Forum: help

RE: Missing values, Sample size, Instruments, and Logical Operations [ Reply ]
By: Arne Henningsen on 2014-09-14 10:24

Dear Mohammed

Next time, please ask different questions in separate threads.

I guess that you estimate a single-equation model with different sets of coefficients for each individual, right?

1a) Yes.

1b) Missing observations (individual-year combinations) are not imputed (e.g. by values from the preceding year) but ignored.

2) Be aware of that the estimation of systems of equations with unequal numbers of observations has not been thoroughly tested yet and take a look at the "details" section in the documentation of systemfit, Additionally, you could review the source code, estimate models with different software packages and compare the results, ...

3) Why does R only include 1000 rows? Does it automagically remove rows 1001-3000?

4) You need at least as many instrumental variables (including exogenous explanatory variables) as you have explanatory variables.

5) You can set argument "data" to a subset of your data set, e.g. if your data set has the name "myData": systemfit( ..., data = myData[ myData$income > 12345, ] )

Best regards,
Arne

Missing values, Sample size, Instruments, and Logical Operations [ Reply ]
By: Mohammed Abdullah on 2014-08-31 15:27

[forum:41409]

Dear Sir,

I am estimating growth regressions applying simultaneous equation panel data models and using systemfit. When I estimated the models, I experienced some problems and I have a few questions based on these problems. I have listed these questions below. I would appreciate if you kindly answer these questions.

1. In my data set I have missing value for some years. Is it possible to estimate simultaneous equation panel models by OLS, SUR, 2SLS, 3SLS and WLS using systemfit with missing data for some years?

If I assume that value of the preceding year remains constant during the years for which data are not available, can I estimate use systemfit for the above method?

2. When I estimated the models I found the following message? What should I do?

"the estimation of systems of equations with unequal numbers of observations has not been thoroughly tested yet"

3. I am working with large dataset. Data is recorded in 3000 rows. R only includes 1000 rows in estimation. How can I handle large data set with systemfit? Please note that my sample has more than 150 countries with 20 years of data for each country. As mentioned above, the sample has also missing values for some years.

4. Can I use single variable as instrument in case of 2SLS and 3SLS?

5. Can I perform logical operations for these methods? For example, if I want to find estimates for a specific income group (subsample of countries), is that possible to find using systemfit?

Looking forward to receiving your kind comments.

With best regards,
Mohammed