FITSTAT: Definition of BIC' ?

[Melissa Fiona]

Hello,

I am using the Fitstat command to compare two logistic regression models and am wondering what the different BIC stand for.
Especially BIC', since STATA gives the message that it provides strong support for my current model, which I would like to use.
Unfortunately, I was not able to find any information. If there is already a post for that, please let me know.

This is the output I an referring to:

.

fitstat, saving(m31) using(m3)

Measures of Fit for logit of sup_dem1

                               Current             Saved        Difference
Model:                           logit             logit
N:                                1702              1702                 0
Log-Lik Intercept Only        -947.643          -947.643             0.000
Log-Lik Full Model            -878.049          -874.545            -3.504
D                             1756.098(1661)    1749.090(1648)       7.008(13)
LR                             139.187(40)       146.195(53)         7.008(13)
Prob > LR                        0.000             0.000             0.902
McFadden's R2                    0.073             0.077            -0.004
McFadden's Adj R2                0.030             0.020             0.010
ML (Cox-Snell) R2                0.079             0.082            -0.004
Cragg-Uhler(Nagelkerke) R2       0.117             0.123            -0.006
McKelvey & Zavoina's R2          0.135             0.141            -0.006
Efron's R2                       0.091             0.094            -0.004
Variance of y*                   3.802             3.829            -0.028
Variance of error                3.290             3.290             0.000
Count R2                         0.767             0.769            -0.002
Adj Count R2                     0.048             0.058            -0.010
AIC                              1.080             1.091            -0.011
AIC*n                         1838.098          1857.090           -18.992
BIC                         -10601.010        -10511.304           -89.706
BIC'                           158.395           248.102           -89.706
BIC used by Stata             2061.120          2150.826           -89.706
AIC used by Stata             1838.098          1857.090           -18.992

Difference of   89.706 in BIC' provides very strong support for current model.

Note: p-value for difference in LR is only valid if models are nested.

Thank you!!

[Richard Williams]

There are different formulas for BIC. It doesn't matter which you use so long as you are consistent. For details, see

https://www3.nd.edu/~rwilliam/xsoc73994/L05.pdf

[Richard Williams]

Also, notice that the difference in BIC, 89.706, is the same regardless of which of the 3 BIC measures you use. Personally I like Stata's BIC formula because it doesn't make me go through the mental gymnastics of looking at negative values.

[Melissa Fiona]

Hello Richard,

pardon my late response, thank you so much for your insight!