Re: st: pweight for households

[Nick Cox <njcoxstata@gmail.com>]

Daniel wrote

>  summarize weight if hhdrel==1
>
> is closer - you get the number of records and and the mean weight so you
> could multiply the two together to find your answer.

The sum is also calculated by -summarize- and accessible in r(sum).
So, you don't need to multiply. Just a small detail,

Nick

On Sat, May 14, 2011 at 1:15 PM, Daniel Feenberg <feenberg@nber.org> wrote:
>
> On Fri, 13 May 2011, Constance Kelly wrote:
>
>> I'm using the Current Population Survey (March Supplement) for a project,
>> and I need to derive an estimate of the number of households in the
>> population. There is a household population weight built into the data.  The
>> problem is that the data is structured by person, with individual household
>> identifier numbers, i.e.:
>>
>> Person I.D.     |       Household I.D.
>> 1                       |               1
>> 2                       |               1
>> 3                       |               1
>> 4                       |               2
>> 5                       |               3
>> 6                       |               3
>> etc.
>>
>> I know that if I reshape the data from long to wide, I can get the
>> population household estimate, but I'm trying to figure out if there is a
>> simper way to derive this value.
>> Any insight would be appreciated.
>
> Althought Stata experts will believe this is a trivial problem, I want to
> show that a beginner might take a while to find the solution. You
> need to know that hhdrel==1 for the household head, so all you need
> to do is get the sum of the weights of those records.
>
> First, she might try
>
>  count if hhdrel==1 [pweight=weight]
>
> but -count- doesn't take weights. Then try:
>
>  summarize if hhdrel==1 [pweight=weight]
>
> won't work because summarize doesn't take pweights
>
>  summarize if hhdrel==1 [aweight=weight]
>
> won't work because -summarize- doesn't offer a weighted count. Everything
> else is weighted, but not the count.
>
>  summarize weight if hhdrel==1
>
> is closer - you get the number of records and and the mean weight so you
> could multiply the two together to find your answer.
>
> That isn't desirable. What about -tabulate-?
>
>  tabulate hhdrel [pw=w] if hhdrel==1
>
> won't work because tabulate doesn't take pweights.
>
>  tabulate hhdrel [aw=w] if hhdrel==1
>
> gives the wrong answer.
>
>  tabulate hhdrel [iw=w] if hhdrel==1
>
> gives the right answer, but the definition of iweights isn't documented.
> Better to use:
>
>  table hhdrel [pw=w] if hhdrel==1
>
> which is documented or even
>
>  egen sum=sum(weight) if hhdrel==1
>  summarize sum
>
> or -table-, -collapse-, and possibly 7 other ways, but the lack
> of pweights on many of the commands is a mystery to me - I never have
> any other kind.

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/