Different BIC scores in HBSMS and BFF?

[Gunnar Björnsson]

Calculating BIC for the very same model based on the very same data set in HBSMS on Tetrad 5.0.0 and BFF on Tetrad 5.2.1 yields quite different scores. (Chi Square and P Value almost identical, but BIC scores difference is huge (≈20).) Has there been a change in how BIC is calculated? If so, what?

Best,

Gunnar 

[Joseph Ramsey]

You're right--I hadn't noticed. The score in the current Tetrad is right. The formula should be:

chisq - p ln n

where p is the degrees of freedom. I got it wrong before in places and used

chisq + p ln n

But as Peter Spirtes point out to me, the former is correct.

Joe

--
You received this message because you are subscribed to the Google Groups "Tetrad Users Group" group.
To unsubscribe from this group and stop receiving emails from it, send an email to tetrad-users-gr...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

--

Joseph D. Ramsey
Special Faculty and Director of Research Computing
Department of Philosophy
143 Baker Hall
Carnegie Mellon University
Pittsburgh, PA 15213

jsph....@gmail.com
Office: (412) 268-8063
http://www.andrew.cmu.edu/user/jdramsey

[Gunnar Björnsson]

Thanks, that's very good to know!

By the way, it looks as if BFF has replaced HBSMS. I have failed to find any discussion of this, or of BFF. Is there anything available somewhere?

Best,

Gunnar

On Saturday, 20 June 2015 21:17:57 UTC+2, Joe wrote:

You're right--I hadn't noticed. The score in the current Tetrad is right. The formula should be:

chisq - p ln n

where p is the degrees of freedom. I got it wrong before in places and used

chisq + p ln n

But as Peter Spirtes point out to me, the former is correct.

Joe

On Sat, Jun 20, 2015 at 12:26 PM, Gunnar Björnsson <gom...@gmail.com> wrote:

Calculating BIC for the very same model based on the very same data set in HBSMS on Tetrad 5.0.0 and BFF on Tetrad 5.2.1 yields quite different scores. (Chi Square and P Value almost identical, but BIC scores difference is huge (≈20).) Has there been a change in how BIC is calculated? If so, what?

Best,

Gunnar 

-- 

[Joseph Ramsey]

Actually, it was just a name change. Clark suggested it. I thought it was OK since he had suggested the original name, which no one could remember.

Joe

--

You received this message because you are subscribed to the Google Groups "Tetrad Users Group" group.
To unsubscribe from this group and stop receiving emails from it, send an email to tetrad-users-gr...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Gunnar Björnsson]

I see. But the functionality seems to be tweaked too: I need to set the alphas differently to get similar search results as I got with HBSMS in Tetrad 5.0.0. Since the way BIC is calculated was changed, maybe something else changed too?

Gunnar

You received this message because you are subscribed to a topic in the Google Groups "Tetrad Users Group" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/tetrad-users-group/L4HQRRWn4lU/unsubscribe.
To unsubscribe from this group and all its topics, send an email to tetrad-users-gr...@googlegroups.com.