[DDI-SRG] FW: stats question please

Tue Mar 24 19:54:57 EDT 2009

Wendy:

No - there's wasn't any approach that worked that I could see. They've
started using "bootstrap weights" at Stats Can, so this will be an on-going
issue, and they have to describe the data as they receive it, so inventing a
variable isn't something they can do.

As I understand the requirement, what is needed is to have several sets of
statistics per variable - the fix we envisioned was to make the "Statistics"
element repeatable. I should probably turn you over to Pascal, as I believe
he understand the issue better than I do.

Cheers,

Arofan

-----Original Message-----
From: ddi-srg-bounces at icpsr.umich.edu
[mailto:ddi-srg-bounces at icpsr.umich.edu] On Behalf Of Wendy Thomas
Sent: Tuesday, March 24, 2009 2:56 PM
To: arofan.gregory
Cc: 'Mary Vardigan'; 'Sanda Ionescu'; 'TIC list'
Subject: Re: [DDI-SRG] FW: stats question please

Different sets of statistics are easy as you can simply provide multiple 
summary statistics for the same variable. The question is when you use 
multiple weights to create a statistics. I recommended the use of a 
variable that reflects the generated weight using a generation instruction 
to describe the process. Then that weight can be referred to from the 
summary statistics. Did you have another approach??

wendy

On Tue, 24 Mar 2009, arofan.gregory wrote:

