Pseudo R2

[Abebaw Assaye]

Dear everyone,

I used the Tobit model with 14 variables and I found Pseudo R² negative.
I need some explanations about it. Can someone give me some advice and possible reasons why it is negative?
Thank you

[Andrew Musau]

Can you show your exact command and output within CODE delimiters?

ADDED IN EDIT: See this Stata FAQ for an explanation: https://www.stata.com/support/faqs/s...ics/pseudo-r2/. From this link, this can happen due to positive maximized log-likelihoods, unlike logit or probit where the maximized log-likelihoods are always negative.

[Abebaw Assaye]

Dear Andrew Musa,
Thank you so much
Here is the tobit command that I used and the result tables.

tobit Input_c EXP_HEAD SEX_HEAD Dependency_ratio Edu_status USED_CREDIT_01 TLU Non_off_farm Main_mrkt_KM_01 Membership_social Rice_area LnFarm_Asset fertilizer_ext Hired_labor , ll robust
Tobit regression Number of obs = 594
F( 13, 581) = 2.74
Prob > F = 0.0009
Log pseudolikelihood = 519.71613 Pseudo R2 = -0.0438

Robust
Input_c Coef. Std. Err. t P>t [95% Conf. Interval]

EXP_HEAD .0007244 .0003492 2.07 0.038 .0000386 .0014103
SEX_HEAD -.0294337 .0206909 -1.42 0.155 -.0700717 .0112043
Dependency_ratio .0015085 .0027658 0.55 0.586 -.0039237 .0069407



/sigma .0940837 .0108504 .0727728 .1153946

21 left-censored observations at Input_c <= 0
573 uncensored observations
0 right-censored observations

[Andrew Musau]

Yes, the issue is positive log-pseudolikelihoods. You probably should rescale the variable "EXP_HEAD" so that its coefficients are readable and you avoid reporting coefficients such as "0.000".

replace EXP_HEAD = EXP_HEAD/1000
label variable EXP_HEAD  "Expenditure of household head in thousands"