OT compare & contrast Linear Mixed Models :: Complex Samples

[Art Kendall]

Since the mid 90's I have have had an entry on my TO DO List, that says that "there are many logical similarities between Linear Mixed Models and Complex Samples".

I retired in 2001 and have decided that I'll never have a chance to work this out.

Over the years there have been many instances where different dialects of statistics have been found to be very much the same thing ANOVA and Regression (Cohen); Clustering, Pattern Detection, Unsupervised Learning; Discriminant function analysis, Pattern Recognition, Supervised Learning.

When I mention the similarity of Mixed models and Complex samples to a top math psychologist/statistician, he said "of course thy are similar." A top sampling expert said that they are often identical but sometimes differ. We did not have time to get into when they would differ.

It would make a good publication to
(1) compare/contrast the logic
(2) compare/contrast the math
(3) run some actual or textbook analyses both ways and compare/contrast the details and conclusions.

Or has anybody already worked this out?

[Bruce Weaver]

Excellent question, Art.  I'd be very interested in reading the article or chapter you're proposing.  

One practical issue that came up the other day in discussion with a colleague (an epidemiologist) is that with big national surveys, the raw data needed for a Linear Mixed Model approach may not be available, but the survey weights needed for a Complex Samples approach are included.  

Cheers,
Bruce

Art Kendall wrote

Since the mid 90's I have have had an entry on my TO DO List, that says that "there are many logical similarities between Linear Mixed Models and Complex Samples".

I retired in 2001 and have decided that I'll never have a chance to work this out.

Over the years there have been many instances where different dialects of statistics have been found to be very much the same thing ANOVA and Regression (Cohen); Clustering, Pattern Detection, Unsupervised Learning; Discriminant function analysis, Pattern Recognition, Supervised Learning.

When I mention the similarity of Mixed models and Complex samples to a top math psychologist/statistician, he said "of course thy are similar." A top sampling expert said that they are often identical but sometimes differ. We did not have time to get into when they would differ.

It would make a good publication to
(1) compare/contrast the logic
(2) compare/contrast the math
(3) run some actual or textbook analyses both ways and compare/contrast the details and conclusions.

Or has anybody already worked this out?

[Ryan]

Art,

There are theoretical differences and differences in the algorithms which could lead to different results. 

Please read this SUGI paper. The last example is particularly relevant. 

http://www2.sas.com/proceedings/sugi31/193-31.pdf

Ryan 

Sent from my iPhone

On Feb 22, 2017, at 9:32 AM, Art Kendall <[hidden email]> wrote:

Since the mid 90's I have have had an entry on my TO DO List, that says that
"there are many logical similarities between Linear Mixed Models and Complex
Samples".

I retired in 2001 and have decided that I'll never have a chance to work
this out.

Over the years there have been many instances where different dialects of
statistics have been found to be very much the same thing ANOVA and
Regression (Cohen); Clustering, Pattern Detection, Unsupervised Learning;
Discriminant function analysis, Pattern Recognition, Supervised Learning.

When I mention the similarity of Mixed models and Complex samples to a top
math psychologist/statistician, he said "of course thy are similar." A top
sampling expert said that they are often identical but sometimes differ. We
did not have time to get into when they would differ.

It would make a good publication to
(1) compare/contrast the logic
(2) compare/contrast the math
(3) run some actual or textbook analyses both ways and compare/contrast the
details and conclusions.

Or has anybody already worked this out?



-----
Art Kendall
Social Research Consultants
--
View this message in context: http://spssx-discussion.1045642.n5.nabble.com/OT-compare-contrast-Linear-Mixed-Models-Complex-Samples-tp5733875.html
Sent from the SPSSX Discussion mailing list archive at 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

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

[James Algina]

Hi Art,

This reference may be of some interest

Sterba, S. K. (2009) Alternative Model-Based and Design-Based Frameworks for Inference From Samples to Populations:
From Polarization to Integration. MBR. 44:6, 711-740. DOI:10.1080/00273170903333574  

Jamie

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Art Kendall
Sent: Wednesday, February 22, 2017 9:32 AM
To: [hidden email]
Subject: OT compare & contrast Linear Mixed Models :: Complex Samples

Since the mid 90's I have have had an entry on my TO DO List, that says that "there are many logical similarities between Linear Mixed Models and Complex Samples".

I retired in 2001 and have decided that I'll never have a chance to work this out.

Over the years there have been many instances where different dialects of statistics have been found to be very much the same thing ANOVA and Regression (Cohen); Clustering, Pattern Detection, Unsupervised Learning; Discriminant function analysis, Pattern Recognition, Supervised Learning.

When I mention the similarity of Mixed models and Complex samples to a top math psychologist/statistician, he said "of course thy are similar." A top sampling expert said that they are often identical but sometimes differ. We did not have time to get into when they would differ.

It would make a good publication to
(1) compare/contrast the logic
(2) compare/contrast the math
(3) run some actual or textbook analyses both ways and compare/contrast the details and conclusions.

Or has anybody already worked this out?



