Bookmark and Share

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

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

Re: st: XTLOGIT: Predicted Diminishing Returns

From   Maarten Buis <>
Subject   Re: st: XTLOGIT: Predicted Diminishing Returns
Date   Tue, 17 May 2011 19:06:18 +0200

On Tue, May 17, 2011 at 6:37 PM, Dennis Kramer wrote:
> I am using a --xtlogit-- model to predict the probability of passing
> an assessment based on student and teacher absences. I am using the
> following commands:
> xi: xtlogit Test_Score STD_Absent Tch_Absent STD_TCH_Absence
> i.Course_Grade male i.race frl lep migrant gifted CLASS_MEAN i.year,
> or
> However, I want to get the predicted probabilities of passing for each
> individual absence value. Meaning, I want to look at the probability
> of passing the test with missing one, two, three, etc days.
> Theoretically there must be some diminishing impact on passing as
> going from 1 to 2days absent might be more influential then going from
> 15 to 16 days.

I can see two solutions:

1) Treat  STD_Absent and Tch_Absent as categorical variables, that is
include them as  i.STD_Absent i.Tch_Absent. This is a literary
interpretation of your question. You will get lots of coefficients,
and you may have to combine some of the rarer "categories", e.g. 1, 2,
3, 4-5, 5-10, etc. in order to get meaningful (or even converging)

2) Enter  STD_Absent and Tch_Absent as non-linear functions. Entering
the logarithm of  STD_Absent and Tch_Absent would make sense. I
actually like linear splines for this thing, as they are easy to
interpret (*), allow for decreasing effects but do not force it, and
they are nested in a model with simple linear effects. See -help

Hope this helps,

(*), see for example:

Maarten L. Buis
Institut fuer Soziologie
Universitaet Tuebingen
Wilhelmstrasse 36
72074 Tuebingen
*   For searches and help try:

© Copyright 1996–2018 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   Site index