This is what multiple regression is for. What you are describing is an interaction; you can test interactions between variables using multiple regression.

This is what multiple regression is for. What you are describing is an interaction; you can test interactions between variables using multiple regression.

Hey Matteo

There are msny ways to do this. If you are using regression, partial regression and analyzing partial correlation and etc can help. Also you can use gender as dummy variable in your partial regression.

Regards

I think the new methods in data mining maybe helpful such as ANN, SVM or RBF

Thank you everyone. Supposing that there is some correlation between some of the tested variables, how can I estimate the impact on the overall relationship?

For instance:

Testing income for multiple regression between gender and nationality, and supposing that both have an influence, but not the same weight on the relationship, which is the best way to weigh the impact of both on income? I am thinking of Spearman/Pearson correlation.

Any other advice?

Thank you.

It's the regression coefficients.

Add in the variables that believe may be correlated with the variables of interest. If any of these change the coefficient value for your variable of interest, then you have evidence for an interaction that should be included in the final analysis. Failure to do so biased your estimate for the variable of interest since without controlling for these variables you cannot know the proportion of variance that it alone contributes to your regression equation. Leave out any of the correlated variables only if they have no effect on the main variable. Loosen the p-value for those correlated variables making it p≤0.10 instead of p≤0.05. You do this not because the correlated variable are themselves significantly different but because without them you cannot get an unbiased estimate of the effect on your main variable without them. This is particularly true, as I have suggested, for case where the main variable coefficient is changed by the addition of the other variables.

Patrice Showers Corneli

M.Statisitics

PhD. Evolutionary Biology

Multiple regression & multiple correlation

with

Partial regression & partial correlations

can be valid analysis in achieving the desired objective of the study .

This is clearly a regression analysis. Simply add gender and age as explanatory variables to your original regression and be sure to check for the significance of two and three-way interactions among the explanatory variables. Let me know if this puzzles you.

Dear Ankit Patel,

Pearson correlation coefficient (r) is a measure of linear correlation between two correlated variables.

In the problem, as mentioned, the objective is to assess / know if the correlation between two variables is influenced by other variables (for instance gender or age).

This is clearly a case of the study on the assessment of the effect of some other variables on the correlation between two variables.

Thus, this is a case of application of analysis of canonical correlation and canonical regression.

Pearson correlation coefficient (r) is not an option here.

I agree with Patrice Showers Corneli that this is clearly a problem of regression analysis.

However, I like to add that this is not a case of simple regression analysis.

This is a case of multiple regression analysis.

Analysis of canonical correlation/regression suits this problem.

Dhritikesh Chakrabarty Thanks a lot for your kind correction.

Dear Ankit Patel,

Welcome your acknowledgement.

Research is a continuous process of searching for truth.

One of the many ways of searching for truth is trial-error process.

Any researcher may be wrong during the progress in researcher process.

Right can only be achieved through wrong or through a series of wrongs.

So, don't be worry. carry on your research activity.

Truth is subject of philosophy, not of (empirical) research / natural science.