Let's say there are two groups of data A and B, each uses a different technique. In group A, there are independent variables X (numerical), for each x, the associated technique produces corresponding value y (numerical). Similar for group B.

How do I test for statistically significant difference between two techniques? If there is only one value in each group then I can use ANOVA, but now each group has (X, y) so I have no idea what to do.

I am thinking of fitting a linear regression using OLS for each group then test for significant difference of respective coefficients between two regressions. But I am not sure this approach is correct.

I am a beginner in Statistics, I don't know what keyword to search for so I ask here. I am happy and willing to learn new things.

I am also wondering for the more generalized version, if X and Y are categorical variables, or X is numerical and Y is categorical, ... then how the above problem can be solved?


1 Answer 1


Assuming your experiment is designed correctly and there are no confounding factors, your suggestion to analyze the data via OLS is fine.

A model

$$ y = \beta_0 + \beta_1 x + \beta_2I(\mbox{group}=B)$$

should suffice assuming you posit no interaction between group and $x$. The test for $\beta_2$ should tell you if, conditioned on $x$, the difference in groups is consistent with a null effect or not.

If $y$ is categorical, you can perform a multinomial regression or ordinal regression (depending on if $y$ is an ordinal variable or not) where things are slightly different but similar. In this case, I would refer you to Agresti's Categorical Data Analysis for more.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.