Scale and Location to TDist in Distributions

Is there any reason not to have a scale and location parameter in TDist in the Distributions package? I was trying to do a MLE with a Generalized T distribution when I realized it wasn’t available out of the box.

Not hard to code up on my own, just curious if it wasn’t in there for a reason.

4 Likes

I believe the idea is that location-scale distributions are better defined using the primitive standardized distribution and the LocationScale function. Like this:

TDist(μ, σ, ν) = LocationScale(μ, σ, TDist(ν))

5 Likes

Thanks. I had not actually seen that. I will give it a shot.

2 Likes

In my opinion this should definitely be part of the distributons package. I spent a good hour searching through the documentation and didn’t figure this out. :joy:

5 Likes

After seeing this post, I went on to make this PR to improve documentation.

And it turns out that one should use * to scale and + to shift distributions - LocationScale is on it’s way out. Follow the effort to create good documentation over at the PR.

2 Likes

Sorry if I’m being dim. But how do we then define TDist in terms of it’s location and scale? We still need to supply 3 parameters here.

Since LocationScale is now deprecated, I now use:

TDist(μ, σ, ν) = μ + TDist(ν)*σ

Does this answer your question?

3 Likes

Thainks for the quick reply! Yes, that’s helpful thanks.

2 Likes