Lee Sieswerda <[email protected]> asks about the contents of -e(b)- and
-e(V)- after -svytab-:
> Its late in the day, so maybe I'm just tired, but -svytab- appears to be
> acting strangely. If you run -svytab- with no options, a table is printed
> with the cell proportions, and a corresponding vector e(b) is saved. If you
> run -svytab- with the -col- option it prints the column proportions, but
> e(b) still contains the regular cell proportions. This may not seem
> exceedingly strange since it is easy enough to derive the column proportions
> from the cell proportions for display purposes. However, if, in addition to
> the -col- option, you also specify the -se- option (which prints the
> standard errors in the table beneath the proportion estimates), then e(b)
> contains the column proportions not the cell proportions. Is it just me, or
> is that a little counter-intuitive? Wouldn't it be better to always return
> the column proportions in e(b) when the -col- option is specified rather
> than making it dependent on also specifying the -se- option? Also,
> everything that I just said about the estimate vector e(b) applies equally
> to the variance matrix e(V), which is even more sinister because its harder
> to spot an error in the variance than in the estimate itself.
-svytab- saves in e(setype) the type of variance (thus standard errors) that
it returns in e(V). The default is -cell- proportions, unless you use the
-se- option.
If you want to work with the variances from the -column- proportions, then use
the -column- and -se- options. Note that -e(V)- is the (co)-variance matrix
of -e(b)-, thus if you are asking for the -column- proportion variances to be
returned in -e(V)-, you are also asking for the -column- proportions
themselves to be placed in -e(b)-.
It is the variance (SE) calculations that specify what is put in -e(b)- and
-e(V)- for -svytab-.
***** BEGIN
. use auto
. svytab for rep
pweight: <none> Number of obs = 69
Strata: <one> Number of strata = 1
PSU: <observations> Number of PSUs = 69
Population size = 69
----------------------------------------------------
| Repair Record 1978
Car type | 1 2 3 4 5 Total
----------+-----------------------------------------
Domestic | .029 .1159 .3913 .1304 .029 .6957
Foreign | 0 0 .0435 .1304 .1304 .3043
|
Total | .029 .1159 .4348 .2609 .1594 1
----------------------------------------------------
Key: cell proportions
Pearson:
Uncorrected chi2(4) = 27.2640
Design-based F(4.00, 271.76) = 6.8334 P = 0.0000
. di e(setype)
cell
. mat li e(b)
e(b)[1,10]
p11 p12 p13 p14 p15 p21
y1 .02898551 .11594203 .39130435 .13043478 .02898551 0
p22 p23 p24 p25
y1 0 .04347826 .13043478 .13043478
. svytab for rep, col
pweight: <none> Number of obs = 69
Strata: <one> Number of strata = 1
PSU: <observations> Number of PSUs = 69
Population size = 69
----------------------------------------------------
| Repair Record 1978
Car type | 1 2 3 4 5 Total
----------+-----------------------------------------
Domestic | 1 1 .9 .5 .1818 .6957
Foreign | 0 0 .1 .5 .8182 .3043
|
Total | 1 1 1 1 1 1
----------------------------------------------------
Key: column proportions
Pearson:
Uncorrected chi2(4) = 27.2640
Design-based F(4.00, 271.76) = 6.8334 P = 0.0000
. di e(setype)
cell
. mat li e(b)
e(b)[1,10]
p11 p12 p13 p14 p15 p21
y1 .02898551 .11594203 .39130435 .13043478 .02898551 0
p22 p23 p24 p25
y1 0 .04347826 .13043478 .13043478
. svytab for rep, col se
pweight: <none> Number of obs = 69
Strata: <one> Number of strata = 1
PSU: <observations> Number of PSUs = 69
Population size = 69
----------------------------------------------------------------
| Repair Record 1978
Car type | 1 2 3 4 5 Total
----------+-----------------------------------------------------
Domestic | 1 1 .9 .5 .1818 .6957
| (0) (0) (.0552) (.1187) (.1171) (.0558)
|
Foreign | 0 0 .1 .5 .8182 .3043
| (0) (0) (.0552) (.1187) (.1171) (.0558)
|
Total | 1 1 1 1 1 1
|
----------------------------------------------------------------
Key: column proportions
(standard errors of column proportions)
Pearson:
Uncorrected chi2(4) = 27.2640
Design-based F(4.00, 271.76) = 6.8334 P = 0.0000
. di e(setype)
column
. mat li e(b)
e(b)[1,10]
p11 p12 p13 p14 p15 p21
y1 1 1 .9 .5 .18181818 0
p22 p23 p24 p25
y1 0 .1 .5 .81818182
. qui log close
***** END
--Jeff
[email protected]
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
