re: standard least square regression question

[Jonathan David]

Hi, but I am trying to run a standard least square regression with a REML method, but I want to run it with a random factor.

I am not sure how to do this on stata, as the only method that would allow me to do this is to use the multilevel mixed-effects linear regression command, as this allows me to enter a random factor.

The problem with this is that it only gives me the Z score, and I need t score.

Can anyone help me?

[ben earnhart]

Well, if you have a random factor with REML, it's not OLS. Yes, you would need the multilevel mixed-effects approach. Any strong reason you need T instead of Z? With any decent sample size (which you would need for the multilevel approach) Z is equivalent to T, out to several digits.

[Jonathan David]

Essentially what I want to do is use a Standard Least Square Regression, with an REML. So the DV would be on of the four activities (foraging, resting, social, or moving). The IV would be calendar month and group they belonged to, I would also like to do an interaction between month and group. Additionally I want to use the individual ID as a random factor.

[Jonathan David]

Honestly i am just replicating someone's data and they used T...

[ben earnhart]

Hmmm... messier than I thought. You don't even want an OLS equivalent, you need a multinomial-logit equivalent. I don't know or care what the previous people did, but outcomes "foraging, resting, social, or moving" are not conceivably linear. "mlogit" would be the immediate choice, but since you have repeated observations, and groups, then it gets complicated and I hesitate to give advice. The more information you can provide about your data structure, the better.

[Jonathan David]

surely a xtmixed would be a good fit also? and as for the Z score, i mean if its equivalent, than so be it

[Maarten Buis]

I suspect that the person you want to replicate made a mistake. These statistics are computed in exactly the same way (coeffiicient divided by the standard error), so it is easy to see how such a mistake can happen. However, you dependent/left-hand-side/y variable (foraging, resting, social, or moving) is definately not suited for xtmixed. You could think of a random effects multinomial logit model, see: http://www.stata.com/manuals13/semexample41g.pdf

[Clyde Schechter]

I sense some confusion about what the dependent variable(s) in this problem are. Maarten Buis has understood that there is a single variable that takes on one of four values (foraging, resting, social, or moving). But jonathan... (please reregister under your real name) may have in mind four separate regressions, each having as its dependent variable the amount of time spent in one of those four activities. If the latter, then, yes, -mixed- will be suitable.

[Jonathan David]

Thank you Clyde, yes they are treated as separate regressions. How do I go about changing my username to my full name.

[Clyde Schechter]

You can change your user name by clicking on the Contact Us button and requesting the people who run this forum to change it.