Weighting

[John B. Mitchell]

Hello folks -

I am a little off topic here but not sure where else to go. I have not done
sample weighting in about a thousand years and the people I used to work
with have all gone on to bigger and better things and I can't find them.

I am trying to weight a sample on three variables:
Region (4 values)
Type of business (2 values)
Volume of business (5 values).

We know from industry data what the values for each variable should look
like and what I want to do is to weight the real sample to reflect the
target values.

My brain has become simplistic over the years and I have tried one formula
which brings me right back to my actual values rather than the target values
(W = final weight, w = variable weight):
W = 1-(1-w1)(1-w2)(1-w3)

Does anyone have a solution they would like to share? Much thanks in
advance, and pardon my off-topic post.

John

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

[Steve Simon, P.Mean Consulting]

John B. Mitchell wrote:

> Hello folks -
>
> I am a little off topic here but not sure where else to go. I have not done
> sample weighting in about a thousand years and the people I used to work
> with have all gone on to bigger and better things and I can't find them.
>
> I am trying to weight a sample on three variables:
> Region (4 values)
> Type of business (2 values)
> Volume of business (5 values).
>
> We know from industry data what the values for each variable should look
> like and what I want to do is to weight the real sample to reflect the
> target values.

If all you want is an unbiased estimate of the overall mean/proportion,
here's an approach that will work.

There are 40 (=4*2*5) strata in your group. If the sample proportion in
a strata is equal to the target population proportions, then set the
weight equal to 1. If the sample proportion is twice as large as it
should be, set the weight equal to 1/2. If the sample proportion is half
as large as it should be, set the weight equal to 2. It's Friday
afternoon, so I don't want to write out the general case, but you should
see where this is going.
--
Steve Simon, Standard disclaimer.
I'm now on Facebook and LinkedIn and
looking for new friends/connections
www.facebook.com/pmean
www.linkedin.com/in/pmean

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

[Ruben Geert van den Berg]

Dear John,

You wrote: "> We know from industry data what the values for each variable should look

> like".

As I understand this, you only have the marginal distributions on these variables, right? I mean, you don't have a crosstabulation of [region * type * volume] for your target population?

If so, you could try the RAKE extension written by Jon Peck. It carries out rim weighting IIRC (I never used it since we've other software for calculating probablility weights). I'm not sure but I think it requires V17 or V18 and you need to have Python installed but no Python knowledge is required to use it. It is downloadable from uh... well, nevermind, if you want to give it a shot, I'll send you the .zip file off-list.

HTH!

Ruben van den Berg

Methodologist

TNS NIPO

E: [hidden email]

P: [phone]

I: www.tns-nipo.com

> Date: Fri, 12 Feb 2010 09:27:05 -0500
> From: [hidden email]
> Subject: Weighting
> To: [hidden email]
>
> Hello folks -
>
> I am a little off topic here but not sure where else to go. I have not done
> sample weighting in about a thousand years and the people I used to work
> with have all gone on to bigger and better things and I can't find them.
>
> I am trying to weight a sample on three variables:
> Region (4 values)
> Type of business (2 values)
> Volume of business (5 values).
>
> We know from industry data what the values for each variable should look
> like and what I want to do is to weight the real sample to reflect the
> target values.
>
> My brain has become simplistic over the years and I have tried one formula
> which brings me right back to my actual values rather than the target values
> (W = final weight, w = variable weight):
> W = 1-(1-w1)(1-w2)(1-w3)
>
> Does anyone have a solution they would like to share? Much thanks in
> advance, and pardon my off-topic post.
>
> John
>
> =====================
> 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

---

Express yourself instantly with MSN Messenger! MSN Messenger

[Henrik Lolle]

Dear John,

You can also use Joao Duarte's SPSS macro which you can get from
Raynald's web side:

http://www.spsstools.net/Syntax/Compute/WeightDataBasedOn2orMoreVars.txt

This works very fine.

Best,
Henrik

Quoting Ruben van den Berg <[hidden email]>:

