Hello,
I am using SPXTREGRESS to estimate an SAR panel model with fixed effects for cross-sectional units. The specific command I am using is:
. spxtregress y x1-x3 year1-year8, fe dvarlag(W1) errorlag(W1)
The standard errors produced by the default approach (I cannot figure out what that approach is) seem to be incorrect. The reason I believe they are incorrect is that they are almost exactly equal in magnitude for several parameters, even though the parameter estimates are very different in magnitude. As an example, these are the estimates for the predictors:
free_mr_count | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-----------------+----------------------------------------------------------------
y |
x1 | 40.07524 22.03305 1.82 0.069 -3.108751 83.25923
x2 | 851.2171 22.17951 38.38 0.000 807.746 894.6881
x3 | 79.93515 22.07906 3.62 0.000 36.66099 123.2093
|
year_visit |
year1 | -.1465619 .0173548 -8.45 0.000 -.1805767 -.1125471
year2 | -.050435 .0173503 -2.91 0.004 -.084441 -.0164289
year3 | .0410452 .0172945 2.37 0.018 .0071486 .0749418
year4 | .139818 .017296 8.08 0.000 .1059185 .1737175
year5 | .2124655 .0173566 12.24 0.000 .1784472 .2464839
year6 | .2425511 .0173584 13.97 0.000 .2085292 .2765729
year7 | .2832503 .0173554 16.32 0.000 .2492343 .3172662
year8 | .3415652 .017397 19.63 0.000 .3074678 .3756626
Note that the standard errors are almost all of magnitude 22 for x1-x3, and of magnitude 0.017 for year1-year8, while the coefficients are dramatically different in magnitude. I have tried models with different specifications in terms of the set of predictors, as well as inclusion of dependent variable lags or spatial error lags, and the same phenomenon happens.
My questions are as follows:
1. What is the default method being used by SPXTREGRESS to produce the standard errors?
2. How can I determine if the standard errors are correct?
3. Are there options available to compute standard errors. The VCE option seems to be unavailable with SPXTREGRESS.
Thank you
I am using SPXTREGRESS to estimate an SAR panel model with fixed effects for cross-sectional units. The specific command I am using is:
. spxtregress y x1-x3 year1-year8, fe dvarlag(W1) errorlag(W1)
The standard errors produced by the default approach (I cannot figure out what that approach is) seem to be incorrect. The reason I believe they are incorrect is that they are almost exactly equal in magnitude for several parameters, even though the parameter estimates are very different in magnitude. As an example, these are the estimates for the predictors:
free_mr_count | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-----------------+----------------------------------------------------------------
y |
x1 | 40.07524 22.03305 1.82 0.069 -3.108751 83.25923
x2 | 851.2171 22.17951 38.38 0.000 807.746 894.6881
x3 | 79.93515 22.07906 3.62 0.000 36.66099 123.2093
|
year_visit |
year1 | -.1465619 .0173548 -8.45 0.000 -.1805767 -.1125471
year2 | -.050435 .0173503 -2.91 0.004 -.084441 -.0164289
year3 | .0410452 .0172945 2.37 0.018 .0071486 .0749418
year4 | .139818 .017296 8.08 0.000 .1059185 .1737175
year5 | .2124655 .0173566 12.24 0.000 .1784472 .2464839
year6 | .2425511 .0173584 13.97 0.000 .2085292 .2765729
year7 | .2832503 .0173554 16.32 0.000 .2492343 .3172662
year8 | .3415652 .017397 19.63 0.000 .3074678 .3756626
Note that the standard errors are almost all of magnitude 22 for x1-x3, and of magnitude 0.017 for year1-year8, while the coefficients are dramatically different in magnitude. I have tried models with different specifications in terms of the set of predictors, as well as inclusion of dependent variable lags or spatial error lags, and the same phenomenon happens.
My questions are as follows:
1. What is the default method being used by SPXTREGRESS to produce the standard errors?
2. How can I determine if the standard errors are correct?
3. Are there options available to compute standard errors. The VCE option seems to be unavailable with SPXTREGRESS.
Thank you
