How to deal with panel dataset with variables in fractional form?

[Muhammad Abid Shahzad]

I am dealing with a panel dataset in which both the DV and IVs are in fractional form. i.e. 0 to 1. I came across the model called "beta regression". Can anybody please tell me about the fixed effects/panel data versions of commands for betareg? Is there any xt version of the betareg model?
Thanks

[Joro Kolev]

Most nonlinear models do not allow simple elimination of the fixed effects like in the linear case. For example there is fixed effects logit, but there is no fixed effects probit.

As for random effects panel data models, their only advantage is efficiency.

I do not think that there is neither fixed nor random effects beta regression.

[Jeff Wooldridge]

In my paper with Leslie Papke, 2008 Journal of Econometrics, we propose correlated random effects fractional probit. Joro is correct that, without a large T, fixed effects approaches should be avoided. But the correlated RE approach is simple. If the panel is unbalanced it is a bit trickier, but there are solutions that use the complete cases (as any default procedure does). Betareg is not necessary. You can use -glm- or -fracreg- and include the time averages of the covariates and cluster the standard errors.

Papke-Wooldridge(2008)

JW

[Muhammad Abid Shahzad]

Jeff Wooldridge "You can use -glm- or -fracreg- and include the time averages of the covariates and cluster the standard errors."

In my case, the dataset is of EU 27 countries. And to account for different sizes of countries I wish to take "number of total farms" as weights. How can we incorporate that into 'fracreg' model?

More information: The dataset covers data for 27 countries for 5 different years. The Y variable is the proportion of part-time farms while Xs are proportions of farms by farm size categories, farm type categories and farmer age categories. All calculated by "number of holdings". e.g, In 2005 in Austria, the proportions of farms by size was: 0.50 small farms, 0.25 medium farms and 0.25 large farms.....etc.
basically, the aim is to see the how has the structural change in agriculture affected part-time farming?

[Jeff Wooldridge]

But what is your T? How many years? What is your dependent variable? You only have to have the mean correctly specified to use fractional methods. That's where you should put your effort. Cluster-robust standard errors take care of other issues.

[Muhammad Abid Shahzad]

But what is your T? How many years? What is your dependent variable? You only have to have the mean correctly specified to use fractional methods. That's where you should put your effort. Cluster-robust standard errors take care of other issues.

The data is available for five years. And the dependent variable is the proportion of part-time farms in each of the countries in EU-27, while independent variables are the proportions of farms size, farm type and farms by age of the farm manager.
e.g., In Austria, in 2005, the proportions of farms by age was: young farmers=0.50, middle-aged=0.25, old=0.25.

[Santosh Pathak]

But what is your T? How many years? What is your dependent variable? You only have to have the mean correctly specified to use fractional methods. That's where you should put your effort. Cluster-robust standard errors take care of other issues.

Jeff Wooldridge : Dear Prof. Jeff, I am dealing with unbalanced panel data (T = 24, n = 64) to find the effect of monetary incentives on the rate of program success at the county level. My dependent variable is the rate of success (ranges from 0 to 1 and both extreme values are observed). I have seven independent variables.I tried to use a panel fractional regression approach for estimation but I am confused with its implementation. Could you please kindly provide me with some hints for moving forward? I went through this link https://www.stata.com/meeting/chicag...wooldridge.pdf but found it difficult to implement in my case. Hope to hear from you. Thanks.

[Hugo Rocha]

In my paper with Leslie Papke, 2008 Journal of Econometrics, we propose correlated random effects fractional probit. Joro is correct that, without a large T, fixed effects approaches should be avoided. But the correlated RE approach is simple. If the panel is unbalanced it is a bit trickier, but there are solutions that use the complete cases (as any default procedure does). Betareg is not necessary. You can use -glm- or -fracreg- and include the time averages of the covariates and cluster the standard errors.

Papke-Wooldridge(2008)

JW

Hi Professor Wooldridge! What would be on average an acceptable number of time periods?