Statistics Book

[Edward Boadi]

Hi  List.
What will be a good statistical reference  book for a Statistician who has just joined a
Statistical Research consulting firm with specialization in Research Design and Statistical Analysis.
Regards.

[Art Kendall]

Of course, a lot depends on what kinds of clients you have.

 From a broad perspective see Raynald Levesque book on data handling,
etc. and Norusis help about SPSS
http://www.spss.com/PDFs/SB13INS-0405.pdf

The older editions  of the following have been very useful to me in
consulting of something like over 1000 Congressional
evaluations/investigations, and 200 doctoral dissertations in social,
behavioral, and health studies.


clips from Library of Congress catalog

 for design this is the latest version of the classic Cook and Campbell
books

Personal Name:  Shadish, William R.
<http://catalog.loc.gov/cgi-bin/Pwebrecon.cgi?SC=Author&SEQ=20061129155139&PID=24540&SA=Shadish,+William+R.>

Main Title:     Experimental and quasi-experimental designs for generalized
causal inference / William R. Shedish, Thomas D. Cook, Donald T. Campbell.
Published/Created:      Boston : Houghton Mifflin, [2001?]
Related Names:  Cook, Thomas D.
<http://catalog.loc.gov/cgi-bin/Pwebrecon.cgi?SC=Author&SEQ=20061129155139&PID=24540&SA=Cook,+Thomas+D.>


        Campbell, Donald Thomas, 1916-
<http://catalog.loc.gov/cgi-bin/Pwebrecon.cgi?SC=Author&SEQ=20061129155139&PID=24540&SA=Campbell,+Donald+Thomas,+1916->

Description:    xxi, 623 p. : ill. ; 23 cm.
ISBN:   0395615569
Notes:  Includes bibliographical references (p. 514-591) and indexes.
Subjects:       Causation--Experiments.
<http://catalog.loc.gov/cgi-bin/Pwebrecon.cgi?SC=Subject&SEQ=20061129155139&PID=24540&SA=Causation+Experiments.>

LC Classification:      BD591 .S48 2001
Dewey Class No.:        122/.07/2 21


For solid understanding of a broad array of statistical methods

Main Title:     Applied multiple regression/correlation analysis for the
behavioral sciences / Jacob Cohen ... [et al.].
Edition Information:    3rd ed.
Published/Created:      Mahwah, N.J. : L. Erlbaum Associates, 2003.
Related Names:  Cohen, Jacob, 1923-
<http://catalog.loc.gov/cgi-bin/Pwebrecon.cgi?SC=Author&SEQ=20061129160147&PID=3738&SA=Cohen,+Jacob,+1923->


        Cohen, Jacob, 1923- Applied multiple regression/correlation analysis
for the behavioral sciences.
<http://catalog.loc.gov/cgi-bin/Pwebrecon.cgi?SC=Author&SEQ=20061129160147&PID=3738&SA=Cohen,+Jacob,+1923->

Description:    xxviii, 703 p. : ill. ; 26 cm. + 1 CD-ROM (4 3/4 in.)
ISBN:   0805822232 (hard cover : alk. paper)
Notes:  Rev. ed. of: Applied multiple regression/correlation analysis
for the behavioral sciences / Jacob Cohen, Patricia Cohen. 2nd ed. 1983.

        The CD-ROM contains the data for almost all examples as well as the
command codes for each of the major statistical packages for the tabular
and other findings in the book.

        Includes bibliographical references (p. 655-669) and indexes.
Subjects:       Regression analysis.
<http://catalog.loc.gov/cgi-bin/Pwebrecon.cgi?SC=Subject&SEQ=20061129160147&PID=3738&SA=Regression+analysis.>


        Correlation (Statistics)
<http://catalog.loc.gov/cgi-bin/Pwebrecon.cgi?SC=Subject&SEQ=20061129160147&PID=3738&SA=Correlation+%28Statistics%29>


        Social sciences--Statistical methods.
<http://catalog.loc.gov/cgi-bin/Pwebrecon.cgi?SC=Subject&SEQ=20061129160147&PID=3738&SA=Social+sciences+Statistical+methods.>

LC Classification:      HA31.3 .A67 2003
Dewey Class No.:        519.5/36 21



Art Kendall
Social Research Consultants


Edward Boadi wrote:

>Hi  List.
>What will be a good statistical reference  book for a Statistician who has just joined a
>Statistical Research consulting firm with specialization in Research Design and Statistical Analysis.
>Regards.
>
>
>
>

[Edward Boadi]

Thanks Art , I am very grateful.
 

