My full question is, would you report a 'correlation table' before a multiple hierarchical regression, if you have dummy variables? If so, would you report ONLY the continuous variables in the correlations table, or would you ALSO report the primary variables in their original consolidated form BEFORE they became dummy variables for the multiple hierarchical regresson?
Any advice would be greatly appreciated.
I totally endorse the answers provided by Professor Thom S. Baguley, and Mathew Kerry. Professor David Eugene Booth was right too, but it may be better to err on the side of caution by adding the correlations. You can remove it if the reviewer or editor recommends it. I have experienced some situations where a reviewer's general recommendation concerning an article is also influenced by the lack of correlation results (Table) in the results section. Some reviewers are very interested in seeing the Table and you need not provide an argument that the Table is unneccessary apriori.
On the inclusion of dummy varibles, it is preferable to run correlations with only dichotomous variables that are in two categories (e.g., gender coded 0 and 1, or 1 and 2) and the continuous variables. When the categories exceed two, you will have difficulty in interpreting the results (e.g., occupation). There may be further expert advice on this issue. Thank you.
In general, it's helpful for the reader to report the simple correlations either in a table or a combination plot that includes all the relevant information. I'm not sure about the issue of dummy variables.
It is generally helpful to see the simple correlations - including for categorical variables. Correlations between predictors produce multicollinearity - so these can be helpful in understanding why multicollinearity is violated (for instance).
For me it depends on whether or not the argument I am making depends on correlations. If it only depends on regression results I wouldn't clutter the paper. Give the reader the information s/he needs to follow your argument. Don't risk confusion unless the editor demands the extra information. Best, David Booth
It is useful to have the full matrix if you don't provide raw data. You reproduce the regression output from the correlation matrix. This is why some journals require it.
I totally endorse the answers provided by Professor Thom S. Baguley, and Mathew Kerry. Professor David Eugene Booth was right too, but it may be better to err on the side of caution by adding the correlations. You can remove it if the reviewer or editor recommends it. I have experienced some situations where a reviewer's general recommendation concerning an article is also influenced by the lack of correlation results (Table) in the results section. Some reviewers are very interested in seeing the Table and you need not provide an argument that the Table is unneccessary apriori.
On the inclusion of dummy varibles, it is preferable to run correlations with only dichotomous variables that are in two categories (e.g., gender coded 0 and 1, or 1 and 2) and the continuous variables. When the categories exceed two, you will have difficulty in interpreting the results (e.g., occupation). There may be further expert advice on this issue. Thank you.
Thank you so, so much to everyone for their input thus far, it has helped to no end!
I decided to include a correlation table but to add to the legend below it, as follows...
“*p < .05, **p < .01, p < .001
‘—‘ = dummy variable“
So, I have only shown the values that, in my eyes, have meaning, in relation to showing no issues of multicoliniarity, but acknowledged that the dummy variables do still exist! Can I ask your thoughts on this? I have discussed assumptions fully and after this table, including multicoliniarity with dummy variables.
Oh! And I should mention that my dummy variables are primarily being used as control variables and have relatively low VIF values, so am maintaining the assumption that there are no violations of the multicoliniarity assumption. :)
Yes Alistair Page. It is also important for you to indicate the coding of the categorical variables as a legend under the correlation Table. It will enable the reader to understand the correlations. Best regards.
JohnBosco Chika Chukwuorji
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Regression