Notice: On March 31, it was **announced** that Statalist is moving from an email list to a **forum**. The old list will shut down on April 23, and its replacement, **statalist.org** is already up and running.

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

From |
Austin Nichols <austinnichols@gmail.com> |

To |
statalist@hsphsun2.harvard.edu |

Subject |
Re: st: Problem with margins after logit on a person period data |

Date |
Thu, 9 Jun 2011 15:22:21 -0400 |

Urmi Bhattacharya <ub3@indiana.edu> : I don't see how using -cloglog- instead of -logit- changes anything. Either way you are modeling the hazard. On Thu, Jun 9, 2011 at 1:36 PM, Urmi Bhattacharya <ub3@indiana.edu> wrote: > Hi Austin, > > I wonder if I could circumvent this problem you mentioned by > considering the clogclog regression instaed of logit because then what > I am modeling is the discrete hazard of dropping out in an interval > conditional upon having survived till before that. > > > So I could do the following > > clogclog school_left childage i.childfemale i.urban i.scstobc > i.casteother i.dadp i.dadm i.momp i.momm wagep wage5 wage8 wage9 distp > distm disth percapcons durat1 durat2 durat3 durat4 durat5 durat6 > durat7 durat8 durat9 durat10 durat11, nocons nolog > > My goal is to find investigate the if the effect of mothers > education(momp) on the probability of dropping out varies with > duration. > > So if I find the predicted probability when momp=1 and durat1=1 and > the predicted probability when momp=1 and durat2=1, then look at the > estimated change in probability and see that the estimated change is > significantly positive, then could I use this as evidence to say that > in risk period 2, momp=1 matters more than in risk period1 in terms of > interval hazard of dropping out? > > Best > > Urmi > > On Thu, Jun 9, 2011 at 12:17 PM, Austin Nichols <austinnichols@gmail.com> wrote: >> Urmi Bhattacharya <ub3@indiana.edu> : >> >> This whole exercise is highly suspect--you are computing marginal >> effects over a sample of periods at risk, not people. Note that >> people are in your model for very different numbers of periods, but >> you are averaging over all periods; what is the goal here? You said >> you are "interested in the marginal effects of the variables on the >> probability of hazard" which I think means you want to measure the >> marginal effects of the variables on the conditional probability of >> leaving school (conditional on not having left yet) at different >> durations, which means you should calculate marginal effects for each >> sample of people still at risk, at different durations. These are >> unlikely to be very informative in the probability metric, however; >> odds ratios are used for a reason for such applications. >> >> On Wed, Jun 8, 2011 at 10:31 PM, Urmi Bhattacharya <ub3@indiana.edu> wrote: >>> Hi, >>> >>> I dropped one of the duration dummies durat1 and ran the following >>> logit regression >> .... * * For searches and help try: * http://www.stata.com/help.cgi?search * http://www.stata.com/support/statalist/faq * http://www.ats.ucla.edu/stat/stata/

**References**:**st: Problem with margins after logit on a person period data***From:*Urmi Bhattacharya <ub3@indiana.edu>

**Re: st: Problem with margins after logit on a person period data***From:*Richard Williams <richardwilliams.ndu@gmail.com>

**Re: st: Problem with margins after logit on a person period data***From:*Urmi Bhattacharya <ub3@indiana.edu>

**Re: st: Problem with margins after logit on a person period data***From:*Austin Nichols <austinnichols@gmail.com>

**Re: st: Problem with margins after logit on a person period data***From:*Urmi Bhattacharya <ub3@indiana.edu>

- Prev by Date:
**Re: st: Query about finding predicted change in probability after logit for changing two variables** - Next by Date:
**Re: st: Re: Re: st: oaxaca decomposition** - Previous by thread:
**Re: st: Problem with margins after logit on a person period data** - Next by thread:
**Re: st: Problem with margins after logit on a person period data** - Index(es):