-----Original Message-----
From: Art Kendall [mailto:[hidden email]]
Sent: Wednesday, November 29, 2006 4:38 PM
To: Edward Boadi
Cc: [hidden email]
Subject: Re: Statistics Book


Of course, a lot depends on what kinds of clients you have.

From a broad perspective see Raynald Levesque book on data handling, etc. and Norusis help about SPSS
http://www.spss.com/PDFs/SB13INS-0405.pdf

The older editions  of the following have been very useful to me in consulting of something like over 1000 Congressional evaluations/investigations, and 200 doctoral dissertations in social, behavioral, and health studies.


clips from Library of Congress catalog

 for design this is the latest version of the classic Cook and Campbell books


Personal Name:   Shadish,  <http://catalog.loc.gov/cgi-bin/Pwebrecon.cgi?SC=Author&SEQ=20061129155139&PID=24540&SA=Shadish,+William+R.> William R.
Main Title:      Experimental and quasi-experimental designs for generalized causal inference / William R. Shedish, Thomas D. Cook, Donald T. Campbell.
Published/Created:       Boston : Houghton Mifflin, [2001?]
Related Names:   Cook,  <http://catalog.loc.gov/cgi-bin/Pwebrecon.cgi?SC=Author&SEQ=20061129155139&PID=24540&SA=Cook,+Thomas+D.> Thomas D.

        Campbell,  <http://catalog.loc.gov/cgi-bin/Pwebrecon.cgi?SC=Author&SEQ=20061129155139&PID=24540&SA=Campbell,+Donald+Thomas,+1916-> Donald Thomas, 1916-
Description:     xxi, 623 p. : ill. ; 23 cm.
ISBN:    0395615569
Notes:   Includes bibliographical references (p. 514-591) and indexes.
Subjects:        Causation--Experiments. <http://catalog.loc.gov/cgi-bin/Pwebrecon.cgi?SC=Subject&SEQ=20061129155139&PID=24540&SA=Causation+Experiments.>
LC Classification:       BD591 .S48 2001
Dewey Class No.:         122/.07/2 21

For solid understanding of a broad array of statistical methods


Main Title:      Applied multiple regression/correlation analysis for the behavioral sciences / Jacob Cohen ... [et al.].
Edition Information:     3rd ed.
Published/Created:       Mahwah, N.J. : L. Erlbaum Associates, 2003.
Related Names:   Cohen,  <http://catalog.loc.gov/cgi-bin/Pwebrecon.cgi?SC=Author&SEQ=20061129160147&PID=3738&SA=Cohen,+Jacob,+1923-> Jacob, 1923-

        Cohen,  <http://catalog.loc.gov/cgi-bin/Pwebrecon.cgi?SC=Author&SEQ=20061129160147&PID=3738&SA=Cohen,+Jacob,+1923-> Jacob, 1923- Applied multiple regression/correlation analysis for the behavioral sciences.
Description:     xxviii, 703 p. : ill. ; 26 cm. + 1 CD-ROM (4 3/4 in.)
ISBN:    0805822232 (hard cover : alk. paper)
Notes:   Rev. ed. of: Applied multiple regression/correlation analysis for the behavioral sciences / Jacob Cohen, Patricia Cohen. 2nd ed. 1983.        

        The CD-ROM contains the data for almost all examples as well as the command codes for each of the major statistical packages for the tabular and other findings in the book.

        Includes bibliographical references (p. 655-669) and indexes.
Subjects:        Regression  <http://catalog.loc.gov/cgi-bin/Pwebrecon.cgi?SC=Subject&SEQ=20061129160147&PID=3738&SA=Regression+analysis.> analysis.