>
> Dear John,
> You wrote: "> We know from industry data what the values for each
> variable should look> like".
> As I understand this, you only have the marginal distributions on
> these variables, right? I mean, you don't have a crosstabulation of
> [region * type * volume] for your target population?
> If so, you could try the RAKE extension written by Jon Peck. It
> carries out rim weighting IIRC (I never used it since we've other
> software for calculating probablility weights). I'm not sure but I
> think it requires V17 or V18 and you need to have Python installed
> but no Python knowledge is required to use it. It is downloadable
> from uh... well, nevermind, if you want to give it a shot, I'll send
> you the .zip file off-list.
> HTH!
> Ruben van den Berg
> Methodologist
> TNS NIPO
> E: [hidden email]
> P: [phone]
> I: www.tns-nipo.com
>
>
>
>> Date: Fri, 12 Feb 2010 09:27:05 -0500
>> From: [hidden email]
>> Subject: Weighting
>> To: [hidden email]
>>
>> Hello folks -
>>
>> I am a little off topic here but not sure where else to go. I have not done
>> sample weighting in about a thousand years and the people I used to work
>> with have all gone on to bigger and better things and I can't find them.
>>
>> I am trying to weight a sample on three variables:
>> Region (4 values)
>> Type of business (2 values)
>> Volume of business (5 values).
>>
>> We know from industry data what the values for each variable should look
>> like and what I want to do is to weight the real sample to reflect the
>> target values.
>>
>> My brain has become simplistic over the years and I have tried one formula
>> which brings me right back to my actual values rather than the target values
>> (W = final weight, w = variable weight):
>> W = 1-(1-w1)(1-w2)(1-w3)
>>
>> Does anyone have a solution they would like to share? Much thanks in
>> advance, and pardon my off-topic post.
>>
>> John
>>
>> =====================
>> 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
>
> _________________________________________________________________
> Express yourself instantly with MSN Messenger! Download today it's FREE!
> http://messenger.msn.click-url.com/go/onm00200471ave/direct/01/

************************************************************
Henrik Lolle
Department of Economics, Politics and Public Administration
Aalborg University
Fibigerstraede 1
9200 Aalborg
Phone: [phone]
************************************************************

=====================
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 Webb-5]

If you're into R look for Rake.

Mark Webb

Line [phone]
Cell [phone]
Fax to email [phone]
Skype  webbmark
Email  [hidden email]


On 2010/02/14 11:00 PM, Henrik Lolle wrote:

> Dear John,
>
> You can also use Joao Duarte's SPSS macro which you can get from
> Raynald's web side:
>
> http://www.spsstools.net/Syntax/Compute/WeightDataBasedOn2orMoreVars.txt
>
> This works very fine.
>
> Best,
> Henrik
>
> Quoting Ruben van den Berg <[hidden email]>:
>
>>
>> Dear John,
>> You wrote: "> We know from industry data what the values for each
>> variable should look> like".
>> As I understand this, you only have the marginal distributions on
>> these variables, right? I mean, you don't have a crosstabulation of
>> [region * type * volume] for your target population?
>> If so, you could try the RAKE extension written by Jon Peck. It
>> carries out rim weighting IIRC (I never used it since we've other
>> software for calculating probablility weights). I'm not sure but I
>> think it requires V17 or V18 and you need to have Python installed
>> but no Python knowledge is required to use it. It is downloadable
>> from uh... well, nevermind, if you want to give it a shot, I'll send
>> you the .zip file off-list.
>> HTH!
>> Ruben van den Berg
>> Methodologist
>> TNS NIPO
>> E: [hidden email]
>> P: [phone]
>> I: www.tns-nipo.com
>>
>>
>>
>>> Date: Fri, 12 Feb 2010 09:27:05 -0500
>>> From: [hidden email]
>>> Subject: Weighting
>>> To: [hidden email]
>>>
>>> Hello folks -
>>>
>>> I am a little off topic here but not sure where else to go. I have
>>> not done
>>> sample weighting in about a thousand years and the people I used to
>>> work
>>> with have all gone on to bigger and better things and I can't find
>>> them.
>>>
>>> I am trying to weight a sample on three variables:
>>> Region (4 values)
>>> Type of business (2 values)
>>> Volume of business (5 values).
>>>
>>> We know from industry data what the values for each variable should
>>> look
>>> like and what I want to do is to weight the real sample to reflect the
>>> target values.
>>>
>>> My brain has become simplistic over the years and I have tried one
>>> formula
>>> which brings me right back to my actual values rather than the
>>> target values
>>> (W = final weight, w = variable weight):
>>> W = 1-(1-w1)(1-w2)(1-w3)
>>>
>>> Does anyone have a solution they would like to share? Much thanks in
>>> advance, and pardon my off-topic post.
>>>
>>> John
>>>
>>> =====================
>>> 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
>>
>> _________________________________________________________________
>> Express yourself instantly with MSN Messenger! Download today it's FREE!
>> http://messenger.msn.click-url.com/go/onm00200471ave/direct/01/
>
>
>
> ************************************************************
> Henrik Lolle
> Department of Economics, Politics and Public Administration
> Aalborg University
> Fibigerstraede 1
> 9200 Aalborg
> Phone: [phone]
> ************************************************************
>
> =====================
> 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