Direction of change in panel data model.

[Resuf Ahmed]

Hello everyone!

I am very new to STATA. And I tried to make this question as clear as I can by using "dataex" but my data is very big around 5 million observation, so I am not sure how to do this. I will try to describe my question as efficiently as I can:

I have a panel of around 500,000 villages for 10(2004-2013) years so in total I have 5 million observation. Here is the description of variable

1. I have a y-variable of interest
2. There is a cut-off date 2009 when there is exogenous variation because of which there could be 4 ways in which the status of a village may change.

• Previously no treatment, and after 2009 no treatment as well (dummy M1)
• Previously no treatment, and after 2009 treatment (dummy M2)
• Previously treatment, and after 2009 no treatment (dummy M4)
• Previously treatment, and after 2009 treatment as well (dummy M5)

3. I create a variable call "Treatment"= 1 (0, otherwise) when a village is treated either after or before 2009 or both.
4. I create a variable call "Control"= 1 (0, otherwise) when a village is not treated either after or before 2009 or both.
5. Key – unique identifier for the village
7. Year – time variable

I run the following code to to see the effect on y when the status of village changes from control to treated, I run the following command:

reghdfe y Treatment , absorb(i.key , i.year)

It gives me the following output (SC is treated here):

This looks fine, To check whether the direction (when Treatment become Control or Control become Treatment) matter or not: I estimate the following (without clustered S.E.):

reghdfe avg_light SC##M2 SC##M4, absorb(i.key i.year)

And I get the following :

And here most of variable and interaction is omitted and I dont understand why. How can I solve this issue ?

[Phil Bromiley]

You didn't get a quick answer. You will increase your chances of useful answer by following the FAQ on asking questions-provide Stata code in code delimiters, readable Stata output, and sample data using dataex. Generally it is best to avoid pictures that they are often hard to read.

Your posting is far too long and complicated. Try to simplify so that you have a simple question that illustrate your problem.

You need to check whether the omitted variables vary within your panels. What it suggests in the output is that they simply aren't varying and so are collinear with the fixed effects. Remember, any variable that doesn't vary within panels cannot be estimated with the fixed effects estimator.