[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

RE: st: Survey data settings

From   "Jackson, Theron Keith (UMSL-Student)" <[email protected]>
To   <[email protected]>
Subject   RE: st: Survey data settings
Date   Mon, 2 Nov 2009 20:41:54 -0600

Stas Kolenikov,
I appoligize, you lost me.  
So it is possible to have stata perform 3 different svyset commands in the same dataset?
Yes city=county.


From: [email protected] on behalf of Stas Kolenikov
Sent: Mon 11/2/2009 5:58 PM
To: [email protected]
Subject: Re: st: Survey data settings

On 11/2/09, Jackson, Theron Keith (UMSL-Student) <[email protected]> wrote:
> Counties are not random.  Just the populations within jails
> within the counties.  The ADAM data is set up like 37 different random samples.

OK. Then counties are the highest level strata.

> 1. Where there were only one jail in the city a single jail design
> was used that randomly selected people arrested.

Where are the cities coming from? Is city = county? Are Here, PSU = person.

> 2. For counties with a few booking agencies a stratified jail design was used.

Cross those strata with counties.

> 3. For counties with several jails a stratified cluster sample design was used

Cross those strata with counties, cluster = PSU, and you will also
have some SSUs, I imagine.

> 4. For counties that had several jails that feed into one jail a feeder
> jail design was used where arrestees were sampled only from the central
> jail but represented arrestees from each of the feeder jails.

Person = PSU; feeder jails should be analyzed as domains (-subpop-).

So as far as you tell yourself: "I want to keep this geographical unit
intact" -- that's a stratum. When you tell yourself: "I want to take a
random sample" -- that's a PSU level, and what you sample from should
be a stratum (be that the level of a county, for a smaller county, a
city, for a medium sized city, a central jail, for the feeder jail
arrangement, etc.)

You will have some funny interpretation issues. You can do any
descriptive statistics you like (percent this, percent that, median
age, average term, whatever), but if you want to do some causal
crime-and-punishment modeling, you would have to keep in mind that
both smaller and larger crimes will be underrepresented in your

Stas Kolenikov, also found at <> 
Small print: I use this email account for mailing lists only.
*   For searches and help try:


© Copyright 1996–2024 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index