-----
Art Kendall
Social Research Consultants
--
View this message in context: http://spssx-discussion.1045642.n5.nabble.com/OT-compare-contrast-Linear-Mixed-Models-Complex-Samples-tp5733875.html
Sent from the SPSSX Discussion mailing list archive at 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

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

[Bruce Weaver]

Here's another one I just found that appears to be on topic.

Carle, A. C. (2009). Fitting multilevel models in complex survey data with design weights: Recommendations. BMC Medical Research Methodology, 9:49.  DOI: 10.1186/1471-2288-9-49

http://bmcmedresmethodol.biomedcentral.com/articles/10.1186/1471-2288-9-49

James Algina wrote

Hi Art,

This reference may be of some interest

Sterba, S. K. (2009) Alternative Model-Based and Design-Based Frameworks for Inference From Samples to Populations:
From Polarization to Integration. MBR. 44:6, 711-740. DOI:10.1080/00273170903333574  

Jamie

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Art Kendall
Sent: Wednesday, February 22, 2017 9:32 AM
To: [hidden email]
Subject: OT compare & contrast Linear Mixed Models :: Complex Samples

Since the mid 90's I have have had an entry on my TO DO List, that says that "there are many logical similarities between Linear Mixed Models and Complex Samples".

I retired in 2001 and have decided that I'll never have a chance to work this out.

Over the years there have been many instances where different dialects of statistics have been found to be very much the same thing ANOVA and Regression (Cohen); Clustering, Pattern Detection, Unsupervised Learning; Discriminant function analysis, Pattern Recognition, Supervised Learning.

When I mention the similarity of Mixed models and Complex samples to a top math psychologist/statistician, he said "of course thy are similar." A top sampling expert said that they are often identical but sometimes differ. We did not have time to get into when they would differ.

It would make a good publication to
(1) compare/contrast the logic
(2) compare/contrast the math
(3) run some actual or textbook analyses both ways and compare/contrast the details and conclusions.

Or has anybody already worked this out?



-----
Art Kendall
Social Research Consultants
--
View this message in context: http://spssx-discussion.1045642.n5.nabble.com/OT-compare-contrast-Linear-Mixed-Models-Complex-Samples-tp5733875.html
Sent from the SPSSX Discussion mailing list archive at 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

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

We started a coalition of science organizations and individuals to work on human rights issues . Working on the Coalition itself,  the the On-Call Scientists program of the Coalition, Statistics Without Borders, the ASA Committee on Scientific Freedom and Human Rights, and the Advisory Board of the Society for Terrorism Research I won't have time to take on any more activities.

https://www.aaas.org/program/science-human-rights-coalition

I posted that message with the hope that somebody would either put together such an article or chapter or have a student do it as a project or something like that.

[Art Kendall]

Thank you.
"Could" is a very important word.  I cannot get to it, but the reasonings have similarities.

The oints on the link  are some of the reasons I pushed SPSS to develop Complex Samples decades ago. Since that time MLM has come on the scene.

Are you aware of times when one could not do all of these things in SPSS?

Over the years, the different dialects of statistics often fell into the problem of applying the "law of the instrument".  As time went on people figured out that some procedures were special cases  of others. More and more general concepts became more widely known. (I believe a lot of this progress is due to better conceptualizing, better software and better hardware.)

In the early 70's when SPSS came out, there were papers showing that many of the correlation coefficients were short cuts that made hand calculation easier when the input data were ranks, dichotomies, etc. But in several instances  using PEARSON turned out to produce the same coefficient and significance whcih was a revelation to many.  Cohen's work a little later Multiple Regression As a General Data Analytic Method brought together different analyses as special cases of more general methods.  Multiple regression became known as a special case of canonical correlation. Doug Carrol's N-battery Canonical Correlation generalized canonical correlation beyond just 2 sets of variables. etc. Today we have GEE, MLD, etc. etc.

[Art Kendall]

Thank you.

That looks like an interesting article.
Sterba, S. K. (2009) Alternative Model-Based and Design-Based Frameworks for Inference From Samples to Populations:
From Polarization to Integration. MBR. 44:6, 711-740. DOI:10.1080/00273170903333574  

I no longer have access to libraries that carry MBR does anybody have a pdf that (s)he could send me?

[Art Kendall]

Thanks for Carle, A. C. (2009).

Apologies to the list. I thought my replies would be with the posts I was replying to rather than at the end of the thread.

[Mike]

Art,
An HTML version of the article is available on the PubMed website
(PDF versions also available but they are of the HTML version,
not copies of the pages of the journal article); see:
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC2856970/

-Mike Palij
New York University
[hidden email]


----- Original Message -----
On Friday, March 03, 2017 9:00 AM, Art Kendall wrote:

> Thank you.
>
> That looks like an interesting article.
> Sterba, S. K. (2009) Alternative Model-Based and Design-Based
> Frameworks for
> Inference From Samples to Populations:
> From Polarization to Integration. MBR. 44:6, 711-740.
> DOI:10.1080/00273170903333574
>
> I no longer have access to libraries that carry MBR does anybody have
> a pdf
> that (s)he could send me?

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