Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.

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

From |
Maarten Buis <maartenlbuis@gmail.com> |

To |
statalist@hsphsun2.harvard.edu |

Subject |
st: Re: Regarding logit and probit model |

Date |
Fri, 6 Jul 2012 13:03:42 +0200 |

Such questions need to be asked on Statalist and not to its individual member, as is clearly stated on the Statalist FAQ: http://www.stata.com/support/faqs/resources/statalist-faq/#private Problem 1: I don't even think R-squared is useful in linear regression, so I certainly don't thing the many "pseudo- R-squares" that exist for models like -logit- or -probit- have any value. Problem 2: I don't understand that question: you cannot estimate a -probit- with -nlogit-. The -logit- and -probit- is so close (up to a fixed factor) that the choice between them is purely arbitrary. The most important factor influencing that choice is typically the tribal habits of different (sub-(sub-))disciplines. It certainly has nothing to do with one model being better at "eliminating unobserved error". Problem 3: There is nothing you can meaningfully do about heteroscedasticity in logit or probit models. Models exist, but they are so sensitive that the results are basically meaningless. -- Maarten On Fri, Jul 6, 2012 at 11:19 AM, KADALI RAGHURAM wrote: > Dear Maarten, > > > Good morning, > > > Greetings from B R KADALI, > > > This is BR Kadali Research Scholar from IIT Bombay India, > > I working in Transportation Engineering, my area of Interest is "Pedestrian > Flow Modeling under Mixed Traffic condition". Presently i am working with > Pedestrian Gap acceptance behaviour, for this data analysis i am trying to > uses discrete choice modeling LOGIT and PROBIT Models. > > Problem-1 > > The problem here, i am struggling with MacFedden R-squared value. I got > R-squared value around 0.784. But it is unfeasible values. I used LOGIT > MODEL- There are 4200 sample vehicular gaps i collected out these only 200 > gaps are accepted by pedestrian i.e., (Probability of accepted gaps=1=200 > sample) remain data samples are 4000 rejected values i.e., o coded values. > So, my doubt here the R-squared value is acceptable or not. Now I converted > data as maximum rejected gap and accepted gap now total data points 400 in > this case also I got same R-sq 0.789. This R-sq value is acceptable or not. > > > Problem-2 > > When I build PROBIT model in the NLOGIT, I couldn’t get any much difference > when compared with LOGIT model. The only difference I observed is > coefficient values. But literature says that there is an improvement by > probit model when compared to logit model by eliminating unobserved error > like heteroscedasticity. It will be eliminated by PROBIT model, what are the > difference can I observe here. In this case how can judge my model is better > than the LOGIT model. If there is difference between LOGIT and PROBIT in the > case of binary data, what is the reason behind it. > > Problem-3 > > If I take model with LOGIT as HETROSCADACITY model, what improvement I can > get, is it eliminating the unobserved error by gender and Age of the people. > What are the main advantages over the simple LOGIT model and PROBIT model of > HETROSCADACITY model?. If suppose in the HETROSCADACITY model the gender is > getting significant, how can I represent the utility equations of this case. > > I am really sorry for the inconvenience, but I am really want some useful > information about these doubts. > I am waiting for valuable reply and suggestions > > Thank you > > > > -- > Regards > B R Kadali -------------------------- Maarten L. Buis Institut fuer Soziologie Universitaet Tuebingen Wilhelmstrasse 36 72074 Tuebingen Germany http://www.maartenbuis.nl -------------------------- * * 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/

- Prev by Date:
**Re: st: R squared of OLS with dummy variables** - Next by Date:
**Re: st: test of repeated measurements - ordinal data (likert scale)** - Previous by thread:
**st: R squared of OLS with dummy variables** - Next by thread:
**st: r(198) error using lrtest to test for heteroskedasticity.** - Index(es):