Problem with Pseudo R squared. ".z" displayed. What is it?

[Firangiz Aghayeva]

Hello!

I am working on a binomial logistic regression with around 1.6mln observations (panel data). I use xtlogit with fixed effects. Independent variables are both dummy and numerical variables. When I use asdoc function, pseudo R squared is displayed as ".z" (with esttab function pseudo R squared is displayed as blank). There is no pseudo R squared shown among estimation results in STATA. Did you have anything like that in your analysis? How can I solve this issue?

Kind regards,
Firangiz Aghayeva

[daniel klein]

This is a problem with the comunity-contributed asdoc. Attaullah Shah, the author of asdoc, might be able to explain.

[Joro Kolev]

.z is a missing value. -xtlogit, fe- allegedly calculates Pseudo R-quared and saves it in e(r2 p).

Check what you have after your estimation, by say typing
-ereturn list- and seeing what is in e(r2 p),
or typing -display e(r2 p)- and seeing what is being displayed.

[William Lisowski]

In case it's not apparent, something erased underscore characters in Joro's post - running

display e(r2_p)

after xtlogit, fe will confirm that a Pseudo R-squared was calculated, as documented in help xtlogit. The fact that it's not displayed in the output of xtlogit can be misleading.

[Firangiz Aghayeva]

Thank you for your answers. I tried "display e(r2_p)", but nothing is displayed. "ereturn list" also doesn't display pseudo R squared. What could it mean? Does it mean my model is not correct?

[Rich Goldstein]

I don't know anything about esttab, but -xtlogit- with FE does have a pseudo-r2; please show us, as requested by the FAQ, exactly what you typed and exactly what Stata (not esttab) gave back to you

[Joro Kolev]

As Rich said show exactly what you typed at Stata, and exactly what Stata returned, where you type:

xtlogit y x, fe

ereturn list

[Firangiz Aghayeva]

Dear Rich Goldstein and @Joro Kolev,
Please, see below the STATA output of what I typed and what STATA returned:

5 . set sortseed 123


6 . xtset LP_ID

Panel variable: LP_ID (unbalanced)

7 . xtlogit Dummy_invested_or_not Dummy_LP_Fund_PRI Dummy_fund_PRI Dummy_LP_PRI Prior_relationship Dummy_home_bias log_fund_sequence_number i.Fund_region_num i.Fund_industry_num i.Fund_vintage, vce(cluster LP_region)

Fitting comparison model:

Iteration 0: log pseudolikelihood = -48013.658
Iteration 1: log pseudolikelihood = -46895.703
Iteration 2: log pseudolikelihood = -38906.955
Iteration 3: log pseudolikelihood = -38403.311
Iteration 4: log pseudolikelihood = -37171.3
Iteration 5: log pseudolikelihood = -37166.422
Iteration 6: log pseudolikelihood = -37166.421

Fitting full model:

tau = 0.0 log pseudolikelihood = -37166.421
tau = 0.1 log pseudolikelihood = -36371.442
tau = 0.2 log pseudolikelihood = -36279.049
tau = 0.3 log pseudolikelihood = -36281.203

Iteration 0: log pseudolikelihood = -36278.278
Iteration 1: log pseudolikelihood = -36124.382
Iteration 2: log pseudolikelihood = -36119.408
Iteration 3: log pseudolikelihood = -36119.393
Iteration 4: log pseudolikelihood = -36119.393

Calculating robust standard errors ...

Random-effects logistic regression Number of obs = 1,595,341
Group variable: LP_ID Number of groups = 2,057

Random effects u_i ~ Gaussian Obs per group:
min = 116
avg = 775.6
max = 3,622

Integration method: mvaghermite Integration pts. = 12

Wald chi2(7) = .
Log pseudolikelihood = -36119.393 Prob > chi2 = .

