Complex Samples and post-stratification

[Robert L]

In an earlier attempt to come to grips with the Complex Samples module, Jon replied that a lot of the weighting I need now can be made with the latest version of CTABLES. This seems really promising, however, the plan is to continue with some regression models so the tables will not be enough

So the question is still relevant: is it possible to use the Complex Samples module for post-stratification? Weights are requested in the setup, and I have computed such weights using a standard population. I have also found an IBM support page saying that the Complex Samples module cannot perform post-stratification. This page can't be reached now. So I would like to know if anyone can put me in the right direction. Post-stratification with the Complex Samples module, is it possible? If so, any hints for how it should be set up would be appreciated.

In case you need some more background: the goal is to analyze data from health screenings covering 30 years. The screenings have been made every fifth year since 1986. The sampling has always been done in the same fashion, stratification over sex and age group, with 250 people in each combination, making up a total of 2500 people invited in each screening. The main focus now is to compare the two counties involved with respect to different health indicators. The post-stratification comes in to compensate for changes in demography. Or it is hoped to do so anyway.

Robert

=====================
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

[Jon Peck]

The CS procedures do not handle post stratification, but here is an approximate solution.

Step 1: Run your regression model using the CS procedure.  Then run the same model with the regular REGRESSION procedure using the CS weight as the WLS weight (/REGWGT) - do not use the regular Statistics weight.  If the results are pretty close, you are in luck.  Proceed to step 2.

Step 2: Set the CS weight as the regular Statistics weight.  Then use the SPSSINC RAKE extension command to compute a new set of weights based on the weight and the control totals you want to match.  If you have a recent version of RAKE, you will also get a heatmap that shows the adjusted pattern for the weights.

Step 3: Turn off the SPSS weight and run REGRESSION with the newly adjusted weight as the WLS weight.

These results will be, of course, approximate, so if the RAKE adjustments make a large change in the weights, I would be cautious.

On Wed, Aug 24, 2016 at 12:43 AM, Robert Lundqvist <[hidden email]> wrote:

In an earlier attempt to come to grips with the Complex Samples module, Jon replied that a lot of the weighting I need now can be made with the latest version of CTABLES. This seems really promising, however, the plan is to continue with some regression models so the tables will not be enough

So the question is still relevant: is it possible to use the Complex Samples module for post-stratification? Weights are requested in the setup, and I have computed such weights using a standard population. I have also found an IBM support page saying that the Complex Samples module cannot perform post-stratification. This page can't be reached now. So I would like to know if anyone can put me in the right direction. Post-stratification with the Complex Samples module, is it possible? If so, any hints for how it should be set up would be appreciated.

In case you need some more background: the goal is to analyze data from health screenings covering 30 years. The screenings have been made every fifth year since 1986. The sampling has always been done in the same fashion, stratification over sex and age group, with 250 people in each combination, making up a total of 2500 people invited in each screening. The main focus now is to compare the two counties involved with respect to different health indicators. The post-stratification comes in to compensate for changes in demography. Or it is hoped to do so anyway.

Robert

=====================
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