# How many correlations are too many?

Following my earlier posts here, here and here, I understand the dangers of 'fishing' for a significant correlation. I also understand the 'problem' of running several correlations on the same data set.

However, I am testing the relationship between variables A and B by several demographic characteristics, so I cannot help but run several correlation tests. For example, on gender I have 2 correlation tests (one for male and one for female), on age I have 3 correlation tests (one for less than 18 years, one for 18 to 44 years and one for over 45 years), on marital status, I have 3 correlation tests (one for married, one for never married and one for other).

With all the concerns expressed in the earlier posts, is there anything I am missing on the use of Pearson r?

Remember, I am just focusing on the relationship between A and B by demographic characteristics of the respondents. Is there an alternative way to test this relationship?

If one variable (A or B) is "dependent" and the other is "independent" then you could use regression with all the demographic variables in the equation as well, and possibly interactions between the main independent variable and the demographic variables.

The former will control for the effects of the the demographic variables; the latter will, in addition, look for differences in the relationship between A and B at different levels of the demographic variables.

• It's difficult to construe any pair of "demographic characteristics" as constituting one IV and one DV. Usually they have to be treated as correlated random variables.
– whuber
Oct 18, 2011 at 14:10
• Hi @whuber As I understand it, A and B are not demographic variables. But it would help if adhesh told us what they were Oct 18, 2011 at 18:03
• Ah, I see. There's some vagueness here.
– whuber
Oct 18, 2011 at 18:49
• @PeterFlom Thanks! Variable A is the number of times a respondent has dinner outside of the home (i.e. at an eatery that is not a home) and Variable B is total price he or she pays for the food. I know this sounds simple and perhaps obvious but please bear with me - I am a market researcher! Oct 18, 2011 at 22:49
• @AdheshJosh What are you trying to find out? Why are you doing these analyses? It sounds to me like B would be the DV, A the main IV and the demographics would be covariates. Then, depending on how big a sample you've got, you could look at multiple variables at once Oct 19, 2011 at 10:21