> Wendy:
>
> We're here in Ottowa working with the Canadian RDC people (Stats Can data)
> and they have exactly the same issue - they need to have different weights
> for a single study, and therefore different sets of statistics.
>
> This is *very* common in Stats Canada data, apparently.
>
> I'm happy to show you how I think it could be fixed.
>
> Cheers,
>
> Arofan
>
> -----Original Message-----
> From: ddi-srg-bounces at icpsr.umich.edu
> [mailto:ddi-srg-bounces at icpsr.umich.edu] On Behalf Of Wendy Thomas
> Sent: Monday, March 23, 2009 11:26 AM
> To: Sanda Ionescu
> Cc: Mary Vardigan; TIC list
> Subject: Re: [DDI-SRG] FW: stats question please
>
> We require the ability to make this machine actionable. We are creating
> variable representations of measures, concatonated measures providing
> unique identifiers and links. We are no longer just documenting data.
>
> Wendy
>
>
> On Mon, 23 Mar 2009, Sanda Ionescu wrote:
>
>> Wendy,
>>
>> It is for case 2) - a combination of weights.
>> But ... I don't know...
>> Creating a "fake" weight in DDI only (not in the data) seems a little
> far-fetched.
>> And really, I'm not sure we should be going that way.
>> Aren't we just documenting data?
>> Why resort to creating artifices like that - assigning fake values,
> creating imaginary weights?
>> And again, I think this makes it so much more difficult to automate -
both
> direct markup, and conversions.
>>
>> Sanda.
>>
>> Sanda Ionescu
>> ICPSR
>> University of Michigan
>> P.O. Box 1248
>> Ann Arbor, MI 48106
>>
>> Phone, Fax: 734-615-7890
>>
>> -----Original Message-----
>> From: Wendy Thomas [mailto:wlt at pop.umn.edu]
>> Sent: Monday, March 23, 2009 12:48 PM
>> To: Sanda Ionescu
>> Subject: Re: FW: stats question please
>>
>> Please clarify
>>
>> If they mean that a variable can have more than one weight, a separate
>> statistic is created for each weight used.
>>
>> If they mean they use a combination of weights, then DDI 3.0 requires
they
>> create a variable (not expressed in the data set physical content) that
>> provides the expression of that multiple weighting through its generation
>> instruction.
>>
>> For example: If I had a standard weight A and also applied weight
variable
>> X and weight variable Y in calculating the statistic then I would need to
>> create a variable Z which using the generation instruction  of
>> "StandardWeightA * variableX * variableY.
>>
>> It may not be how they want to express it, but this is the way its done.
>>
>> Wendy
>>
>>
>>  On Mon, 23 Mar 2009, Sanda Ionescu wrote:
>>
>>> Hi, Wendy.
>>> This is a different issue.
>>> Our researchers/ analysis specialists seem to think that it is necessary
> to be able to reference multiple weights for a statistic.
>>> Please see exchange below.
>>> Could you submit this as a request for a change to TIC?
>>> Thanks
>>> Sanda.
>>>
>>> Sanda Ionescu
>>> ICPSR
>>> University of Michigan
>>> P.O. Box 1248
>>> Ann Arbor, MI 48106
>>>
>>> Phone, Fax: 734-615-7890
>>>
>>> -----Original Message-----
>>> From: Felicia LeClere
>>> Sent: Monday, March 23, 2009 10:14 AM
>>> To: Sanda Ionescu
>>> Subject: Re: stats question please
>>>
>>> It happens...so it is necessary to be able to have more than one weight
>>>
>>> -----Original Message-----
>>> From: Sanda Ionescu
>>> To: Lynette Hoelter; Felicia LeClere
>>> CC: Mary Vardigan
>>> Sent: Mon Mar 23 09:23:52 2009
>>> Subject: RE: stats question please
>>>
>>> It's about DDI.
>>>
>>> Version 3 only allows 1 weight to be defined per statistical value.
>>>
>>> And we need to know if it is necessary to ask for the weight to become
> repeatable (so that one can link to more than one).
>>>
>>> This is not about whether it's LIKELY or not (because, yes, it is
> UNLIKELY) but rather is it necessary to cover for that possibility too,
> however unlikely it may be?
>>> Sanda.
>>>
>>>
>>>
>>> Sanda Ionescu
>>> ICPSR
>>> University of Michigan
>>> P.O. Box 1248
>>> Ann Arbor, MI 48106
>>>
>>> Phone, Fax: 734-615-7890
>>>
>>> From: Lynette Hoelter
>>> Sent: Friday, March 20, 2009 4:22 PM
>>> To: Sanda Ionescu; Felicia LeClere
>>> Cc: Mary Vardigan
>>> Subject: RE: stats question please
>>>
>>>
>>>
>>> In theory, I don't see why not because the weights just multiple the
> values for the cases by some fraction, so you could multiply by two.
>>> Why?
>>>
>>>
>>>
>>> ....................................................
>>> Lynette F. Hoelter, Ph.D.
>>> Dir., Instructional Resources & Development, ICPSR
>>> Institute for Social Research
>>> University of Michigan
>>>
>>>
>>>
>>> From: Sanda Ionescu
>>> Sent: Friday, March 20, 2009 3:13 PM
>>> To: Felicia LeClere; Lynette Hoelter
>>> Cc: Mary Vardigan
>>> Subject: stats question please
>>>
>>>
>>>
>>> Felicia, Lynette,
>>>
>>> In Variable Statistics is it POSSIBLE  to obtain ONE value (say, mean,
or
> mode, etc.) by applying TWO different weights - not sequentially, but at
the
> same time?
>>> Thanks
>>>
>>> Sanda.
>>>
>>>
>>>
>>> Sanda Ionescu
>>> ICPSR
>>> University of Michigan
>>> P.O. Box 1248
>>> Ann Arbor, MI 48106
>>>
>>> Phone, Fax: 734-615-7890
>>>
>>>
>>
>> Wendy L. Thomas                          Phone: +1 612.624.4389
>> Data Access Core Director		 Fax:   +1 612.626.8375
>> Minnesota Population Center              Email: wlt at pop.umn.edu
>> University of Minnesota
>> 50 Willey Hall
>> 225 19th Avenue South
>> Minneapolis, MN 55455
>>
>
> Wendy L. Thomas                          Phone: +1 612.624.4389
> Data Access Core Director		 Fax:   +1 612.626.8375
> Minnesota Population Center              Email: wlt at pop.umn.edu
> University of Minnesota
> 50 Willey Hall
> 225 19th Avenue South
> Minneapolis, MN 55455
>

Wendy L. Thomas                          Phone: +1 612.624.4389
Data Access Core Director		 Fax:   +1 612.626.8375
Minnesota Population Center              Email: wlt at pop.umn.edu
University of Minnesota
50 Willey Hall
225 19th Avenue South
Minneapolis, MN 55455
_______________________________________________
DDI-SRG mailing list
DDI-SRG at icpsr.umich.edu
http://www.icpsr.umich.edu/mailman/listinfo/ddi-srg