Question on Weighting

[Barry DeCicco]

Hello,

I think that I might be doing some analyses incorrectly; I've been using weights, which have averaged over 1  
(they are based on volumes/response counts).  In SPSS, are these interpreted as frequencies?  If so, I believe that I've been falsely inflating the significances.  Should I normalize the weights (divide by the mean, so that the new average is 1)?

Thank you very much,

Barry

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

[Maguin, Eugene]

I believe that to weight the dataset to the sample size, the population weight for each record must be divided by the sum (not the mean) of the population weights.
You probably have already done so but if not, you should read the weight command documentation very carefully.
Gene Maguin



-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Barry DeCicco
Sent: Friday, September 12, 2014 9:31 AM
To: [hidden email]
Subject: Question on Weighting

Hello,

I think that I might be doing some analyses incorrectly; I've been using weights, which have averaged over 1 (they are based on volumes/response counts).  In SPSS, are these interpreted as frequencies?  If so, I believe that I've been falsely inflating the significances.  Should I normalize the weights (divide by the mean, so that the new average is 1)?

Thank you very much,

Barry

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

[Mark Miller]

In either case, the weights are still proportional, so Div by Mean or Div by Sum will not change proportions.

Ordinarily, the choice would be to transform the weights so that Sum of Weights = N of Observations.

That would be dividing by the Mean Population Weight.

... Mark Miller

On Fri, Sep 12, 2014 at 8:11 AM, Maguin, Eugene <[hidden email]> wrote:

I believe that to weight the dataset to the sample size, the population weight for each record must be divided by the sum (not the mean) of the population weights.
You probably have already done so but if not, you should read the weight command documentation very carefully.
Gene Maguin

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Barry DeCicco
Sent: Friday, September 12, 2014 9:31 AM
To: [hidden email]
Subject: Question on Weighting

Hello,

I think that I might be doing some analyses incorrectly; I've been using weights, which have averaged over 1 (they are based on volumes/response counts).  In SPSS, are these interpreted as frequencies?  If so, I believe that I've been falsely inflating the significances.  Should I normalize the weights (divide by the mean, so that the new average is 1)?

Thank you very much,

Barry

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

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

Bear in mind that the proper weight treatment really depends on the nature of the weights and the procedure to be used.  Standard SPSS weights are frequency weights and can expand the number of cases correspondingly.  However, if you have survey weights from complex sampling designs, the procedures in the Complex Samples option are designed to take these fully into account.


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




From:        Barry DeCicco <[hidden email]>
To:        [hidden email]
Date:        09/12/2014 07:31 AM
Subject:        [SPSSX-L] Question on Weighting
Sent by:        "SPSSX(r) Discussion" <[hidden email]>

---

Hello,

I think that I might be doing some analyses incorrectly; I've been using weights, which have averaged over 1  
(they are based on volumes/response counts).  In SPSS, are these interpreted as frequencies?  If so, I believe that I've been falsely inflating the significances.  Should I normalize the weights (divide by the mean, so that the new average is 1)?

Thank you very much,

Barry

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

[Barry DeCicco]

"In either case, the weights are still proportional, so Div by Mean or Div by Sum will not change proportions.
Ordinarily, the choice would be to transform the weights so that Sum of Weights = N of Observations.
That would be dividing by the Mean Population Weight."

Yes, it looks like that's what I should be doing.  The main thing that I'm now concerned about are p-values.  It seems that whenever doing any analysis with weights, I'll need to normalize first.

Barry

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

[Mark Miller]

Barry

Exactly right.

Normalization (dividing by Mean Weight) gives you the proper N for p-values.

... Mark Miller

On Fri, Sep 12, 2014 at 12:48 PM, Barry DeCicco <[hidden email]> wrote:

"In either case, the weights are still proportional, so Div by Mean or Div by Sum will not change proportions.
Ordinarily, the choice would be to transform the weights so that Sum of Weights = N of Observations.
That would be dividing by the Mean Population Weight."

Yes, it looks like that's what I should be doing.  The main thing that I'm now concerned about are p-values.  It seems that whenever doing any analysis with weights, I'll need to normalize first.

Barry

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

[Barry]

Thanks, John! I think that one way to express it is that in general, I'd like to treat weights as proportionality coefficients, not as frequencies. I'll look up the documentation for complex samples. Barry

Jon K Peck wrote

Bear in mind that the proper weight treatment really depends on the nature of the weights and the procedure to be used. Standard SPSS weights are frequency weights and can expand the number of cases correspondingly. However, if you have survey weights from complex sampling designs, the procedures in the Complex Samples option are designed to take these fully into account. Jon Peck (no "h") aka Kim Senior Software Engineer, IBM [hidden email] phone: [phone] From: Barry DeCicco <[hidden email]> To: [hidden email] Date: 09/12/2014 07:31 AM Subject: [SPSSX-L] Question on Weighting Sent by: "SPSSX(r) Discussion" <[hidden email]> Hello, I think that I might be doing some analyses incorrectly; I've been using weights, which have averaged over 1 (they are based on volumes/response counts). In SPSS, are these interpreted as frequencies? If so, I believe that I've been falsely inflating the significances. Should I normalize the weights (divide by the mean, so that the new average is 1)? Thank you very much, Barry ===================== 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

Jon K Peck wrote

Bear in mind that the proper weight treatment really depends on the nature of the weights and the procedure to be used. Standard SPSS weights are frequency weights and can expand the number of cases correspondingly. However, if you have survey weights from complex sampling designs, the procedures in the Complex Samples option are designed to take these fully into account. Jon Peck (no "h") aka Kim Senior Software Engineer, IBM [hidden email] phone: [phone] From: Barry DeCicco <[hidden email]> To: [hidden email] Date: 09/12/2014 07:31 AM Subject: [SPSSX-L] Question on Weighting Sent by: "SPSSX(r) Discussion" <[hidden email]> Hello, I think that I might be doing some analyses incorrectly; I've been using weights, which have averaged over 1 (they are based on volumes/response counts). In SPSS, are these interpreted as frequencies? If so, I believe that I've been falsely inflating the significances. Should I normalize the weights (divide by the mean, so that the new average is 1)? Thank you very much, Barry ===================== 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