        Correlation  <http://catalog.loc.gov/cgi-bin/Pwebrecon.cgi?SC=Subject&SEQ=20061129160147&PID=3738&SA=Correlation+%28Statistics%29> (Statistics)

        Social  <http://catalog.loc.gov/cgi-bin/Pwebrecon.cgi?SC=Subject&SEQ=20061129160147&PID=3738&SA=Social+sciences+Statistical+methods.> sciences--Statistical methods.
LC Classification:       HA31.3 .A67 2003
Dewey Class No.:         519.5/36 21


Art Kendall
Social Research Consultants


Edward Boadi wrote:


Hi  List.

What will be a good statistical reference  book for a Statistician who has just joined a

Statistical Research consulting firm with specialization in Research Design and Statistical Analysis.

Regards.





 

[Anthony Babinec]

Two that I like are:

Gerald van Belle's "Statistical Rules of Thumb" published
by Wiley - Some good material on research design and consulting
in there.

Richard Berk's "Regression Analysis: A Constructive Critique"
published by Sage -- Not a reference book on regression, but
instead a good read to get you thinking about the logic of
analysis and the use of regression and related methods on
observational data.

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
Edward Boadi
Sent: Wednesday, November 29, 2006 1:55 PM
To: [hidden email]
Subject: Statistics Book

Hi  List.
What will be a good statistical reference  book for a Statistician who has
just joined a
Statistical Research consulting firm with specialization in Research Design
and Statistical Analysis.
Regards.

[Art Kendall-2]

To get a handle on why we do statistics in the first place.

Personal Name:  Abelson, Robert P.
<http://catalog.loc.gov/cgi-bin/Pwebrecon.cgi?SC=Author&SEQ=20061130105250&PID=1830&SA=Abelson,+Robert+P.>

Main Title:     Statistics as principled argument / Robert P. Abelson.
Published/Created:      Hillsdale, N.J. : L. Erlbaum Associates, 1995.
Description:    xv, 221 p. ; 24 cm.
ISBN:   0805805273 (acid-free)

        0805805281 (pbk. : acid-free)
Notes:  Includes bibliographical references (p. 199-211) and inde xes.
Subjects:       Statistics.
<http://catalog.loc.gov/cgi-bin/Pwebrecon.cgi?SC=Subject&SEQ=20061130105250&PID=1830&SA=Statistics.>

LC Classification:      QA276 .A22 1995
Dewey Class No.:        001.4/22 20


------------------------------------------------------------------------
CALL NUMBER:    QA276 .A22 1995
<http://catalog.loc.gov/cgi-bin/Pwebrecon.cgi?SC=CallNumber&SEQ=20061130105250&PID=1830&SA=QA276>





Art Kendall
Social Research Consultants
Edward Boadi wrote:

>Hi  List.
>What will be a good statistical reference  book for a Statistician who has just joined a
>Statistical Research consulting firm with specialization in Research Design and Statistical Analysis.
>Regards.
>
>
>
>

[Mark A Davenport MADAVENP]

I have always liked the Cohen and Cohen book 'Applied Multple
Regression/Correlation Analysis for the Behavioral Sciences'.   I have an
older copy, don't know if it's even in print anymore.  I also like
Pedhazur's 'Multiple Regression in Behavioral Research.   For
multivariate, Steven's 'Applied Multivariate Statistics for the Social
Sciences'.

I have always been a big fan of the Norusis books and am VERY glad to see
that she is writing again.

***************************************************************************************************************************************************************
Mark A. Davenport Ph.D.
Senior Research Analyst
Office of Institutional Research
The University of North Carolina at Greensboro
336.256.0395
[hidden email]

'An approximate answer to the right question is worth a good deal more
than an exact answer to an approximate question.' --a paraphrase of J. W.
Tukey (1962)






Anthony Babinec <[hidden email]>
Sent by: "SPSSX(r) Discussion" <[hidden email]>
11/30/2006 10:12 AM
Please respond to
Anthony Babinec <[hidden email]>


To
[hidden email]
cc

Subject
Re: Statistics Book






Two that I like are:

Gerald van Belle's "Statistical Rules of Thumb" published
by Wiley - Some good material on research design and consulting
in there.

