Do two variables influence each other?

[Florian Berger]

Greetings,

I have two variables. The first shows the happiness of a person on a scale between 0 and 10. The second one shows how satisfied a person with his/her health is. It's also on a scale from 0 to 10.

I ran a regression and the health of a person has an significant effect on happiness, obviously.

Now I wonder if this correlation can be even used in a regression, because happiness could also have an influence on health, right?

So I ran the following command

 tab happy health, chi

and got this result

Pearson chi2(110) =  1.8e+03   Pr = 0.000

As fas as I know this means that there is a correlation between the two, but does this result mean, that the two influence each other or are influenced simultaneously by another variable? If not which command can I use to get an answear for this question?

Some help would be really aprreciated. Thank you

[Clyde Schechter]

First, the chi square test associated with the -tab- command treats both of your scales as discrete variables. It does not imply that there is an orderly correlation between the two. It could be, for example, that people who are extremely high and people who are extremely low on one of the scales tend to be very high (or very low) on the other and people in between on the first scale are opposite. That wouldn't be a correlation.

As fas as I know this means that there is a correlation between the two, but does this result mean, that the two influence each other or are influenced simultaneously by another variable? If not which command can I use to get an answer for this question?

This cannot be done statistically with any commands at all. You have observational data that do not even tell you which measure preceded the other in time. But even if you knew that, you could not using anything from the data discern which direction is causal, or whether both are caused by something else. And then there are methodological issues such as the extent to which the questions in the two instruments might cue each other based on wording, etc.

To definitively answer a causal question you need data from a causal study. That would be nearly impossible to do here. You could not randomly assign people to low or high levels of happiness, nor of health. Probably the closest you could come to this would be to have a large longitudinal data set in which some people experienced abrupt changes in either health or happiness and then you could see what happens to the other measure over time thereafter and compare that to people who had no abrupt changes in health or happiness. Even this would be imperfect and would not completely exclude a common cause acting with a time lag. Also the effects of being healthy (resp. unhappy) may be different from the effects of becoming healthy (resp. unhappy). In short, I think the best you could hope for is a tentative answer to this question, and even that would require a much more complex study design.

[Florian Berger]

Allright, thanks a lot