CTABLES and effective base weights

[Robert L]

The weighting option in Custom Tables in version 24 seemed promising, so I set up a simple data set for testing. However, it does not work as I thought it would. The chisquare test and test of column proportions give incorrect answers, even though the table itself looks just as it should. For comparison, I tried “ordinary” weighting and CROSSTABS which works fine. Is there something special about “effective base weights” used in CTABLES that makes these weights less useful for the situation I tried?

 

DATA LIST LIST /disease(F1) exposed(F1) weights(F3).

BEGIN DATA

0 0 60

0 1 240

1 0 120

1 1 80

END DATA.

 

*A standard crosstabulation for comparison.

WEIGHT BY weights.

CROSSTABS exposed BY disease

/STATISTICS=CHISQ.

 WEIGHT OFF.

 

*The corresponding table with CTABLES.

CTABLES

  /WEIGHT VARIABLE=weights

  /VLABELS VARIABLES=exposed disease DISPLAY=DEFAULT

  /TABLE exposed [C] BY disease [C][COUNT F40.0]

  /CATEGORIES VARIABLES=exposed disease ORDER=A KEY=VALUE EMPTY=EXCLUDE

  /CRITERIA CILEVEL=95

  /SIGTEST TYPE=CHISQUARE ALPHA=0.05 INCLUDEMRSETS=YES CATEGORIES=ALLVISIBLE.

 

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]

When you use weights in CROSSTABS, they are interpreted as replication weights, so you crosstab operates as if there are 500 cases.

That is a quite different calculation from effective base.  For effective  base, these are sampling weights, which is more like the way complex samples would treat these.  I would not expect the results to be the same.

You can google around for an explanation of effective base, but if you have the SPSSINC RAKE extension command installed, which you probably do, there is tutorial document saved in the SPSSINC RAKE subdirectory on your extension command search path that has an explanation and examples of effective base.

On Fri, Aug 26, 2016 at 7:48 AM, Robert Lundqvist <[hidden email]> wrote:

The weighting option in Custom Tables in version 24 seemed promising, so I set up a simple data set for testing. However, it does not work as I thought it would. The chisquare test and test of column proportions give incorrect answers, even though the table itself looks just as it should. For comparison, I tried “ordinary” weighting and CROSSTABS which works fine. Is there something special about “effective base weights” used in CTABLES that makes these weights less useful for the situation I tried?

 

DATA LIST LIST /disease(F1) exposed(F1) weights(F3).

BEGIN DATA

0 0 60

0 1 240

1 0 120

1 1 80

END DATA.

 

*A standard crosstabulation for comparison.

WEIGHT BY weights.

CROSSTABS exposed BY disease

/STATISTICS=CHISQ.

 WEIGHT OFF.

 

*The corresponding table with CTABLES.

CTABLES

  /WEIGHT VARIABLE=weights

  /VLABELS VARIABLES=exposed disease DISPLAY=DEFAULT

  /TABLE exposed [C] BY disease [C][COUNT F40.0]

  /CATEGORIES VARIABLES=exposed disease ORDER=A KEY=VALUE EMPTY=EXCLUDE

  /CRITERIA CILEVEL=95

  /SIGTEST TYPE=CHISQUARE ALPHA=0.05 INCLUDEMRSETS=YES CATEGORIES=ALLVISIBLE.

 

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