weights in SPSS and inferential statistics

classic Classic list List threaded Threaded
4 messages Options
Reply | Threaded
Open this post in threaded view
|

weights in SPSS and inferential statistics

Zdaniuk, Bozena-3
Hello everyone,
I am combining a number of samples from different studies which differ greatly in size. I think I should use weights before I run any analyses. However, I have read somewhere online that SPSS treats weights incorrectly in inferential statistics. Is that true?
cheers,
bozena zdaniuk
=====================
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
Reply | Threaded
Open this post in threaded view
|

Re: weights in SPSS and inferential statistics

Rich Ulrich

Analyses with weights should always spur suspicions of validity for inference.

Weighting (IMHO) provides more consistency in a set of estimate of tested effects,

at the sacrifice of pretending that some contrasts have Ns greater than they actually

do. When the Ns are not adjusted by much, that is not much sacrifice.


Once you add in "studies which differ greatly in size", your best hope of validity

lies with a meta-analytic approach which does careful testing of the how badly

assumptions are being violated - for one thing, "homogeneity of results" within

sets of samples for which you are computing means. Good meta-analysis is tougher

to achieve than ordinary "good analysis" since it wants expertise in the subject matter

plus expertise in statistics.


I remember reading, long ago, that the preferred approach for surveys did weighting

differently than SPSS allowed at the time. That's probably irrelevant for "differ greatly

in size".


--

Rich Ulrich


From: SPSSX(r) Discussion <[hidden email]> on behalf of Zdaniuk, Bozena <[hidden email]>
Sent: Thursday, May 3, 2018 2:15:51 PM
To: [hidden email]
Subject: weights in SPSS and inferential statistics
 
Hello everyone,
I am combining a number of samples from different studies which differ greatly in size. I think I should use weights before I run any analyses. However, I have read somewhere online that SPSS treats weights incorrectly in inferential statistics. Is that true?
cheers,
bozena zdaniuk
=====================
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
Reply | Threaded
Open this post in threaded view
|

Re: weights in SPSS and inferential statistics

Jon Peck
In reply to this post by Zdaniuk, Bozena-3
Statistics supports three general kinds of weights in addition to some specialized weights such as for weighted least squares or importance weights in trees.  How to weight is a complicated statistical issue.

The general SPSS weight is a replication or frequency weight that is equivalent to duplicating cases according to the weight.

Complex samples weights arise from nonrandom sampling where the probability of selection varies.  These are used by the complex samples procedures.

Effective base weighting is available in custom tables and would typically be used for fractional weights as would arise from weighting to control totals using the SPSSINC RAKE procedure or similar.

So whether and how to weight depends on what the weights mean and what statistical procedures you want to use.

On Thu, May 3, 2018 at 12:15 PM, Zdaniuk, Bozena <[hidden email]> wrote:
Hello everyone,
I am combining a number of samples from different studies which differ greatly in size. I think I should use weights before I run any analyses. However, I have read somewhere online that SPSS treats weights incorrectly in inferential statistics. Is that true?
cheers,
bozena zdaniuk
=====================
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
[hidden email]

===================== 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
Reply | Threaded
Open this post in threaded view
|

Re: weights in SPSS and inferential statistics

Art Kendall
"Complex samples weights arise from nonrandom sampling where the probability
of selection varies.  These are used by the complex samples procedures."

The samples in Complex sample are not SRS -- Simple random samples from a
complete population. The overall sampling although not simple is scientific
aka random.  Estimates of the uncertainty due to sampling are in accord with
good statistical practice.  Ther are at least strong analogies between
Complex  Samples and Multilevel Modelling.

Instead of being simple. they are designed to take account of information
about the population.  Different parts are sampled by SRS within those
parts.  
The latest editing of Chance has an example of using the same size simple
random samples within each of the Court Districts in Maryland.

The whole issue on statistics in human rights is available for free at.
https://chance.amstat.org/archive/
Scroll down to February 2018 to see the Table of contents.


The specific article is at
https://chance.amstat.org/2018/02/maryland-rent-courts/



-----
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
Art Kendall
Social Research Consultants