Category: Stata/Econometrics


Why not LPM instead of probit or logit?

An interesting post from a blog held by the World Bank try to give  valid answer as to why LPM could be adopted.

This is a very interesting post, because it might be of help to economists who are stuck with referees who insist on the necessity of changing the LPM into a non-linear model.

I came across this post, and I decided to post it here, because I have had a similar discussion with my supervisors (although I was the one insisting for the LPM), and I feel like in the future I will have many other discussion about the same topic.

I had to find out what observations in my dataset were such that the individual i was a person instead of a firm, this in formation was represented by individual’s i name in a string variable…how could I do that in a not time-consuming way?…Using:

tab id if regexem(xxx)

where xxx is a condition, which can be simply a part of the name. For instance, many Italian firms’ names end with s.p.a. (= joint stock company); in order to find which individuals are firms all the firms of this kind:

tab id if regexem(individual, “S.p.a”)

where “individual” is the name of the var and “S.p.a.” is the portion of the string variable.

To make have a clearer idea of what I am talking about, you can visit the following web-page: http://www.ats.ucla.edu/stat/stata/faq/regex.htm

 

If you have my same fundamental question, I suggest to read the article in scienceblogs.com

http://scienceblogs.com/deltoid/2003/09/10/cluster/

It is actually one of the best illustrations I have ever read!

LaTex, should I even say something about it?

LaTex online manual

R-project

It does not need any introduction to economists/statisticians/etc… however, I find necessary to spend some words and space on it too, since in this blog I have been widely speaking of everything an economist should have in his backpack (LaTex, and Stata, but also WXmaxima and Graph -for those who, like me, would love to get some help when doing TA in math for economist)

http://en.wikipedia.org/wiki/R_(programming_language)

http://www.r-project.org/

http://en.wikibooks.org/wiki/R

This is a very useful package for R, which can help you while you are still a beginner:

http://socserv.mcmaster.ca/jfox/Misc/Rcmdr/

how to install it?

http://socserv.mcmaster.ca/jfox/Misc/Rcmdr/installation-notes.html

but what is this?

John Fox (2005),  The R Commander: A Basic-Statistics Graphical User Interface to R, Journal of Statistical Software, 14, No. 9.

Odds ratios

In this period I am working on the second paper for my PhD thesis.

Because of my strategy and what I am studying, I am using two logit models with an interaction term (dummy-by-dummy and dummy-by-continuous). It seems like the logit regression can give an interpretational advantage when you include interaction terms in your binary model.

Indeed, in case of binary models, the interaction effects are often misinterpreted:

Computing interaction effects and standard errors in logit and probit models

For a general overview about the logit model, I suggest to quickly read the following:

important special cases of the logistic model

Inappropriate Interpretation of the Odds Ratio

More specifically, for the correct interpretation of interaction effects, I warmly suggest to read the following.

Interpretation of interactions in non-linear models – m b

To me, this fourth article has been mind-blowing. However, the interpretation of my model is somehow complicated by the presence of a third variable, i.e., I want to compare  logit y x1 x2 x1x2 with logit y x1 x2 x1x2 x3.

(The material I have uploaded is freely downloadable on the internet, if someone has some claim then I will change this post)

very useful link for Stata users (like often, from UCLA)

http://www.ats.ucla.edu/stat/stata/modules/dates.htm

Moreover, someone may need to cluster the SE over dates, like month and year; in that case a second step is needed.

http://www.stata.com/statalist/archive/2009-08/msg01367.html

 

Useful link where is well explained the difference between interaction term and interaction effect, with non linear models. Bibliography is included in the first page. You can also find .do for the implementation of the different why to compute the marginal effect of the interaction term (i.e., the interaction effect).

http://www.crmportals.com/crmnews/Interaction%20term%20vs.%20interaction%20effect%20in%20logit%20and%20probit%20models.pdf

http://www3.unibo.it/medspo/default.htm?http://www3.unibo.it/medspo/medicus/testita’.htm~due

ripasso di nozioni base per il test di significatività.

http://sites.roosevelt.edu/sziliak/w-s-gosset-aka-student/

for those like me who are a bit skeptic about the test for statistical significance….we are not alone.

however, the scholar who manage the web page I have linked is quite extremist in his thought, and I share only partially his view for the topic