How to analyse repeated measures on two continuous variables

[Rena Jenkins]

Hi Stataforumers

A friend has asked me to help analyze her dataset. She has measurements on two continuous variables at four points in time (on about 100 subjects). She is interested in the change in the relationship over time. She proposed to work out the correlation between the two variables at each of the four time points. Is there anything more sophisticated she could do?

Thankyou

Rena

[Charlie Joyez]

Are all the 100 subjects the same for the 4 time period?
If they are, (even partially) you have panel data, if not, you don't, wich will change much your analysis.

We can't such much more, as we don't know the relationship your friend want to test. Is there any causality links she want to test (the change in var1 influence the change in var2)? Does she want to control the relationship for third variables effect? Testing the correlation is fine, but not very convincing; it all depends on the subject, and the other data available.

[daniel klein]

The description is too vague to give good advice. What is "the change in the relationship over time"? Is there a hypothesis, in the sense that one variable, say x, influences the other, say y? If so a model like

xtreg y x ,fe

could be a start. If the strength of this association is assumed to vary over time, then

xtreg y c.x##i.time ,fe

might be a next step.

Anyway, you probably want to have a look into xt.

Best
Daniel


Note: the above assumes panel data, i.e. repeated measures on the same units.

[Rena Jenkins]

Hi Charlie

Thankyou for replying. Yes she has exactly the same subjects for all four time periods.

I don't think that she is sure what she wants to test either! She does have some covariates that she might possibly want to adjust for. I think that whether the change in var1 influences the change in var2 might be a good question. There is a presumed direction of causality in that var1 might be assumed to influence var2 but not vice versa.

Thankyou again

Rena

[Rena Jenkins]

Thankyou Daniel--that's very helpful. Will look into xt commands and xtreg in particular.

[Charlie Joyez]

Actually the best solution would be to advice your friend to post directly on Statalist, she might better know what she wants to do.

So you have panel data, as Daniel sugested, you (she) should use -xtreg- for a regression (after defining the panel with the xtset command), to test the influence of x over y.

For more details, come back with preciser analysis angle, or if you want some idea, check panel data topics on this forum or on the web, you'll see what is possible.