Wald Chi2 not reported (represented as a dot) in conditional logit (clogit) output

[Richard Bui]

Hi everyone,

I hope you are all doing well.
I'm currently working on clogit models with firm and year-fixed effects using unbalanced panel data in Stata. I've encountered an issue that I'm hoping you could help me with.

In my baseline regression, where I don't include any control variables, everything seems to be working fine. However, as I add more control variables to the model, I've noticed that the Wald Chi2 statistic and its associated p-value are being displayed as dots (.) in the output.
Strangely, there are no error messages, and the rest of the output appears normal.

I've searched online but couldn't find anything that addresses this specific situation. I reached out to chatGPT and it mentioned this situation is not uncommon. According to it, the chi2 may not be that meaningful for the clogit models due to its unique structure and assumptions, implying it's "oke" to have chi2 not reported in clogit models. But again, I couldn't find anything online to support this notation.

My best guess so far is something to do with the sample size, which was reduced dramatically (by half) when adding control variables.

Since this is my first time posting here, I want to apologize in advance if I've made any mistakes in the post.
Thank you.

[Carlo Lazzaro]

Richard:
welcome to this forum.
As per FAQ, posters are kindly requested to share what they typed and what Stata gave them back. Thanks.
This way, the chances of getting helpful replies increase.
That said, a tentative example that possibly mimicks yours is pasted below:

. clogit y x1 x2  if id<=1025, group(id) vce(cluster id)
note: multiple positive outcomes within groups encountered.

<snip>

Conditional (fixed-effects) logistic regression         Number of obs =     19
                                                        Wald chi2(0)  =      .
                                                        Prob > chi2   =      .
Log pseudolikelihood = -.69314718                       Pseudo R2     = 0.9097

                                     (Std. err. adjusted for 5 clusters in id)
------------------------------------------------------------------------------
             |               Robust
           y | Coefficient  std. err.      z    P>|z|     [95% conf. interval]
-------------+----------------------------------------------------------------
          x1 |  -36.04486          .        .       .            .           .
          x2 |   48.70868          .        .       .            .           .
------------------------------------------------------------------------------
convergence not achieved
r(430);

[Richard Bui]

Hi Carlo,

Thank you for your reply.
I'm asking my supervisors for permission to post my Stata input and output here and hopefully, I can show you something tomorrow

In my case though, nothing is missing from the output and everything seems fine. I don't have any missing values (robust std err, p-value...) and no error message.
Despite that, the Wald chi2 is just a dot. ChatGPT says that is not important and not uncommon with clogit. But I've not been able to find any evidence to support this.

Have you ever encountered this problem?
Thank you.

[Carlo Lazzaro]

Richard:
yes , the omission is not important (like in all regressions). Usually, those test check whether the the coefficients of all parameters are jointly zero (or, put differently, if a constant-only parameter is equally informative).
See -help j_robustsingular-, though.

[Richard Bui]

Thank you, Carlo. So I can safely ignore the omission of chi2, can't I? My concern is that I see papers often report chi2 in the regression output when working with logit models.

[Carlo Lazzaro]

Richard:
as you can read in the helpfile I've previously pointed you out to, there are different reasons why you may experience that omission.
Therefore, I would not be concerned about that, but rather interested in tracking down the possible causes of the omission, as this drill helps researchers out with improving their knowledge of the quantitative tools they use (it works for me, at least).

[Richard Bui]

Thank you, Carlo for your insights.