Richard Berk's "Regression Analysis: A Constructive Critique"
published by Sage -- Not a reference book on regression, but
instead a good read to get you thinking about the logic of
analysis and the use of regression and related methods on
observational data.

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
Edward Boadi
Sent: Wednesday, November 29, 2006 1:55 PM
To: [hidden email]
Subject: Statistics Book

Hi  List.
What will be a good statistical reference  book for a Statistician who has
just joined a
Statistical Research consulting firm with specialization in Research
Design
and Statistical Analysis.
Regards.

[Maguin, Eugene]

All,

I have some questions about cluster analysis, which I have never done or
used before. I am looking for several things. First, an assessment of
whether cluster analysis would be useful given my data and interests. If
clustering seems feasible, then a recommendation on procedure selection (the
discussion of the past few days on clustering has been helpful) and
recommendations of references.

I have attendance data (yes/no) for 350 persons attending a 14 session
program. I have previously analyzed this data by just counting the number of
sessions attended. I am interested to see if the data can be clustered by
the 14 session attendance variables to yield a different and more
interesting view of the dataset. I don't know that it matters but total
attendance ranges from 0 to 14 with about 14% attending 0 sessions, 10%
attending 14 sessions and between about 3% and 10% attend each number of
sessions from 1 to 13. Given all this, is clustering even feasible? If so,
what method would be recommended? And, references?

Thanks, Gene Maguin

[David Hitchin]

--On 05 December 2006 11:07 -0500 Gene Maguin <[hidden email]>
asked about cluster analysis for attendance data.

I think that there is one important aspect of your data which may be
difficult to include in a cluster analysis. You are recording attendance at
14 sessions of a program, so that probably means that your 14 variables are
ORDERED (assuming that the sessions are a program over a period of time
rather than a "pick and mix" selection of topics.

I can think of some theories that might produce clusters, if people fall
into particular patterns of behaviour. You might think of those people who
start, and then drop out, those that persist to the end, and those with
"irregular" attendance, but it is unlikely that any cluster analysis
program would identify these in any particularly useful way.

You might find that a cluster program finds you several groups of people,
and that then by looking at the data you can puzzle out what characterises
each of the groups.

You have to face the fact that some kinds of data cluster nice and neatly
and stably, while others produce a arbitrary set of solutions which depend
critically on the clustering method used.

Suppose that you have one continuous variable, and when you plot your data
you see two well separated heaps from the normal distribution. Almost any
clustering program will separate these well.

On the other hand, suppose that you see a single normal distribution. How
would you (or the computer) cut this into two groups. In the middle? At one
of the tails? Which tail? Somewhere else?

If the data values are continuous and form a single "clump" then any
attempt to cluster will produce an arbitrary result.

Would you expect your people to fall into discrete groups, or are they just
an amorphous mass?

Are you considering cluster analysis because you can't think of anything
else to do, and it might be worth trying?

The most important omission in your question is a statement of the research
question. Do you have any hypotheses? What questions are you asking of the
data? What do you want to use the results of your analysis for?

I'm sorry to face you with these difficult questions, but if (and only if)
you can answer them, your choice of a method of analysis may be clearer.



David Hitchin

[Bob Schacht-3]

At 08:46 AM 12/5/2006, David Hitchin wrote:
>--On 05 December 2006 11:07 -0500 Gene Maguin <[hidden email]>
>asked about cluster analysis for attendance data.
>
>I think that there is one important aspect of your data which may be
>difficult to include in a cluster analysis. You are recording attendance at
>14 sessions of a program, so that probably means that your 14 variables are
>ORDERED (assuming that the sessions are a program over a period of time
>rather than a "pick and mix" selection of topics. . . .

