Noninteger frequency weights

[Amy Butler]

If noninteger weights worked for the original analysis, why don't they work for the graphing of the results?

I successfully run the following command:

mixed discrimi c.teenrape##c.wave momsed [fweight = TAwt] || id: wave

I then run a postestimation analysis to get a graph:

margins, at(wave=(1 2 3 4) teenrape=(0 1 )) plot

I get the following error message for the second command: may not use noninteger frequency weights


Is there a way graph my results through STATA?

[Clyde Schechter]

Well, the real question is why they did work with -mixed-. They shouldn't have. Frequency weights, by definition, are positive integers. If you have non-integer weights, then they are not fweights, and treating them as such produces seriously incorrect results. So I think you need to rethink whether your TAwt variable is full of data errors (non-integer values), or, if they are the right numbers, what kind of weights they actually are--not fweights.