I just updated several of my programs on SSC: -ineqdeco-, -ineqdec0-, -ineqfac-, -ineqrbd-, -povdeco, and -sumdist-. (Thanks as ever to Kit Baum for his SSC work.)
There are cosmetic changes, e.g. to refer to my current email address in the help files. I now also allow all types of weight to be used.
More substantively, one can now derive -svy bootstrap- estimates, i.e. with SEs, for all the saved results produced by those programs, drawing on Philippe Van Kerm’s -rhsbsample- (SSC; Repeated half-sample bootstrap sampling). For this I needed to be able to refer to pweights. A common issue when bootstrapping weighted estimates is how to deal with the weights appropriately; PVK’s program addresses that.
Some example calculations are shown below with one year of UK income data (from Goodman, A. and S. Webb, For Richer, for Poorer: The Changing Distribution of Income in the United Kingdom, 1961–91, Commentary 42, Institute for Fiscal Studies, London, 1994. Available to registered users from the UK Data Service.)
Reassuringly it turns out that the bootstrapped SEs are very similar to the SEs produced using asymptotic/linearised methods (See my -svylorenz- and -svygei_svyatk- on SSC for Lorenz ordinates, Gini, GE and Atkinson inequality indices.). Examples below
Stephen
.
There are cosmetic changes, e.g. to refer to my current email address in the help files. I now also allow all types of weight to be used.
More substantively, one can now derive -svy bootstrap- estimates, i.e. with SEs, for all the saved results produced by those programs, drawing on Philippe Van Kerm’s -rhsbsample- (SSC; Repeated half-sample bootstrap sampling). For this I needed to be able to refer to pweights. A common issue when bootstrapping weighted estimates is how to deal with the weights appropriately; PVK’s program addresses that.
Some example calculations are shown below with one year of UK income data (from Goodman, A. and S. Webb, For Richer, for Poorer: The Changing Distribution of Income in the United Kingdom, 1961–91, Commentary 42, Institute for Fiscal Studies, London, 1994. Available to registered users from the UK Data Service.)
Reassuringly it turns out that the bootstrapped SEs are very similar to the SEs produced using asymptotic/linearised methods (See my -svylorenz- and -svygei_svyatk- on SSC for Lorenz ordinates, Gini, GE and Atkinson inequality indices.). Examples below
Stephen
.
Code:
. local R = 500
. forvalues i = 1/`R' {
2. qui gen rhsbrw`i' = .
3. qui rhsbsample, weight(rhsbrw`i')
4. qui replace rhsbrw`i' = rhsbrw`i' * wgt
5. }
.
. svyset [pw = wgt], vce(bootstrap) bsrweight(rhsbrw*) mse
pweight: wgt
VCE: bootstrap
MSE: on
bsrweight: rhsbrw1 .. rhsbrw500
Single unit: missing
Strata 1: <one>
SU 1: <observations>
FPC 1: <zero>
.
. svy bootstrap gini = (r(gini)) hscv = (r(ge2)) ///
> , nodots ///
> : ineqdec0 HHinc
Bootstrap results Number of obs = 7,635
Population size = 56,382.626
Replications = 500
command: ineqdec0 HHinc
gini: r(gini)
hscv: r(ge2)
------------------------------------------------------------------------------
| Observed Bstrap *
| Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
gini | .3447334 .0061432 56.12 0.000 .332693 .3567737
hscv | .3736323 .0301236 12.40 0.000 .3145911 .4326736
------------------------------------------------------------------------------
.
. svy bootstrap gini = (r(gini)) hscv = (r(ge2)) ge0 = (r(ge0)) ///
> ge0B = (r(within_ge0)) ge0W = (r(between_ge0)) ///
> ge0_0 = (r(ge0_0)) ge0_1 = (r(ge0_1)) ge0_2 = (r(ge0_2)) /
> //
> ge0_3 = (r(ge0_3)) ge0_4 = (r(ge0_4)) ge0_5 = (r(ge0_5)) /
> //
> , nodots ///
> : ineqdeco HHinc, by(famtype)
Bootstrap results Number of obs = 7,635
Population size = 56,382.626
Replications = 500
command: ineqdeco HHinc, by(famtype)
gini: r(gini)
hscv: r(ge2)
ge0: r(ge0)
ge0B: r(within_ge0)
ge0W: r(between_ge0)
ge0_0: r(ge0_0)
ge0_1: r(ge0_1)
ge0_2: r(ge0_2)
ge0_3: r(ge0_3)
ge0_4: r(ge0_4)
ge0_5: r(ge0_5)
------------------------------------------------------------------------------
| Observed Bstrap *
| Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
gini | .3447334 .0061432 56.12 0.000 .332693 .3567737
hscv | .3736323 .0301236 12.40 0.000 .3145911 .4326736
ge0 | .2099721 .0079288 26.48 0.000 .1944319 .2255122
ge0B | .1878084 .0074155 25.33 0.000 .1732743 .2023425
ge0W | .0221637 .0022436 9.88 0.000 .0177663 .0265611
ge0_0 | .2302079 .0167659 13.73 0.000 .1973473 .2630684
ge0_1 | .205637 .0149452 13.76 0.000 .176345 .234929
ge0_2 | .192224 .0132923 14.46 0.000 .1661717 .2182764
ge0_3 | .1349613 .0261765 5.16 0.000 .0836563 .1862664
ge0_4 | .105247 .0073959 14.23 0.000 .0907513 .1197428
ge0_5 | .1622878 .0175463 9.25 0.000 .1278976 .196678
------------------------------------------------------------------------------
.
