I have posted a blog item about the seven new Bayesian procedures in SPSS Statistics 25 here for those who are interested. |
Dear Jon
that is really helpful
have some additional potentially useful suggestions and would like opinion of SPSS community
GeneraliZed linear models provides BIC.
So why not use it for Bayes FACTOR?
Run both with and without the tested predictor and take the ratio (BIC with predictor)/(BICwithout predictor).
this seems to me simple and direct. Where is the flaw?
Meanwhile, I note that GeneraliZed linear MIXED gives F-tests for predictor BUT GeneraliZed linear gives chi-square tests [Wald or likelihood].
Why is that? completely mystified.
Both provide BIC, values close but not identical
best
Diana
On 18 Aug 2017, at 21:49, Jon Peck <[hidden email]> wrote:
=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD
________________________________________
Professor Diana Kornbrot Work University of Hertfordshire College Lane, Hatfield, Hertfordshire AL10 9AB, UK +44 (0) 170 728 4626 [hidden email] http://dianakornbrot.wordpress.com/ http://go.herts.ac.uk/Diana_Kornbrot skype: kornbrotme Home 19 Elmhurst Avenue London N2 0LT, UK +44 (0) 208 444 2081 |
I don't remember if I ever used BIC or AIC in anything published, so I'm not the best
possible reference here. But I don't see anybody's response.
There are two formulas for BIC - one for maximum likelihood solutions and the other for ANOVA (least squares). When the different criteria don't reduce to the same thing,
the solutions will be slightly different. I'm not sure what you are looking at, but it is certainly possible to compute BIC in both ways on one set of data.
As I understand it, and the Wiki page seems to support it, BICs are compared by their difference, not the ratio. I think of it as approximating a 1 d.f. chi-squared.
-- Rich Ulrich From: SPSSX(r) Discussion <[hidden email]> on behalf of Kornbrot, Diana <[hidden email]>
Sent: Saturday, August 26, 2017 12:16:24 PM To: [hidden email] Subject: Bates procedures Dear Jon
=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD
that is really helpful
have some additional potentially useful suggestions and would like opinion of SPSS community
GeneraliZed linear models provides BIC.
So why not use it for Bayes FACTOR?
Run both with and without the tested predictor and take the ratio (BIC with predictor)/(BICwithout predictor).
this seems to me simple and direct. Where is the flaw?
Meanwhile, I note that GeneraliZed linear MIXED gives F-tests for predictor BUT GeneraliZed linear gives chi-square tests [Wald or likelihood].
Why is that? completely mystified.
Both provide BIC, values close but not identical
best
Diana
On 18 Aug 2017, at 21:49, Jon Peck <[hidden email]> wrote:
===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD
________________________________________
Professor Diana Kornbrot Work University of Hertfordshire College Lane, Hatfield, Hertfordshire AL10 9AB, UK +44 (0) 170 728 4626 [hidden email] http://dianakornbrot.wordpress.com/ http://go.herts.ac.uk/Diana_Kornbrot skype: kornbrotme Home 19 Elmhurst Avenue London N2 0LT, UK +44 (0) 208 444 2081 |
Free forum by Nabble | Edit this page |