can I apply regression with percentage data in STATA?

[Adnan Khan]

I am new STATA user, recently I faced difficulties collecting data, however the data I collected are in percentage form so now I am a bit confused whether it would be wise to implement Quantile Regression on such dataset, and even If it works, will the result will be appropriate or not. please let me know what you think.

[Rich Goldstein]

h fracreg
h betareg

[Maarten Buis]

Also see https://www.maartenbuis.nl/publicati...oportions4.pdf

[Nick Cox]

There are at least two distinct set-ups here:
b
1. Percentage change which for non-negative outcomes presumably is bounded by -100% below and is unbounded above.

2. Percentages which are bounded by 0 and 100% and are identical to proportions given an understanding that % is just notation for /100.

I am guessing at #2 as I think are Rich Goldstein and Maarten Buis.

Two considerations lead the field.

1. Linear variation is often implausible as values beyond the limits are impossible and predictions beyond the limits should seem absurd. Plausibke models start with sigmoid curves.

2. The variability of proportions is different from other counted or measured outcomes, as seen by the fact that means of 0 and 1 are only attainable if all values are 0 and 1 respectively, and so variance is 0 at either limit.

[Romano Piras]

Dear All,
I'm working with a dependent variable (the share of self-employed workers over total employment), which is a proportion strictly bounded between 0 and 1. My model looks like:

y_it = a0 + a1 * x1_it + a2 * x2_it + a3 * (x1_it * x2_it) + other covariates + year_dummies + country_dummies + error_it.

where x1 is the unemployment rate, and x2 is an index (0<index<100) of migrants' ethnic concentration.

My dataset is an unbalanced panel of data covering a variable number of EU countries over a span of approximately 15 years, with gaps in some years. I'm planning to use fracreg or betareg.

1) Is it correct to include year_dummies and country_dummies when using fracreg or betareg? I mean, can I simply add these dummies into a regression using fracreg or betareg?
2) Since I have an interaction between x1 and x2, is it possible to use fracreg or betareg?

Best,
Romano

[Clyde Schechter]

Yes, in terms of the right hand side variables (predictors, independent variables, regressors--whatever you wish to call them) both interactions and indicator variables ("dummies") can be used. As with most other regressions, there really are no limitations on the types of variables that can appear on the right hand side.

[Jeff Wooldridge]

If the data are at the country level then putting in country and year dummies causes an incidental parameters problem. It seems like T is never more than 15, and how many EU countries do you have? Around 25? Nonlinear models typically require a large N for the inference to be reliable. You’d still have an IP problem, though. With larger N I’d use correlated random effects. If you have N around 25, T around 15 I’d use linear FE. Might these data be at the industry level?

[Romano Piras]

Many thanks to Clyde Schechter and Jeff Wooldige,

The maximum number of countries is 18, and the longest time period is 15 years. Data is not at the industry level. Based on #7, I believe the best option would be to use a linear FE model. Unfortunately, I also have outliers, so I am considering applying robust regression (rreg) to account for these outliers. Would it be a good approach?