Ddifference in number of obs -xtreg- versus -xtpoisson-

[Lennart Osses]

Dear Statalisters,

I have made an observation in my data anylsis for which I am struggeling to find an explanation:

Analyzing the exact same model with -xtreg- and -xtpoisson- leaves me with a substantially different number of observations (and groups). -xtreg- gives me ~10.6k observations, -xtpoisson- just ~7K.

My first hypothesis was that it is due to my DV having values below zero, but this is not the case (see below).

Does anyone know what the reason for this might be?

Thanks and best regards,
Lennart

***

Results

Quick explanation of variables:

• pat_filings: My DV, number of patents filed
• ExtFounderCEO2; ProfCEO: My IVs, dummies for specific CEOs
• $ControlsActivity: Set of of controls

. sum pat_filings

    Variable |        Obs        Mean    Std. Dev.       Min        Max
-------------+---------------------------------------------------------
 pat_filings |     14,473    62.11815    304.5335          0       8905

. xtreg pat_filings ExtFounderCEO2 ProfCEO $ControlsActivity i.Year, fe vce(robust)

Fixed-effects (within) regression Number of obs = 10,621
Group variable: Gvkey Number of groups = 815

R-sq: Obs per group:
within = 0.0528 min = 1
between = 0.2036 avg = 13.0
overall = 0.1851 max = 20

F(35,814) = 2.84
corr(u_i, Xb) = 0.2777 Prob > F = 0.0000

. xtpoisson pat_filings ExtFounderCEO2 ProfCEO $ControlsActivity i.Year, fe vce(robust)
note: 22 groups (22 obs) dropped because of only one obs per group
note: 292 groups (3529 obs) dropped because of all zero outcomes

Iteration 0:   log pseudolikelihood = -141669.32  
Iteration 1:   log pseudolikelihood =  -85088.56  
Iteration 2:   log pseudolikelihood = -76207.982  
Iteration 3:   log pseudolikelihood = -74207.777  
Iteration 4:   log pseudolikelihood = -73732.966  
Iteration 5:   log pseudolikelihood = -73631.856  
Iteration 6:   log pseudolikelihood = -73610.757  
Iteration 7:   log pseudolikelihood = -73606.019  
Iteration 8:   log pseudolikelihood =  -73604.85  
Iteration 9:   log pseudolikelihood = -73604.614  
Iteration 10:  log pseudolikelihood = -73604.563  
Iteration 11:  log pseudolikelihood = -73604.552  
Iteration 12:  log pseudolikelihood = -73604.549  
Iteration 13:  log pseudolikelihood = -73604.548  
Iteration 14:  log pseudolikelihood = -73604.548  
Iteration 15:  log pseudolikelihood = -73604.548  

Conditional fixed-effects Poisson regression    Number of obs     =      7,070
Group variable: Gvkey                           Number of groups  =        501

                                                Obs per group:
                                                              min =          2
                                                              avg =       14.1
                                                              max =         20

                                                Wald chi2(35)     =   35822.61
Log pseudolikelihood  = -73604.548              Prob > chi2       =     0.0000

[Clyde Schechter]

There is no mystery here. Stata has already told you what's going on in the -xtpoisson- output.

note: 22 groups (22 obs) dropped because of only one obs per group
note: 292 groups (3529 obs) dropped because of all zero outcomes

3529 + 22 + 7070 = 10,621, so the entire difference is accounted for by these two conditions, which are restrictions imposed by -xtpoisson- but are allowable in -xtreg-.

[Lennart Osses]

Thanks Clyde, should have read that more carefully!