Which model should I choose for ratios as an dependent variable?

[Eason Xu]

Hi all,
I hope to take ratios (the number of newly established firms in one city/ total number of newly established firms in all cities) as dependent variables. And my independent variables are also ratios. The overall data is a panel data. As the ratio ranges from 0 to 1, I wonder the result of OLS will be hard to interpret,in fact, the coefficient is over 1. I did some search but find the models I should choose varies. And it really confused me. Here are some questions I got:
1.logit/probit： as far as I understand, both these models require the dependent variable to be either 1 or 0.But my variable is continuous, ranging from 0 to 1.How can I possibly use these two models, and how should I interpret the coefficients?
2.tobit; I understand the model is suitable for data with "truncation", some research did use this model. But I think my variable doesn't have this problem, because no data is missing. I wonder whether I can use this model?
3. Above 3 models are the most frequently mentioned. I wonder whether there are any other options for my data?
4.I just find other two models, the command in Stata is fracreg -- Fractional response regression and betareg -- Beta regression, are these two models suitable?
Thanks I will be very grateful if anyone can give some suggestions!

[Jeff Wooldridge]

Using fracreg logit or fracreg probit is your best best. Cluster standard errors to allow serial correlation. Include time dummies. Maybe use correlated random effects.

[Eason Xu]

Using fracreg logit or fracreg probit is your best best. Cluster standard errors to allow serial correlation. Include time dummies. Maybe use correlated random effects.

Thanks a lot !

[Eason Xu]

Hi all
I have another question. I take another ratio as a new dependent variable. As the data is in panel form, the sum of the ratios in each year equals to 1. And I learned that the command fracreg requires the varibales to be independent, for example, market share cannot be an dependent variable. So I wonder whether I can still use this command, or should I try another model? And I found the fixed effect of individuals most are not significant. So I only use time dummies.

[Eason Xu]

Using fracreg logit or fracreg probit is your best best. Cluster standard errors to allow serial correlation. Include time dummies. Maybe use correlated random effects.

Professor Wooldridge: I have just encountered another question. I take another ratio as a new dependent variable. As the data is in panel form, the sum of the ratios in each year equals to 1. And I learned that the command fracreg requires the varibales to be independent, for example, market share cannot be an dependent variable. So I wonder whether I can still use this command, or should I try another model? And I found the fixed effect of individuals most are not significant. So I only use time dummies.