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

From |
"Schaffer, Mark E" <M.E.Schaffer@hw.ac.uk> |

To |
<statalist@hsphsun2.harvard.edu> |

Subject |
RE: st: Test of ordered probit vs ordinary probits |

Date |
Thu, 1 Nov 2007 11:14:59 -0000 |

Thanks, Richard (and thanks, Partha), for the very helpful discussion and examples. I have one last question. The examples we've been going through have focused on the constraint that the slope coeffs are the same across equations. The key point is that imposing this constraint also implies that the cutoffs satisfy the ordering constraint. Say that I don't impose the constraint that the slope coeffs are the same. Say also that I have some doubts about whether my outcome variable is genuinely order-able. I know that outcome A is different from B, and B is different from C, and I hypothesize that C>B>A but I have some doubts about it. (Silly example: in terms of education, PhD > BA > high school, but maybe spending so much time getting the PhD means you forget a lot of the useful stuff you learned in high school and university. Hmmm, maybe not so silly.) This takes me back to my original example of a bunch of separate probits, but this time contrasted with, e.g., the default behavior of -gologit2- (with link(pr) to make it a generalized ordered probit). The latter imposes only the constraint that the cutoffs are ordered, corresponding to order-ability of the outcome variable. How would you test this constraint? I looked through the mass of very useful refs and packages Richard suggested yesterday, but didn't see this. Apologies in advance if it's there and I didn't spot it. Cheers, Mark Prof. Mark Schaffer Director, CERT Department of Economics School of Management & Languages Heriot-Watt University, Edinburgh EH14 4AS tel +44-131-451-3494 / fax +44-131-451-3296 email: m.e.schaffer@hw.ac.uk web: http://www.sml.hw.ac.uk/ecomes > -----Original Message----- > From: owner-statalist@hsphsun2.harvard.edu > [mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of > Richard Williams > Sent: 01 November 2007 02:53 > To: statalist@hsphsun2.harvard.edu; statalist@hsphsun2.harvard.edu > Subject: RE: st: Test of ordered probit vs ordinary probits > > At 06:48 PM 10/31/2007, Schaffer, Mark E wrote: > >Thanks, Richard, that's *really* helpful. The only thing I > would add > >is that another reason my test result vs. the test result > from -omodel- > >differ is that I used -suest-, which means it was a heterosk-robust > >Wald test vs. -omodel-'s non-robust approximate LR test. > > Sounds good. > > > >Your use of the -coef- option of -test- is very nifty. If I > understand > >what you've said and what the manual says about -coef- correctly: > > > >(1) -gologit2- imposes during the estimation the one-step > constraints > >that the cutoffs are ordered. > > If you use gologit2 with the pl option, it will require that > the coefficients be equal across equations, which in turn > will force the cutoffs to be ordered. Yes, these are one > step constraints; internally gologit2 is creating a bunch of > constraint commands and then imposing those constraints on > the estimates. In an unconstrained gologit, the coefficients > can all differ across equations; in a totally constrained > gologit the coefficients are all the same across equations > and it becomes the ologit model. > > To be clear, there is nothing in the code that explicitly > says the cutpoints have to be ordered; rather the ordering of > the cutpoints is a consequence of the rest of the model's > requirements/constraints. > > >(2) Separate -probit-s followed by -test- with the -coef- option in > >effect imposes asymptotically equivalent two-step > constraints that the > >cutoffs are ordered. > > Yes, I believe that is correct, although again I would > clarify that the ordering of the cutoffs is a consequence of > the constraint that the coefficients be equal; it isn't a > separate constraint. i.e. equal coefficients across equations > implies/forces ordered cutoff points. (If that isn't clear I > can try to elaborate further on why equal coefficients > implies ordered cutpoints.) > > Weesie's -linest- command (available from his own site; use > -findit-) can be another nifty way of doing this sometimes. > So, for example, picking up from the last example where the > suest command was given, > > . suest probit12 probit23 probit34 > > [ Output deleted] > > . constraint 1 [probit12]yr89 = [probit23]yr89 > > . constraint 2 [probit23]yr89 = [probit34]yr89 > > . linest, c(1 2) modify > > Two-step constrained suest > Dim unrestricted > model = 6 > Dim restricted > model = 4 > # restrictions > = 2 > Wald X2 for > restrictions = 13.1530 > Prob > chi2(2) > = 0.0014 > ( 1) [probit12]yr89 - [probit23]yr89 = 0 > ( 2) [probit23]yr89 - [probit34]yr89 = 0 > -------------------------------------------------------------- > ---------------- > | Coef. Std. Err. z P>|z| > [95% Conf. Interval] > -------------+------------------------------------------------ > ---------- > -------------+------ > probit12 | > yr89 | .363189 .045948 7.90 0.000 > .2731326 .4532454 > _cons | .9988136 .037007 26.99 0.000 > .9262812 1.071346 > -------------+------------------------------------------------ > ---------- > -------------+------ > probit23 | > yr89 | .363189 .045948 7.90 0.000 > .2731326 .4532454 > _cons | -.0067555 .0321153 -0.21 0.833 > -.0697003 .0561892 > -------------+------------------------------------------------ > ---------- > -------------+------ > probit34 | > yr89 | .363189 .045948 7.90 0.000 > .2731326 .4532454 > _cons | -1.063885 .0365802 -29.08 0.000 > -1.135581 -.9921892 > -------------------------------------------------------------- > ---------------- > > We modified the stored results of suest. > Post-estimation commands use the constrained model! > Beware (see online help!) > > One nice (dangerous?) thing about -linest- is that, with the > -modify- option, the constrained estimates replace the > unconstrained estimates. So, any post-estimation command is > now going to use the results you see above. If you have a > command that does not support the -constraints- option, > linest may be the next best thing. Early versions of > gologit2 used linest until I figured out how to do 1 step estimation. > > > ------------------------------------------- > Richard Williams, Notre Dame Dept of Sociology > OFFICE: (574)631-6668, (574)631-6463 > HOME: (574)289-5227 > EMAIL: Richard.A.Williams.5@ND.Edu > WWW: http://www.nd.edu/~rwilliam > > * > * 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/ > * * 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/

**Follow-Ups**:**RE: st: Test of ordered probit vs ordinary probits***From:*Richard Williams <Richard.A.Williams.5@ND.edu>

- Prev by Date:
**st: West Coast Stata Users Group Meetings** - Next by Date:
**RE: st: Test of ordered probit vs ordinary probits** - Previous by thread:
**st: West Coast Stata Users Group Meetings** - Next by thread:
**RE: st: Test of ordered probit vs ordinary probits** - Index(es):

© Copyright 1996–2017 StataCorp LLC | Terms of use | Privacy | Contact us | What's new | Site index |