can a categorical variable be the dependent variable in a regression

[Rushini Goonatilake]

Hello everyone.
I am working in my RAE and would like to figure out the factors that affect labour productivity. Therefore my dependent variable is 'labour productivity'. This variable is categorical and is in a way such that 1- very good levels of labour productivity 2- quite good, 3-neither good nor bad 4- quite bad and 5-very bad. Can I use this as my dependent variable or do I have to make it continuous...and if so is this possible?

[Nick Cox]

Read about models for ordinal (ordered, graded) responses in any categorical data analysis text. They tend to be covered in intermediate econometrics texts too.

(No idea what an RAE is. Please remember that this forum is international and interdisciplinary.)

[Charlie Joyez]

Hello,
You can use a discrete (categorical) variable as a depedant variable, but you have to specify a non-linear model (not to pretend that the value "2" is equal to the double of the value "1", as the numbers here are only codes for the categories.)
My advice would be to consider an ordered probit (or logit) model, but these are not the only possibilities.
However, before launching these regression, you should read the specifities of logistic regressions, and try to understand how it works and how to interpret their results, as they raise some new issues compared to linear regressions.
Take a look on Cameron and Trivedi's book "microeconometrics using stata", especially the non-linear models chapter, which is very clear to begin.

Hope this helped,
Charlie

[Rushini Goonatilake]

thank you! I will look into this. RAE= Research in Applied Economics

[Aspen Chen]

Two things to add to the nice recommendation. First, depending on the distribution of your dependent variable, there might be a need to collapse some categories even if you use an ordinal probit or logit model.

As for materials, there is also Regression Models for Categorical and Limited Dependent Variables by J. Scott Long and Jeremy Freese, which is a useful introductory text dedicated to non-linear regressions. Although I have not read it, they also published with Stata Press another book titled Regression Models for Categorical Dependent Variables Using Stata, which offers more technical advice on applying the models in Stata.

[Carlo Lazzaro]

Rushini:
I second Aspen's references and would strongly recommend you the new edition of J. Scott Long and Jeremy Freese Regression Models for Categorical Dependent Variables Using Stata, 3rd edition. College Station, TX: Stata Press, 2014 (http://www.stata.com/bookstore/regre...ent-variables/).

[Rushini Goonatilake]

These books have been very helpful and just what I needed for the project. I understand that one command to find the marginal effects for a ordered probit model is 'mchange income, brief' , but when I do this stata gives me 'unrecognized command: mchange' why is this please?

[Nick Cox]

You need to install mchange before you can use it. It's user-written.

search mchange

tells you where it can be downloaded from and gives you a link to do that.