estat and fitstat show different BIC

[Neo Zhou]

I want to get the AIC and BIC after fitting a model, but recently I found that estat and fitstat show very different results for BIC, see the log file below:

sysuse auto.dta, clear
qui reg price weight length
fitstat

qui reg price weight length
estat ic

AIC is identical, but BIC is different. Which one should I use for AIC and BIC? Thanks.

[Leonardo Guizzetti]

I can't reproduce the problem you have. Please note that -fitstat- is not a part of Stata, and you are asked in the FAQ to identify its source. Fitstat is a user contributed command by Long and Freese (package -fitstat-), but is also included in their -spost13_ado- package.

You may wish to check that you have the most updated package (see -adoupdate-).

These are the results I get and BIC matches:

. which fitstat
c:\Stata16\ado\plus\f\fitstat.ado
*! 4.1.10 | 2017-01-05 | freese long | fixed bug in tobit (for v <14 and 15+)

. sysuse auto.dta, clear
(1978 Automobile Data)

. qui reg price weight length

. fitstat

                         |     regress
-------------------------+-------------
Log-likelihood           |            
                   Model |    -679.912
          Intercept-only |    -695.713
-------------------------+-------------
Chi-square               |            
         Deviance(df=71) |    1359.825
-------------------------+-------------
R2                       |            
                      R2 |       0.348
             Adjusted R2 |       0.329
                McFadden |       0.023
      McFadden(adjusted) |       0.018
            Cox-Snell/ML |       0.348
  Cragg-Uhler/Nagelkerke |       0.348
-------------------------+-------------
IC                       |            
                     AIC |    1365.825
        AIC divided by N |      18.457
               BIC(df=3) |    1372.737

. estat ic

Akaike's information criterion and Bayesian information criterion

-----------------------------------------------------------------------------
       Model |          N   ll(null)  ll(model)      df        AIC        BIC
-------------+---------------------------------------------------------------
           . |         74  -695.7129  -679.9124       3   1365.825   1372.737
-----------------------------------------------------------------------------
Note: BIC uses N = number of observations. See [R] BIC note.

[Rich Goldstein]

looking at the differences in the results, my guess is that Neo Zhou has an old version of fitstat; try this and see if you get the same result (ignoring the /Users line of course)

. which fitstat
/Users/rich/Library/Application Support/Stata/ado/plus/f/fitstat.ado
*! 4.1.10 | 2017-01-05 | freese long | fixed bug in tobit (for v <14 and 15+)

[Neo Zhou]

Ah, I see the problem. I do have an outdated fitstat. But when I was trying to use adoupdate to update all the packages, Stata did not find an updated version of fitstat.
I uninstall fitstat, and reinstall it using ssc install fitstat, still, I got the version 1.6.4 installed. How you guys updated your fitstat?

By the way, I was using Stata 16 on a Windows machine

[Leonardo Guizzetti]

Try installing -spost13_ado- which will include, among other things, -fitstat-.

[Neo Zhou]

Thanks. much appreciated.