This is a very important theoretical point, but I have a slightly different
take on the consequences.
Normally, your 14 variables are supposed to be *independent*. But if they
are ordered, as noted above, then which session you see first may well
influence your choice of which session you see next.

Because of this linkage, you may want to consider some kind of path
analysis. I am not at all familiar with the kinds of path analysis programs
SPSS has, so that other than making the suggestion and pointing the way, I
can't be of much use to you. But you might find a path analysis to yield
some very interesting results in your situation.

Bob Schacht



Robert M. Schacht, Ph.D. <[hidden email]>
Pacific Basin Rehabilitation Research & Training Center
1268 Young Street, Suite #204
Research Center, University of Hawaii
Honolulu, HI 96814

[Maguin, Eugene]

David,

Thank you for your quick and comprehensive reply. I'll reply to your
comments in a somewhat different order.

>>Are you considering cluster analysis because you can't think of anything
else to do, and it might be worth trying?

>>The most important omission in your question is a statement of the
research
question. Do you have any hypotheses? What questions are you asking of the
data? What do you want to use the results of your analysis for?

I've already done one analysis where I divided the total number of sessions
into several groups and looked at predictors of group membership. Although I
haven't done this yet, I am planning on looking at attendance as a time to
nonattendance (survival) analysis. When I was talking to someone else about
this, it occurred to me that if I could figure out how to do it, I could
include persons who attended, dropped out, and then started back via a
repeated events survival analysis. Lastly, I had thought about calculating
conditional probabilities. The stimulus for the clustering was an article
that mentioned using clustering. However, silly me, I thought the person had
used a cluster algorithm, instead they did a 'hand-clustering' procedure.
Basically, I'm looking for ways to characterize people's patterns of
attendance. The typical way, I think, what I have already done. I'm
wondering about new ways.


>>I think that there is one important aspect of your data which may be
difficult to include in a cluster analysis. You are recording attendance at
14 sessions of a program, so that probably means that your 14 variables are
ORDERED (assuming that the sessions are a program over a period of time
rather than a "pick and mix" selection of topics.

Yes, they are ordered. However, how does that enter into the underlying math
theory of clustering?

>>I can think of some theories that might produce clusters, if people fall
into particular patterns of behaviour. You might think of those people who
start, and then drop out, those that persist to the end, and those with
"irregular" attendance, but it is unlikely that any cluster analysis
program would identify these in any particularly useful way.

>>You might find that a cluster program finds you several groups of people,
and that then by looking at the data you can puzzle out what characterises
each of the groups.

Yes, that is my basic plan. To identify more-or-less homogenous groups of
people with respect to attendance and then look for demo and substantive
differences. It sounds like you hold little hope of that working because of
difficulty of forming clusters


>>Would you expect your people to fall into discrete groups, or are they
just
an amorphous mass?

I don't know what to expect. I don't know if people are inclined to skip
certain sessions. I had thought of looking at the frequencies of an A14
pattern variable to get an idea of patterns. When I look at the individual
session attendance percentages, I see that session 1 attendance was the
highest and then the percentage dropped steadily until session 4 where it
more or less stabilized with a slight downward slope until rising at session
14.

If you were to TRY a cluster procedure, which would you try and why?

Thanks, Gene Maguin

[Maguin, Eugene]

Bob,

Thanks for your reply.

>>This is a very important theoretical point, but I have a slightly
different
take on the consequences.
Normally, your 14 variables are supposed to be *independent*. But if they
are ordered, as noted above, then which session you see first may well
influence your choice of which session you see next.


After seeing your use of the word independent and the following sentence, I
find myself wondering if I have a misunderstanding of how clustering works.
I had thought that the procedure computes a 'distance' from one case to all
others based on the values of the variables for the cases. A mahalanobis
distance measure for instance. Cases near to each other are clumped together
to make clusters. Have I got this all wrong?

Thanks, Gene Maguin

[Art Kendall-2]

TWOSTEP

Art Kendall
Social Research Consultants

Gene Maguin wrote:

>
>If you were to TRY a cluster procedure, which would you try and why?
>
>Thanks, Gene Maguin
>
>
>
>