[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

From |
edoardo masset <edomasset@yahoo.com> |

To |
statalist@hsphsun2.harvard.edu |

Subject |
st: ordered probit with ml, again |

Date |
Thu, 17 Jul 2003 14:51:03 -0700 (PDT) |

It looks like my previous message has gone unnoticed. So I am sending it again. Forgive the repetition, but I am really stuck on this and hoping to get some help. Thanks dear statalisters, I am trying to estimate an ordered probit model using the ML command. The reason for not simply use the oprobit command, is that I want then to modify the likelihood function in order to account for censored observations. I know I could probably use other models then a censored ordered probit to correct for censoring (like Cox regression), but to show this I need first to replicate the results from a published study where the censored ordered probit model was used. I am new to the ML command, so what I did was to follow the instructions in the FAQ ‘Is it possible to include a constant term (intercept) in ordered probit model within Stata? What is the relationship between ordered probit and probit?’(http://www.stata.com/support/faqs/stat/constant.html). In this FAQ W. Gould explains that estimating an ordered probit is equivalent to estimate a series of binary probit with one equation for each outcome, and constraining the coefficients (but not the constants) to be the same across equations, the constants derived from each equation are the cut-off points with reversed sign. My dependent variable (x) has 12 outcomes (from 0 to 11). Therefore I defined first 11 dummies of the type: Dum1= x ==11 Dum2= x>=10 Dum3= x>=9 ……………. Dum10=x>=1 Then, I estimate the 11 equations simultaneously, imposing the constraints on the coefficients. The likelihood function of the model is the sum of the likelihood functions for each of the equations. The model specification is below (the model contains some 50 explanatory variables, I reduced here to 5 or six to save space). Now, the coefficients estimates I obtain are similar to those obtained by simply running the ‘oprobit’ command. However: - the value of the log-likelihood I get is twice the value obtained with oprobit - standard errors are much smaller in my model than in ‘oprobit’ - the constants from my model are ‘similar’ to the cut-off points obtained through oprobit, but still quite different. Any guess about what am I doing wrong? Any other way to estimate this model? Any help really appreciated. Thanks. Edoardo cap program drop maxprob program define maxprob args lnf1 theta1 theta2 theta3 theta4 theta5 theta6 theta7 theta8 theta9 theta10 theta11 qui replace `lnf1' =($ML_y1)*ln(normprob(`theta1'))+(1-$ML_y1)*ln(normprob(-`theta1'))+ /* */ ($ML_y2)*ln(normprob(`theta2'))+(1-$ML_y2)*ln(normprob(-`theta2'))+ /* */ ($ML_y3)*ln(normprob(`theta3'))+(1-$ML_y3)*ln(normprob(-`theta3'))+ /* */ ($ML_y4)*ln(normprob(`theta4'))+(1-$ML_y4)*ln(normprob(-`theta4'))+ /* */ ($ML_y5)*ln(normprob(`theta5'))+(1-$ML_y5)*ln(normprob(-`theta5'))+ /* */ ($ML_y6)*ln(normprob(`theta6'))+(1-$ML_y6)*ln(normprob(-`theta6'))+ /* */ ($ML_y7)*ln(normprob(`theta7'))+(1-$ML_y7)*ln(normprob(-`theta7'))+ /* */ ($ML_y8)*ln(normprob(`theta8'))+(1-$ML_y8)*ln(normprob(-`theta8'))+ /* */ ($ML_y9)*ln(normprob(`theta9'))+(1-$ML_y9)*ln(normprob(-`theta9'))+ /* */ ($ML_y10)*ln(normprob(`theta10'))+(1-$ML_y10)*ln(normprob(-`theta10'))+ /* */ ($ML_y11)*ln(normprob(`theta11'))+(1-$ML_y11)*ln(normprob(-`theta11')) /* */ end do constr1 do constr2 do constr3 ml model lf maxprob (school1=cos for sav semurb rural sex) /* */ (school2=cos for sav semurb rural sex) /* */ (school3=cos for sav semurb rural sex) /* */ (school4=cos for sav semurb rural sex) /* */ (school5=cos for sav semurb rural sex) /* */ (school6=cos for sav semurb rural sex) /* */ (school7=cos for sav semurb rural sex) /* */ (school8=cos for sav semurb rural sex ) /* */ (school9=cos for sav semurb rural sex) /* */ (school10=cos for sav semurb rural sex) /* */ (school11=cos for sav semurb rural sex ) /* */ , constraints(1-590) __________________________________ Do you Yahoo!? SBC Yahoo! DSL - Now only $29.95 per month! http://sbc.yahoo.com * * For searches and help try: * http://www.stata.com/support/faqs/res/findit.html * http://www.stata.com/support/statalist/faq * http://www.ats.ucla.edu/stat/stata/

- Prev by Date:
**Re: st: Matrix subscripts using row and column names** - Next by Date:
**RE: st: PDF Stata 8 manuals** - Previous by thread:
**Re: st: announcing statsbyfast - a faster statsby** - Next by thread:
**st: command line Stata on Mac OS** - Index(es):

© Copyright 1996–2016 StataCorp LP | Terms of use | Privacy | Contact us | What's new | Site index |