I am attempting to manually bootstrap standard errors using bsample, simulate and bstat commands. I have tried doing this using the standard bootstrap procedure to no avail. See previous post:
https://www.statalist.org/forums/for...and-panel-data
I am working from two example samples constructed from Stata's 'auto' dataset:
Sample 1: cross-sectional
Sample 2: panel
The procedure I am using is as follows:
The results look promising in that the code runs and the coefficients are correct, but the bootstrapped standard errors are way off (too small).
Without bootstrap:
With bootstrap:
Does anyone have any ideas what I am doing wrong?
Thank you!
https://www.statalist.org/forums/for...and-panel-data
I am working from two example samples constructed from Stata's 'auto' dataset:
Sample 1: cross-sectional
Code:
* Example generated by -dataex-. For more info, type help dataex clear input int mpg float headroom int weight 14 4 4330 14 3.5 3900 21 3 4290 29 2.5 2110 16 4 3690 22 3.5 3180 22 2 3220 24 2 2750 19 3.5 3430 30 2 2120 18 4 3600 16 4 3600 17 4.5 3740 28 1.5 1800 21 2 2650 12 3.5 4840 12 2.5 4720 14 3.5 3830 22 3 2580 14 3.5 4060 15 3.5 3720 18 3 3370 14 3 4130 20 3.5 2830 21 4 4060 19 2 3310 19 4.5 3300 18 4 3690 19 4.5 3370 24 2 2730 16 3.5 4030 28 2 3260 34 2.5 1800 25 4 2200 26 1.5 2520 18 5 3330 18 4 3700 18 1.5 3470 19 2 3210 19 3.5 3200 19 3.5 3420 24 2 2690 17 3 2830 23 2.5 2070 25 2.5 2650 23 1.5 2370 35 2 2020 24 2.5 2280 21 2.5 2750 21 2.5 2130 25 3 2240 28 2.5 1760 30 3.5 1980 14 3.5 3420 26 3 1830 35 2.5 2050 18 2.5 2410 31 3 2200 18 2 2670 23 2.5 2160 41 3 2040 25 3 1930 25 2 1990 17 2.5 3170 end
Code:
* Example generated by -dataex-. For more info, type help dataex
clear
input float(price mpg headroom t id)
4099 22 2.5 1 1
4186.009 26.934307 1.933958 2 1
4749 17 3 1 2
4843.107 18.91313 3.3818026 2 2
3799 22 3 1 3
3770.704 24.97907 2.9321864 2 3
4816 20 4.5 1 4
4852.0933 22.037195 4.3978457 2 4
7827 15 4 1 5
7734.57 19.414215 3.253971 2 5
5788 18 4 1 6
5762.868 20.123154 4.837951 2 6
4453 26 3 1 7
4521.1724 29.6178 2.0983803 2 7
5189 20 2 1 8
5146.077 23.62131 1.3961287 2 8
10372 16 3.5 1 9
10323.16 15.383166 2.9035466 2 9
4082 19 3.5 1 10
4041.077 16.604582 2.918597 2 10
end
Code:
use sample1.dta, replace
qui reg weight mpg headroom
use sample2.dta, replace
capture drop weight_hat
predict weight_hat
xtset id t
xtreg price weight_hat, fe
matrix b = e(b)
scalar obs = r(N)
capture program drop example
program define example, eclass
use sample1.dta, replace
bsample
qui reg weight mpg headroom
use sample2.dta, replace
capture drop weight_hat
predict weight_hat
xtset id t
xtreg price weight_hat, fe
exit
end
simulate, reps(10000) seed(12345): example
bstat, stat(b) n(`obs')
estat bootstrap, all
Without bootstrap:
Code:
Fixed-effects (within) regression Number of obs = 20
Group variable: id Number of groups = 10
R-squared: Obs per group:
Within = 0.1927 min = 2
Between = 0.3717 avg = 2.0
Overall = 0.3019 max = 2
F(1,9) = 2.15
corr(u_i, Xb) = -0.5577 Prob > F = 0.1768
------------------------------------------------------------------------------
price | Coefficient Std. err. t P>|t| [95% conf. interval]
-------------+----------------------------------------------------------------
weight_hat | -.0608991 .0415549 -1.47 0.177 -.1549027 .0331046
_cons | 5712.816 139.0787 41.08 0.000 5398.199 6027.434
-------------+----------------------------------------------------------------
sigma_u | 2081.0277
sigma_e | 30.10469
rho | .99979077 (fraction of variance due to u_i)
------------------------------------------------------------------------------
F test that all u_i=0: F(9, 9) = 6584.95 Prob > F = 0.0000
Code:
Bootstrap results Number of obs = 20
Replications = 50
-------------------------------------------------------------------------------
| Observed Bootstrap Normal-based
| coefficient std. err. z P>|z| [95% conf. interval]
--------------+----------------------------------------------------------------
_b_weight_hat | -.0608991 .0107155 -5.68 0.000 -.0819011 -.039897
_b_cons | 5712.816 34.90237 163.68 0.000 5644.409 5781.224
-------------------------------------------------------------------------------
. eststo: estat bootstrap, all
Bootstrap results Number of obs = 20
Replications = 50
------------------------------------------------------------------------------
| Observed Bootstrap
| coefficient Bias std. err. [95% conf. interval]
-------------+----------------------------------------------------------------
_b_weight_~t | -.06089906 -.0004936 .01071552 -.0819011 -.039897 (N)
| -.0800628 -.040871 (P)
| -.0800628 -.0400041 (BC)
_b_cons | 5712.8164 1.755529 34.90237 5644.409 5781.224 (N)
| 5646.295 5768.709 (P)
| 5644.807 5768.709 (BC)
------------------------------------------------------------------------------
Key: N: Normal
P: Percentile
BC: Bias-corrected
Thank you!
