Re: st: update seqlogit

[Ángel Rodríguez Laso <angelrlaso@gmail.com>]

Dear Maarten,

I've read with interest your message describing the -seqlogit- package
and your dissertation  chapter where you apply it to describe
inequalities of educational opportunities. I was thinking about the
application of this model to outcomes of contact in a survey. I would
appreciate your opinion very much.

The possibilities of the outcome are incorporated to the variable
inciden, with the codes: 1. death 2. moved 3. absent 4. long absence
5. refuses 6. interviewed. Of course there were more possibilities,
but they were either too infrequent (incorrect address, hospitalised,
ill, doesn’t speak Spanish) or equivocal (other). I understand these
categories as steps in a continuum from being death to willing to
participate in the survey that can be described as a tree (sorry, I
have't managed to get a better presentation):

                                                                 Alive
  ------------------------------   Death
                                                                   |
                           Lives in the address ----------|----------
Has moved
                                         |
                                         |
                At home -----------|--------------  Absent
                      |                                  Long absence
                      |
Interviewed -----|------  Refuses




The explanatory variables are sex (dichotomus), age (continuous; with
allowance for non-linearity using mkspline), rent (continuous; mean
rent of the census tract were the resident lives) and origin
(categorical: Spanish, Latin-American, Eastern European, high-income
countries, other).

There is a cluster variable: censustract

My first idea was to run logistic models with the dependent variables:
death/alive, moved/living in the address, refuses/interviewed, and
restricting the sample to those who have passed the previous
transition (i.e. excluding death individuals in the model for
moved/living in the address). There was also a multinomial model with
the categories at home/absent/long absence. Nevertheless, I've come to
the conclusion that -seqlogit- is a better alternative because it
assesses the influence of each variable on the outcome of interest
(being interviewed).

What I would type then is:

mkspline age1 30 age2 65 age3 = age
xi?: seqlogit inciden sex age1 age2 age3 rent i.origin, tree(1:2 3 4 5
6, 2:3 4 5 6, 3 4?: 5 6, 5:6) ofinterest (sex?) over(origin age1 age2
age3) levels(?) or cluster(censustract)

I've included question marks where I have problems:

1) There is a categorical independent variable. Can xi be used with -seqlogit-?

2) One step of the process has three branches instead of two (at home,
absent, long absence). Can this be modelled with -seqlogit-? Is 3 4:5
6 the correct way to define it?

3) It is difficult for me to give levels to the categories. Of course,
being interviewed is the desired outcome, but I cannot say that it is
x times more important that other outcomes, as you do in your
dissertation. On the other hand, these other outcomes would definitely
have to have the same value.

4) In this example, I try to decompose the effect of the variable sex,
but I would like to investigate the decomposition of the effects of
other explanatory variables. I suppose I only need to change the other
variables for sex in ofinterest. I’m interested in the interaction:
sex*origin and sex*age1 sex*age2 sex*age3. I’ve understood that the
options ofinterest (sex) over(origin age1 age2 age3) generate the
interaction terms.

Many thanks for your time and interest.


Angel Rodriguez-Laso


2010/1/22 Maarten buis <maartenbuis@yahoo.co.uk>:
> Thanks to Kit Baum an update of the -seqlogit- package is now
> available from SSC. The -seqlogit- package can be used to study
> a sequence of discrete events, for example some respondents
> finish high school and others not, and those who finished high
> school can either finish college or not. The -seqlogit- package
> allows one to additionally study the interelationship between
> effects of covariates on the individual transitions and the
> effects of these covariates on the final outcome (in this case
> highest achieved level of education). Moreover, it contains
> tools for doing a sensitivity analysis regarding the potential
> influence of unobserved heterogeneity. A describtion can be
> found here: <http://www.maartenbuis.nl/software/seqlogit.html>.
>
> This update contains a fix for a bug that was introduced in Stata
> 11, which made -uhdesc-, one of the tools for doing a sensitivity
> analysis, fail to work.
>
> This package can be updated by typing in Stata -ssc instal seqlogit, replace-.
>
> -- Maarten
>
> --------------------------
> 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/
>

*
*   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/