weights on fractions

[Kurt Wichtelsacker]

Hello Statalist members,

I got a question about the weight option in the reg command: I got values for the weight variable that are 0>weight<1. If I just use fweight=weight as an option in the reg command STATA gives the error: "may not use noninteger frequency weights".

If I just multiply the weights so that they are >1 and integers, the sd of the coefficions shrinks unreasonably (just due to an artificial increase in observations), I also cant use pweight because this would increase the variance disproportionately.

Any ideas?

Thank you!

Kurt

[Maarten Buis]

They are not fweights, as they are literally the number of persons with those characteristics, so they have to be an integer. Multiplying your weights and using that as an fweight will artificially increase the sample size, so that is definately not what you want.

In order to find out what kind of weights you got you need to look in the documentation of your dataset. My guess would be pweights, but you don't want to base such decisions on a guess. Since we don't have access to the documentation of your dataset, that is pretty much all we can say.

[Kurt Wichtelsacker]

The dataset contains bilateral trade data(dep var) and a treatment (monetary union) and some controls. I did a propensity score matching to match 3 observations from the control group to one treated observation. Now these control observations each get a weight of 0.3333 and the treated of 1. This leads to the problem, that some weights are lower then 1...Do you need any more informations?

Thanks!