RIM Weighting

[MR]

Hi Team,

First thank you to SPSS Python module for developing SPSS RACK that allows RIM Weighting, it works like charm. However, I am stuck with one question and could not find answer neither on google or forums.

I am trying to weight the data by age, gender, and region. Both age and gender has 'prefer not to answer' options. 2010 US census data does not have prefer not to answer data. As such, how would you handle the prefer not to answer responses (~1% of total base)?

Thanks,

=====================
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 K Peck]

Early this year I produced a short course on rake weighting with the SPSSINC RAKE command, and I am hoping to post it on the SPSS Community site as soon as some necessary paperwork is completed.  It addresses this question briefly.  There is no perfect solution for item nonresponse, but there are two approaches in use.

extracted from that material.

Example: age group sometimes not reported
Affects matching national totals

- Use an imputation procedure to fill in the nonresponse before raking
- Calculate missing percentage and subtract from total age counts

With only 1% missing, these adjustments will be small, so I wouldn't be too worried about the issue, although it is possible that the nonresponse is unevenly distributed.

HTH,

Jon Peck (no "h") aka Kim
Senior Software Engineer, IBM
[hidden email]
new phone: [phone]




From:        MR <[hidden email]>
To:        [hidden email]
Date:        09/30/2012 05:48 PM
Subject:        [SPSSX-L] RIM Weighting
Sent by:        "SPSSX(r) Discussion" <[hidden email]>

---

Hi Team,

First thank you to SPSS Python module for developing SPSS RACK that allows RIM Weighting, it works like charm. However, I am stuck with one question and could not find answer neither on google or forums.

I am trying to weight the data by age, gender, and region. Both age and gender has 'prefer not to answer' options. 2010 US census data does not have prefer not to answer data. As such, how would you handle the prefer not to answer responses (~1% of total base)?

Thanks,

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

[MR]

Hi John,

Thank you for providing further clarification. While running the RACK command, I realized that I am forcing the weights within each category. For e.g., let say dimension is gender: male 49% female 51%, dimension 2: region - midwest 30%, west - 30%, south 40%, east 10%. However, as per the US 2010 census data, gender distribution is not consistent across US region. In the current RACK command, I am forcing gender to be same within and across regions, which does not 'truly' reflect the US census. Is there a way to do interlocked weighting where I can define gender proportions within each region?

Weighting is always complicated and I highly appreciate all your efforts and thoughts thus far. I look forward to hearing your thoughts.

J 

On 2012-09-30, at 9:09 PM, Jon K Peck wrote:

Early this year I produced a short course on rake weighting with the SPSSINC RAKE command, and I am hoping to post it on the SPSS Community site as soon as some necessary paperwork is completed.  It addresses this question briefly.  There is no perfect solution for item nonresponse, but there are two approaches in use.

extracted from that material.

Example: age group sometimes not reported
Affects matching national totals

- Use an imputation procedure to fill in the nonresponse before raking
- Calculate missing percentage and subtract from total age counts

With only 1% missing, these adjustments will be small, so I wouldn't be too worried about the issue, although it is possible that the nonresponse is unevenly distributed.

HTH,

Jon Peck (no "h") aka Kim
Senior Software Engineer, IBM
[hidden email]
new phone: [phone]




From:        MR <[hidden email]>
To:        [hidden email]
Date:        09/30/2012 05:48 PM
Subject:        [SPSSX-L] RIM Weighting
Sent by:        "SPSSX(r) Discussion" <[hidden email]>

---

Hi Team,

First thank you to SPSS Python module for developing SPSS RACK that allows RIM Weighting, it works like charm. However, I am stuck with one question and could not find answer neither on google or forums.

I am trying to weight the data by age, gender, and region. Both age and gender has 'prefer not to answer' options. 2010 US census data does not have prefer not to answer data. As such, how would you handle the prefer not to answer responses (~1% of total base)?

Thanks,

=====================
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 K Peck]

First, it is the RAKE command (SPSSINC RAKE), not RACK.

This topic is also addressed in the short course I mentioned.  The solution if you have enough information is to combine gender and region into a single dimension, MW males, MW females, S males, S females, etc.  Then enter the known totals for these combined cells for this composite variable in the raking process.

Jon Peck (no "h") aka Kim
Senior Software Engineer, IBM
[hidden email]
new phone: [phone]




From:        MR <[hidden email]>
To:        Jon K Peck/Chicago/IBM@IBMUS
Cc:        [hidden email]
Date:        10/02/2012 07:44 PM
Subject:        Re: [SPSSX-L] RIM Weighting

---

Hi John,

Thank you for providing further clarification. While running the RACK command, I realized that I am forcing the weights within each category. For e.g., let say dimension is gender: male 49% female 51%, dimension 2: region - midwest 30%, west - 30%, south 40%, east 10%. However, as per the US 2010 census data, gender distribution is not consistent across US region. In the current RACK command, I am forcing gender to be same within and across regions, which does not 'truly' reflect the US census. Is there a way to do interlocked weighting where I can define gender proportions within each region?

Weighting is always complicated and I highly appreciate all your efforts and thoughts thus far. I look forward to hearing your thoughts.

J
On 2012-09-30, at 9:09 PM, Jon K Peck wrote:

Early this year I produced a short course on rake weighting with the SPSSINC RAKE command, and I am hoping to post it on the SPSS Community site as soon as some necessary paperwork is completed.  It addresses this question briefly.  There is no perfect solution for item nonresponse, but there are two approaches in use.

extracted from that material.

Example: age group sometimes not reported
Affects matching national totals

- Use an imputation procedure to fill in the nonresponse before raking
- Calculate missing percentage and subtract from total age counts

With only 1% missing, these adjustments will be small, so I wouldn't be too worried about the issue, although it is possible that the nonresponse is unevenly distributed.

HTH,

Jon Peck (no "h") aka Kim
Senior Software Engineer, IBM
peck@...
new phone: [phone]




From:        MR <manmitmr@...>
To:        [hidden email]
Date:        09/30/2012 05:48 PM
Subject:        [SPSSX-L] RIM Weighting
Sent by:        "SPSSX(r) Discussion" <[hidden email]>

---

Hi Team,

First thank you to SPSS Python module for developing SPSS RACK that allows RIM Weighting, it works like charm. However, I am stuck with one question and could not find answer neither on google or forums.

I am trying to weight the data by age, gender, and region. Both age and gender has 'prefer not to answer' options. 2010 US census data does not have prefer not to answer data. As such, how would you handle the prefer not to answer responses (~1% of total base)?

Thanks,

=====================
To manage your subscription to SPSSX-L, send a message to
LISTSERV@... (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