Weighted counts [message #17465] |
Tue, 26 March 2019 03:04 ![Go to next message Go to next message](/theme/default/images/down.png) |
siddu123
Messages: 1 Registered: March 2019
|
Member |
|
|
I am trying to calculate weighted counts in stata for India 2014-15 MR dataset.
1)From my understanding, individual observations when multiplied with weights, should give the count of the actual population.
For example. The total percent of men who have diabetes is 2.13%. Which in the weighted count (table below) shows a value of 2396. However my understanding is that the weighted count must show some 35-million (approx) cases of men who are actually diabetic in India.
2)Also is there any problem in using scaled singleunit for svy instead of "missing"?
3) is there a difference in iweight and pweight?
These are my codes. MRdataset.
generate wgt = mv005/1000000
svyset mv021 [pweight=wgt], strata(mv022) vce(linearized) singleunit(scaled)
recode sm622a (8=0) , gen (CurrentlyDiabetic)
. svy linearized : tabulate mv102 CurrentlyDiabetic, count row percent format(%16.0g)
(running tabulate on estimation sample)
Number of strata = 2508 Number of obs = 112122
Number of PSUs = 9887 Population size = 112122
Design df = 7379
------------------------------------------------------------ -
type of |
place of | RECODE of sm622a (currently has diabetes)
residence | 0 1 Total
----------+------------------------------------------------- -
urban | 41789.450922453 1163.0585127594 42952.509435213
| 97.292222205285 2.7077777947146 100
|
rural | 67936.373593395 1233.1175029331 69169.491096328
| 98.217252312561 1.782747687439 100
|
Total | 109725.82451585 2396.1760156925 112122.00053154
| 97.862885067754 2.137114932246 100
------------------------------------------------------------ -
Key: weighted counts
row percentages
Pearson:
Uncorrected chi2(1) = 108.4122
Design-based F(1, 7379) = 23.9928 P = 0.0000
Note: Variance scaled to handle strata with a single sampling unit.
|
|
|
Re: Weighted counts [message #17480 is a reply to message #17465] |
Thu, 28 March 2019 11:35 ![Go to previous message Go to previous message](/theme/default/images/up.png) |
Bridgette-DHS
Messages: 3143 Registered: February 2013
|
Senior Member |
|
|
Following is a response from Senior DHS Stata Specialist, Tom Pullum:
The DHS weights are NOT "inflation" weights, which could be used to estimate population counts. Rather, the weights are calculated to add up to the sample size, except that they include a factor of 1000000 (one million) which simply moves the decimal place 6 places to the right. If you want to inflate to population, counts, you will have to use a multiplier. We do not recommend doing this, because of sampling error.
|
|
|