Specifying nu for StudentT

Hello. I’m running a regression model using and using a StudentT distribution for the first time. That said, am I right in assuming the value for nu should be the len(observed Y - 1)?

1 Like

There is no restriction, you can set it to any positive value or random variable. In practice it shapes the tail of the distribution, so it dependence on how heavy tail you want your distribution to be.

Thank you @junpenglao

Ok. Any best practices? I thought it represented the degrees of freedom. Is that right?

It does, but the usage is different than when you are using it for t-test and getting p values.
Best practice is to set it to a small value, after nu > 30 you get something quite close to Gaussian

1 Like

Alright then. Thank you!