Closed awmeade closed 4 years ago
The changes look reasonable to me, though I think there's something unusual with the overall setup of some arguments in this function. What I mean is that, by default in the multipleGroup()
function the lowest level of the group
variable (after it's been changed to a factor) is typically understood to be the reference group (at least, that's how the invariance
arguments work). Hence, the constant identification constraints, theta ~ N(0,1), is fixed during estimation while the other groups are considered as focal groups.
I think I'm going to modify this function a little to remove the ref.group
input for this reason, and always assume the first group is the reference (while also checking whether the mean-variance combo is fixed at 0-1, as this seems to be how many of the fit statistics are setup). Would you have any major objections to this? The function is of course your contribution, so I don't want to distort the logic too much.
Sounds good. Requiring the ref group to be first sounds like a good way to minimize confusion.
Please modify any of this as you see fit! This is your baby and while I'm happy to contribute for the common good, I hope you'll feel at liberty to do with it what you'd like.
Adam
On Wed, Sep 9, 2020 at 10:03 PM Phil Chalmers notifications@github.com wrote:
The changes look reasonable to me, though I think there's something unusual with the overall setup of some arguments in this function. What I mean is that, by default in the multipleGroup() function the lowest level of the group variable (after it's been changed to a factor) is typically understood to be the reference group (at least, that's how the invariance arguments work). Hence, the constant identification constraints, theta ~ N(0,1), is fixed during estimation while the other groups are considered as focal groups.
I think I'm going to modify this function a little to remove the ref.group input for this reason, and always assume the first group is the reference (while also checking whether the mean-variance combo is fixed at 0-1, as this seems to be how many of the fit statistics are setup). Would you have any major objections to this? The function is of course your contribution, so I don't want to distort the logic too much.
— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/philchalmers/mirt/pull/194#issuecomment-689927346, or unsubscribe https://github.com/notifications/unsubscribe-auth/ACA5DEI7WPNDYDKSLBWUAETSFAXW7ANCNFSM4RDGMTUQ .
reversed colors so red = focal, black = reference.
pulling group labels from extract.mirt()