Weighting with a simple design effect

[Anton-24]

Dear SPSS Listers-

We've have been asked to create post-stratification weights based on age (in
categories) and gender counts from the U.S. census.  This has always been
fairly straight forward for us by using either the rake procedure in SPSS or
"balance" in P-Stat.

Our client now wants to incorporate a "design effect" that is simply the
"number of adults in the household" variable in our data set.  For instance,
 a household in our data set with two adults would count twice as much as a
household with only one adult.

Do we simply multiply the age/gender weight variable by the number of adults
in the household to create a final weight?  Then would we multiply this
final weight by a constant to get back down to our original sample N?

We would like the age-by-gender crosstab to match the census percentages
when the final weights are on, however,  multiplying by number in household
throws this off a bit.

Any insight is greatly appreciated.

Anton

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

[zstatman]

Look into the Rake or Rim weighting procedure

W

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
Anton
Sent: Tuesday, November 09, 2010 1:36 PM
To: [hidden email]
Subject: Weighting with a simple design effect

Dear SPSS Listers-

We've have been asked to create post-stratification weights based on age (in
categories) and gender counts from the U.S. census.  This has always been
fairly straight forward for us by using either the rake procedure in SPSS or
"balance" in P-Stat.

Our client now wants to incorporate a "design effect" that is simply the
"number of adults in the household" variable in our data set.  For instance,
a household in our data set with two adults would count twice as much as a
household with only one adult.

Do we simply multiply the age/gender weight variable by the number of adults
in the household to create a final weight?  Then would we multiply this
final weight by a constant to get back down to our original sample N?

We would like the age-by-gender crosstab to match the census percentages
when the final weights are on, however,  multiplying by number in household
throws this off a bit.

Any insight is greatly appreciated.

Anton

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command SIGNOFF SPSSX-L For a list of
commands to manage subscriptions, send the command INFO REFCARD

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

[parisec]

Question of interest:

What factors of a hospital admission make a patient more likely to be readmitted?


Dillema:

Patients bring in factors such as age, SES, sex that will be the same every single time they are admitted to the hospital. But each encounter also has unique factors associated with it such as reason for admission and severity of patient upon admission etc.

If I want unique patients in the study,  then I've got to have criteria to select a specific encounter to choose which, ultimately biases the data.  But if include the same patient more than once I believe I have violated the assumption of independence (which the analysis  will ultimately require) and could weight the dataset with patients who have patient factors that make them more likely to be admitted to begin with.

Based on the question I am asking, can the same patients be included in the data set multiple times if my question is with regard to the *encounter* rather than the patient or the definition of independence non-negotiable ;-) ?



Thanks for your thoughts
Carol

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

[Bruce Weaver]

It sounds like you want some kind  of multilevel model, with admissions nested within patients.  It also sounds like your outcome variable is readmission (yes-no), or time to readmission possibly, so you'll not find a procedure to run this model in SPSS.  I've just taken a quick look in chapter 9 of  Jos Twisk's book "Applied Multilevel Analysis", where he says that Stata, SAS and R all have procedures for multilevel logistic regression.  

At the time of writing, Twisk said that the only software for multilevel Cox regression was the General Linear Latent and Mixed Models (gllamm) procedure in Stata.  I suspect there have been other developments since then.

HTH.

Parise, Carol A. wrote

Question of interest:

What factors of a hospital admission make a patient more likely to be readmitted?


Dillema:

Patients bring in factors such as age, SES, sex that will be the same every single time they are admitted to the hospital. But each encounter also has unique factors associated with it such as reason for admission and severity of patient upon admission etc.

If I want unique patients in the study,  then I've got to have criteria to select a specific encounter to choose which, ultimately biases the data.  But if include the same patient more than once I believe I have violated the assumption of independence (which the analysis  will ultimately require) and could weight the dataset with patients who have patient factors that make them more likely to be admitted to begin with.

Based on the question I am asking, can the same patients be included in the data set multiple times if my question is with regard to the *encounter* rather than the patient or the definition of independence non-negotiable ;-) ?



Thanks for your thoughts
Carol

=====================
To manage your subscription to SPSSX-L, send a message to
[e-mail] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD