis) is non-linear can also be problematic.  More details would be useful, BUT if you reported the ANOVA and regression as if they were testing independent hypotheses, then yes, the reviewer is right. If you reported your ANOVA and then gave an R^2 using the same model as your ANOVA (so if it is a oneway ANOVA with k groups, a linear regression with k-1 dummy variables), then it is only one test and you are just reporting an effect size that lots of people like (though for this example eta^2 would work). A rough rule of thumb is that you should not have two p values for the same hypothesis (if you believe in p values). So, a few more details and probably readers can give more concrete advice.

Dear Vladimir, I think your reviewer makes a good point here. Go back to your hypothesis. Check whether your data are distributed normally. If your aim was to demonstrate a relationship between the variables, and the variables are distributed normally, you can better go for a regression analysis.

Hope this helps!

Piotr

With the type of data you describe, my recommendation would be to report specifics using the correlational/regression method. If desired, you might mention your other analysis in a footnote and invite readers to contact you if interested in these details. Good luck!

In any case, if you've done the choosing properly, reporting both gives you no extra information because the values should be exactly the same. The only difference between the two tests is how you code your variables, other than that, significance should be identical so you should only report the rest statistics that match your data type.

Here's a good explanation as to why the tests tell you the exact same information: 

http://www.theanalysisfactor.com/why-anova-and-linear-regression-are-the-same-analysis/

Vlad,

I don't quite understand what you mean by artificially grouping.  Normally I have had very specific reasons for whatever grouping I choose.  For example in many cases I am really interested in either the outliers or those groups of individuals who represent the greatest deviations from the mean.  So I might divide a sample up into 3rds or quartiles and examine the top and bottom.  Median splits typically should not be done if you are attempting to examine differences as it usually masks differences between high and low responders.  So in short, which analysis tells you what  you want to know?

I think it depends on the type of variables you choice the statistical tools to be ANOVA or regression. I think it is not appropriate to report both of them at the same time. Which means either of one is correct analyses or you might need further selection of statistical tool/s/. You may refer how you can select on the attached document.

This following article might help you

Article Behavioural phenotypes in the cuprizone model of central ner...