Howdy, Stranger!

It looks like you're new here. If you want to get involved, click one of these buttons!

Supported by

bayesian t-test for beta?

MSBMSB
edited May 2016 in JASP & BayesFactor

Hi all (again),

I was wondering - I know that in JASP regression is shown by comparing models, but I was wondering if it would make sense to compute BFs on individual betas by using their t-statistic?
What I want to show is that adding parameter X2 to the model has no affect on the beta X1.

Thanks again,
M

Comments

  • EJEJ
    edited 6:34PM

    Hi M,

    The first point is that individual betas only have meaning in the context of a specific model. You can pick the full model, but that's just one (arbitrary?) choice. With respect to your specific question, one easy way would be to plot the posterior for beta X1 with and without including also X2 (you would still have to pick a specific model of course). However, JASP does not plot these posterior distributions right now. It is high on the to-do list. I do believe the BayesFactor package will let you do this though.

    Cheers,
    E.J.

  • MSBMSB
    edited 6:34PM

    Hi EJ,

    Thanks for your quick response.

    I'm not sure I understand what you mean by comparing posterior.

    I find that X1's beta is 0.5~, and i want to show that not only is this model better than one containing also X2, but also that X1's beta does not change (and it is still around 0.5~).
    Computing BFs for t's will not allow me to do this very specific thing?

    Thanks again (again)!

  • EJEJ
    edited 6:34PM

    Hi M,

    In JASP you can compare models to assess whether inclusion of beta for X2 helps or hurts. However, in case of regression and ANOVA, JASP does not (yet!) return the parameter estimates.

    Cheers,
    E.J.

Sign In or Register to comment.