(Std. err. adjusted for 8 clusters in LP_region)
------------------------------------------------------------------------------------------
| Robust
Dummy_invested_or_not | Coefficient std. err. z P>|z| [95% conf. interval]
-------------------------+----------------------------------------------------------------
Dummy_LP_Fund_PRI | .360916 .1494298 2.42 0.016 .068039 .653793
Dummy_fund_PRI | -.5526997 .1271336 -4.35 0.000 -.8018771 -.3035224
Dummy_LP_PRI | -.0267808 .179703 -0.15 0.882 -.3789922 .3254306
Prior_relationship | 4.484332 .2249435 19.94 0.000 4.04345 4.925213
Dummy_home_bias | .6723201 .1494194 4.50 0.000 .3794633 .9651768
log_fund_sequence_number | .3110024 .6432726 0.48 0.629 -.9497888 1.571794
|
Fund_region_num |
2 | .0399756 .3970461 0.10 0.920 -.7382205 .8181717
3 | -.0123732 .4394583 -0.03 0.978 -.8736956 .8489493
4 | -.1256316 .4559336 -0.28 0.783 -1.019245 .7679819
5 | -.0019979 .2965385 -0.01 0.995 -.5832027 .5792068
|
Fund_industry_num |
2 | -.3614807 .1697753 -2.13 0.033 -.6942341 -.0287272
3 | -.5969741 .4020793 -1.48 0.138 -1.385035 .191087
4 | -.3183784 .1922162 -1.66 0.098 -.6951151 .0583584
5 | -.5843853 .2238303 -2.61 0.009 -1.023085 -.1456859
6 | -.525885 .1175568 -4.47 0.000 -.756292 -.295478
7 | -.3058262 .2969135 -1.03 0.303 -.887766 .2761136
8 | -.5488518 .1062835 -5.16 0.000 -.7571637 -.3405399
9 | -.7461087 .5087303 -1.47 0.142 -1.743202 .2509843
10 | -.9997778 .6044649 -1.65 0.098 -2.184507 .1849517
11 | -.409583 .1349732 -3.03 0.002 -.6741257 -.1450403
|
Fund_vintage |
2007 | -.0140132 .0363593 -0.39 0.700 -.0852761 .0572497
2008 | -.0192336 .0848916 -0.23 0.821 -.1856181 .147151
2009 | .0912352 .1614439 0.57 0.572 -.2251891 .4076595
2010 | -.0553971 .2067173 -0.27 0.789 -.4605555 .3497613
2011 | -.2150118 .1255786 -1.71 0.087 -.4611413 .0311178
2012 | -.1713196 .1268857 -1.35 0.177 -.4200109 .0773718
2013 | -.2094859 .2032533 -1.03 0.303 -.6078551 .1888832
2014 | -.3440877 .2471677 -1.39 0.164 -.8285276 .1403522
2015 | -.6053826 .185353 -3.27 0.001 -.9686678 -.2420973
2016 | -.6692628 .1882094 -3.56 0.000 -1.038146 -.300379
2017 | -.4904899 .1533822 -3.20 0.001 -.7911134 -.1898664
2018 | -.511232 .1757973 -2.91 0.004 -.8557883 -.1666756
2019 | -.5290861 .2060593 -2.57 0.010 -.9329549 -.1252173
2020 | -.52424 .2321195 -2.26 0.024 -.9791859 -.0692942
2021 | -.5818161 .2709813 -2.15 0.032 -1.11293 -.0507026
|
_cons | -5.595157 .2272539 -24.62 0.000 -6.040567 -5.149748
-------------------------+----------------------------------------------------------------
/lnsig2u | -.6099625 .7153801 -2.012082 .7921568
-------------------------+----------------------------------------------------------------
sigma_u | .7371372 .2636666 .3656638 1.485986
rho | .1417525 .0870322 .0390556 .4016269
------------------------------------------------------------------------------------------



8 . display e(r2_p)
.

9 . ereturn list

scalars:
e(df_m) = 7
e(chi2) = .
e(p) = .
e(N_clust) = 8
e(ll_c) = -37166.42075489485
e(chi2_c) = 2094.055561565459
e(sigma_u) = .7371371966902871
e(rho) = .1417524951655696
e(k_aux) = 1
e(g_avg) = 775.5668449197861
e(g_min) = 116
e(N_g) = 2057
e(n_quad) = 12
e(k_eq_model) = 1
e(ll) = -36119.39297411212
e(rc) = 0
e(converged) = 1
e(k_dv) = 1
e(k_eq) = 2
e(k) = 40
e(ic) = 4
e(N) = 1595341
e(rank) = 37
e(g_max) = 3622

macros:
e(cmdline) : "xtlogit Dummy_invested_or_not Dummy_LP_Fund_PRI Dummy_fund_PRI Dummy.."
e(cmd) : "xtlogit"
e(marginsdefault) : "predict(pr)"
e(predict) : "xtlogit_re_p"
e(model) : "re"
e(clustvar) : "LP_region"
e(ivar) : "LP_ID"
e(chi2_ct) : "LR"
e(chi2type) : "Wald"
e(vcetype) : "Robust"
e(vce) : "robust"
e(depvar) : "Dummy_invested_or_not"
e(title) : "Random-effects logistic regression"
e(distrib) : "Gaussian"
e(intmethod) : "mvaghermite"
e(opt) : "moptimize"
e(user) : "xtlogit_d2"
e(ml_method) : "d2"
e(technique) : "nr"
e(which) : "max"
e(properties) : "b V"

matrices:
e(b) : 1 x 40
e(V) : 40 x 40
e(Cns) : 3 x 41
e(V_modelbased) : 40 x 40
e(ilog) : 1 x 20
e(gradient) : 1 x 40

functions:
e(sample)

[William Lisowski]

In post #1 you tell us you fit xtlogit with fixed effects. In post #8 we see that the xtlogit command did not include the fe option, so in fact Stata tells you that it fit the default random effects model in the output you displayed

Calculating robust standard errors ...

Random-effects logistic regression Number of obs = 1,595,341
Group variable: LP_ID Number of groups = 2,057

Random effects u_i ~ Gaussian Obs per group:
min = 116
avg = 775.6
max = 3,622

You need to reread the documentation for xtlogit to understand the meaning of a fixed effects model in the panel data setting.

[Rich Goldstein]

In addition to William Lisowski 's comment, with which I fully agree, please read the FAQ and learn how to post things so they are more easily legible

[Firangiz Aghayeva]

Thank you for your valuable comments. I corrected my model and got pseudo R squared properly displayed. Now I use clogit, because I need to cluster robust standard errors. In xtlogit, fe function vce options allow only oim, bootstrap and jackknife.

[Joro Kolev]

Excellent ! We probably also learnt something new today, apparently the -xtlogit, re- does not display a pseudo R-squared...