Re: st: Treatment of missing values in surveys in Stata (subpop)

[Steve Samuels <sjsamuels@gmail.com>]

Ángel:

The theory of subpopulation corrections does not apply to non-response.

A subpopulation is a subset of the population tht can be defined in
advance: (e.g. males, ages 30-40, living in rural areas). The number
selected by a sample will be random. For example, suppose a population
of N members contains a subpopulation of M members. A SRS of size n
taken. You should be able work out the exact probability that the sample
will contain exactly k members of the subpopulation. The theory of the
subpopulation correction is an extension of this, and can be found in
any good text.

In contrast, "responder" is not a characteristic, like gender, that is
known in advance. It is defined only in relation to the particular sample
design and protocol. For identical designs, better protocols can
increase response rates. Thus, sampling theory alone cannot
describe the numbers of  responders and, consequently, the
subpopulation correction is not applicable.


Steve
sjsamuels@gmail.com

On Mar 8, 2013, at 2:37 PM, Ángel Rodríguez Laso wrote:

Dear Statalisters,

I have found two recommended procedures for dealing with individuals
with missing items ('normal' missing answers like 'DK/DA' or equipment
failure) when analysing surveys with Stata:

1) One is based on the recommendation that, unless there is a very
strong reason to do otherwise, whenever you analyse a group of
individuals in a survey with Stata, you have to use subpop. (See for
example: http://www.stata.com/meeting/mexico10/mex10sug_canette.pdf).
Under this perspective, those with valid values would be a
subpopulation. From my point of view, this means that in order to
prevent Stata from dropping them from the calculation of standard
errors, missing codes (".") should be recoded to a numerical value
(like 999) and then a command issued this way:

svy, subpop(if var<999): command var

2) Nevertheless, most of the information I've read does not make any
statement about this, what implicitly means that missing codes don't
need to be recoded. I've even found this piece of advice
(http://www.stata.com/statalist/archive/2012-09/msg01028.html): 'I've
never seen a recommendation to consider observations with non-missing
values as a subpopulation'


I wonder if anyone could throw some ligth on this topic.

Thank you very much.

Angel Rodriguez-Laso
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/


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