Initial Values Panel Data Stata 12

[Dadan Kardiana]

Dear All,

I have a panel data set 1997-2011 (continous variables only) and would like to set an initial value of variable, for example here initial value of road infrastructure. What I understand is that I type this code:

sort country year
bys country: gen road_ini= roadinfra[1]

and the result is that, for each year (from 1997 to 2011), one country has the same initial values of road infrastructure, which is the value of that of 1997. Which seems logic to me.

But when I see an example of panel data on growth regression, initial value gdp for the same country differs for every year. In my understanding, it should be the case that for one country, the initial value of gdp is the same for every year since it refers to the value of that of the first year of observation.

Could someone explain me this ?

Thanks

[Nick Cox]

But when I see an example of panel data on growth regression,

We can't see this example. But code similar to the code fragment you give should indeed give similar results. Whether it's GDP, road infrastructure, or sales of hula hoops, the logic is the same.

[Dadan Kardiana]

That's what I thought. I don't know which code of initial values that he/she applied for initial GDP in that example. However, if my command is right, i'll stick on it.
Thanks

[Dadan Kardiana]

However, I noticed that when I use initial value of infrastructure variable as the code mentioned above:

sort country year
bys country: gen road_ini= roadinfra[1]

and try to run the panel fixed effects, it is directly omitted. How can I avoid this? and how to make inclusion of initial values of any variable not to be omitted in fixed effect panel data?

Thanks

[Stephen Jenkins]

I don't think your wish can be granted. With a continuous outcome variable, the FE (within) estimator works in effect using differences over time for each subject in the values of variables. The longitudinal difference between 2 variables that are constant is zero, so the predictor is dropped. Put another way, the effect on the outcome of that predictor cannot be identified using the FE estimator. If you want to estimate its effect, you have to use a different estimator, e.g. a RE one, and pay the price of the additional assumptions that come with that.

[Dadan Kardiana]

Stephen thanks for your answer.

Then my question is, why in many articles there is an estimation with fixed effect panel, including initial value of GDP as controls, and it is not omitted. Or, how should I create an initial GDP control in fixed effect estimation ?

I thought I can do the same thing by making an initial value of infrastructure variables.

[Stephen Jenkins]

I don't know this literature very well, and can't really help you more at this stage. However, I would recommend that you post full references (author, date, title, journal/book) to the articles that apparently do what you would like to do. Also mention the Stata commands (if any) that you have been looking at in this connection. Showing example Stata code (exactly what you typed) and the resulting output may also be helpful. Without such information (recommended in the FAQ), people can only guess at what you want -- not every reader is a specialist in your field or expert at telepathy -- and so most won't bother.