Discrepancies in findings within SPSS

[Janel Schuh]

Hi, Bill Ware recommended that I reach out to this listserv with a question I have about SPSS.

 

My problem:

 

When I run weighted proportions in SPSS through their traditional interface (e.g., frequencies, crosstabs), I get results (counts, %s) that don't exactly match those I get when I use the Tables add-on.

• Results are typically very close, but not exactly the same.
• In a few cases, there are substantial differences (e.g., differences of more than +/-2%).

 

I’d appreciate any explanations, resources, etc. that can help me understand why the stats differ and any insight into why SPSS would use 2 different ways of calculating the same statistics, etc.

 

I’ve heard this may have something to do with whether weighted or unweighted n sizes are used in calculating statistics, but have had trouble nailing down any citations about this topic.

 

Regards, Janel

 

Janel S. Schuh, PhD

Senior Research Scientist, Health Communication

CommunicateHealth, Inc. 

 

202.277.8088

www.communicatehealth.com 

@CommunicateHlth

 

 

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

[Art Kendall]

Please send some syntax that demonstrates the problem.

data list ...
some data

frequencies ....

Crosstabs ...

CTABLES ...

Remember you can always draft syntax by using the GUI.






-----
Art Kendall
Social Research Consultants
--
Sent from: http://spssx-discussion.1045642.n5.nabble.com/

=====================
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]

CROSSTABS offers five options for how noninteger weights are treated.  You can round or truncate the case weights or the cell counts, or you can make no adjustments.  Personally, I would choose no adjustment, but the default is to round the cell counts.  All this is based on the weight set by the WEIGHT command (or menu equivalent).

CTABLES does not offer these options, but it can handle two different kinds of weights:  frequency or replication weights vs effective base weights.  Which is appropriate depends on the meaning of the weights.  Frequency weights are specified by the same weight variable that crosstabs uses.  Effective base weights are sampling weights specified within the CTABLES command or the Options tab in the gui.   I am leaving aside the complex sampling weights using in the specialized Complex Samples procedures such as CSTABLES.

On Mon, Nov 25, 2019 at 8:59 AM Janel Schuh <[hidden email]> wrote:

Hi, Bill Ware recommended that I reach out to this listserv with a question I have about SPSS.

 

My problem:

 

When I run weighted proportions in SPSS through their traditional interface (e.g., frequencies, crosstabs), I get results (counts, %s) that don't exactly match those I get when I use the Tables add-on.

• Results are typically very close, but not exactly the same.
• In a few cases, there are substantial differences (e.g., differences of more than +/-2%).

 

I’d appreciate any explanations, resources, etc. that can help me understand why the stats differ and any insight into why SPSS would use 2 different ways of calculating the same statistics, etc.

 

I’ve heard this may have something to do with whether weighted or unweighted n sizes are used in calculating statistics, but have had trouble nailing down any citations about this topic.

 

Regards, Janel

 

Janel S. Schuh, PhD

Senior Research Scientist, Health Communication

CommunicateHealth, Inc. 

 

202.277.8088

www.communicatehealth.com 

@CommunicateHlth

 

 

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