. svy bootstrap cush1 = (r(cush1)) cush2 = (r(cush2)) cush3 = (r(cush3)) ///
> cush4 = (r(cush4)) cush5 = (r(cush5)) cush6 = (r(cush6)) ///
> cush7 = (r(cush7)) cush8 = (r(cush8)) cush9 = (r(cush9)) ///
> , nodots ///
> : sumdist HHinc
Bootstrap results Number of obs = 7,635
Population size = 56,382.626
Replications = 500
command: sumdist HHinc
cush1: r(cush1)
cush2: r(cush2)
cush3: r(cush3)
cush4: r(cush4)
cush5: r(cush5)
cush6: r(cush6)
cush7: r(cush7)
cush8: r(cush8)
cush9: r(cush9)
------------------------------------------------------------------------------
| Observed Bstrap *
| Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
cush1 | .0304713 .0005963 51.10 0.000 .0293025 .0316401
cush2 | .0755657 .0010748 70.31 0.000 .0734592 .0776722
cush3 | .1300278 .0016566 78.49 0.000 .126781 .1332747
cush4 | .1950421 .0023288 83.75 0.000 .1904777 .1996065
cush5 | .2709849 .0030788 88.02 0.000 .2649505 .2770192
cush6 | .3587181 .0038956 92.08 0.000 .3510828 .3663534
cush7 | .4598146 .0047223 97.37 0.000 .4505591 .4690702
cush8 | .5786008 .0055712 103.86 0.000 .5676814 .5895203
cush9 | .7250403 .0063702 113.82 0.000 .7125549 .7375257
------------------------------------------------------------------------------
.
. *** linearised SEs for comparison ****
. svyset [pw = wgt]
pweight: wgt
VCE: linearized
Single unit: missing
Strata 1: <one>
SU 1: <observations>
FPC 1: <zero>
.
. svylorenz HHinc
Quantile group shares, cumulative shares (Lorenz ordinates),
generalized Lorenz ordinates, and Gini
Number of strata = 1 Number of obs = 7635
Number of PSUs = 7635 Population size = 56382.63
Design df = 7634
---------------------------------------------------------------------------
Group | Linearized
share | Estimate Std. Err. z P>|z| [95% Conf. Interval]
---------+-----------------------------------------------------------------
1 | 0.030471 0.000564 54.047 0.000 .0293663 .0315763
2 | 0.045094 0.000568 79.433 0.000 .0439818 .0462071
3 | 0.054462 0.000667 81.677 0.000 .0531552 .055769
4 | 0.065014 0.000754 86.203 0.000 .0635361 .0664925
5 | 0.075943 0.000825 92.065 0.000 .074326 .0775595
6 | 0.087733 0.000902 97.221 0.000 .0859646 .089502
7 | 0.101097 0.000988 102.361 0.000 .0991608 .103032
8 | 0.118786 0.001107 107.319 0.000 .116617 .120956
9 | 0.146439 0.001285 113.952 0.000 .143921 .148958
10 | 0.274960 0.006213 44.256 0.000 .262783 .287137
---------+-----------------------------------------------------------------
Cumul. |
share |
1 | 0.030471 0.000564 54.047 0.000 .0293663 .0315763
2 | 0.075566 0.001043 72.447 0.000 .0735214 .07761
3 | 0.130028 0.001622 80.179 0.000 .126849 .133206
4 | 0.195042 0.002278 85.609 0.000 .190577 .199507
5 | 0.270985 0.002994 90.514 0.000 .265117 .276853
6 | 0.358718 0.003761 95.376 0.000 .351347 .36609
7 | 0.459815 0.004580 100.393 0.000 .450838 .468792
8 | 0.578601 0.005438 106.402 0.000 .567943 .589259
9 | 0.725040 0.006213 116.700 0.000 .712863 .737217
10 | 1.000000
---------+-----------------------------------------------------------------
Gen. |
Lorenz |
1 | 10.689 0.178 60.018 0.000 10.340 11.038
2 | 26.507 0.290 91.305 0.000 25.938 27.076
3 | 45.611 0.441 103.345 0.000 44.746 46.476
4 | 68.416 0.622 109.975 0.000 67.197 69.635
5 | 95.055 0.818 116.245 0.000 93.452 96.658
6 | 125.830 1.029 122.269 0.000 123.813 127.847
7 | 161.292 1.260 128.039 0.000 158.823 163.761
8 | 202.959 1.537 132.011 0.000 199.946 205.972
9 | 254.327 1.927 131.949 0.000 250.549 258.104
10 | 350.776 4.246 82.621 0.000 342.455 359.097
---------+-----------------------------------------------------------------
Gini | 0.3447334 .00595649 57.875 0.000 .3330589 .3564079
---------------------------------------------------------------------------
.
. svygei HHinc
Complex survey estimates of Generalized Entropy inequality indices
pweight: wgt Number of obs = 7635
Strata: <one> Number of strata = 1
PSU: <observations> Number of PSUs = 7635
Population size = 56382.626
---------------------------------------------------------------------------
Index | Estimate Std. Err. z P>|z| [95% Conf. Interval]
---------+-----------------------------------------------------------------
GE(-1) | 1.01961 .29284732 3.48 0.000 .4456402 1.593581
MLD | .2099721 .00755174 27.80 0.000 .1951709 .2247732
Theil | .2307089 .01121549 20.57 0.000 .208727 .2526909
GE(2) | .3736323 .02940012 12.71 0.000 .3160092 .4312555
GE(3) | .9688198 .12483202 7.76 0.000 .7241536 1.213486
---------------------------------------------------------------------------
