Subject: Re: Cronbach alpha
From: Rick DeShon
Date: Tue, 26 Nov 1996 08:01:54 -0500
Dennis Roberts wrote:
>Cronbachs alpha is not really a type of correlation ... but rather an index
>of internal consistency reliability. One reference would be: Hoi Suen
>(1990), Principle of test theories, Lawrence Erlbaum, Hillsdale: NJ ... see
>pages 35, 57, and 61.
Although alpha is used as an index of internal consistency it is a type of
correlation. Specifically, it is an intra-class correlation. Shrout and
Fleiss (1979) Psychological Bulletin and McGraw and Wong (1996)
Psychological Methods highlight this interpretation.
Rick
Richard DeShon
Dept. of Psychology
Michigan State University
East Lansing, MI 48824-1117
E-mail: deshon@pilot.msu.edu
Voice: (517) 353-4624
Fax: (517) 353-4873
Subject: help: robust estimation
From: Roel.Leus@student.kuleuven.ac.be (Roel Leus)
Date: 26 Nov 1996 13:40:48 GMT
I am a fourth year student of Trade Engineer at the KU Leuven.
In the framework of a paper I am to write on detecting outliers
in multivariate samples, I am at the moment looking into robust
estimation of multivariate location and shape, with which I could
calculate the Mahalanobis-distances of each point to the center.
I already found : MVE, MCD, M- and S-estimators.
Are there other suggestions, and what (basic) books and articles
should I consult ?
Subject: The "crud" factor! [Meehl]
From: rwhite@superior.carleton.ca (Robert White)
Date: 26 Nov 96 14:17:42 GMT
Pervin, L.A.(1990)_Handbook of Personality Theory & Research_
The Guilford Press New York
"As Meehl [in press] notes, our fascination with
correlations and tests of statistical significance
often leads us to assume progress in understanding
a phenomenon when none is present; this is due in part
to what he calls the "crud" factor. The history of
fads in personality research, noted in the intro-
ductory chapter, is undoubtedly relevant here." p724
Date: 25-Nov-96
Database: PsycINFO <1984 to November 1996>
Set Search Results
---------------------------------------------------------------------------
005 crud factor.tw. 2
<1>
Accession Number
Journal Article: 79-18428.
Authors
Standing, Lionel; Sproule, Robert; Khouzam, Nelly.
Institution
Bishop's U, Lennoxville, PQ, Canada.
Title
Empirical statistics: IV. Illustrating Meehl's sixth law of soft
psychology: Everything correlates with everything.
Source
Psychological Reports. Vol 69(1) 123-126, Aug 1991.
Abstract
Computed a 135 * 135 matrix of correlations, using
educational/biographical data on 2,058 Canadian grade-school children, to
find the background level of statistically significant correlations, or P.
Meehl's (1990) "crud factor." All of the 135 variables, except student
identification number, displayed more statistically significant
correlations with the other variables than could be predicted from chance.
Results reinforce Meehl's criticisms of overreliance on significance
levels for correlational data. (PsycINFO Database Copyright 1992 American
Psychological Assn, all rights reserved).
<2>
Accession Number
Chapter: 91-011021-001.
Authors
Meehl, Paul E.
Institution
U Minnesota, Professor of Psychology, MN, US.
Chapter Title
Why summaries of research on psychological theories are often
uninterpretable. [References].
Book Citation
Improving inquiry in social science: A volume in honor of Lee J.
Cronbach. (Richard E. Snow, David E. Wiley, Eds.), pp. 13-59.
Source
Lawrence Erlbaum Associates, Inc, Hillsdale, NJ, US; xiv, 423 pp. 1991.
Content Representation
(from the book) begins ...with a general rejection of the hypothesis
testing tradition that has dominated so much of psychology and social
science # argues that most of our research literatures are rendered
uninterpretable because of a series of obfuscating factors the effects of
which are often sizable, variable, operating in opposite directions, and
left unmeasured # urges that we come to full appreciation of how little we
really know, and how little can ever be learned through null hypothesis
testing. (from the chapter) (discusses) surveys of research evidence
sharing three properties ...(a) theories in so called " soft areas,"
(namely, clinical, counseling, personality theory, and social psychology),
(b) data correlational, and (c) positive findings consisting of refutation
of the null hypothesis /// ten obfuscating factors that make (null
hypothesis) refutation in the soft areas largely uninterpretable # loose
derivation chain # problematic auxiliary theories # problematic ceteris
paribus clause # experimenter error # inadequate statistical power # crud
factor # pilot studies # selective bias in submitting reports # selective
editorial bias # detached validation claim for psychometric instruments.
---------------------------------------------------------------------------
--
----------------------------------------- Carleton University ----------
Robert G. White Dept. of Psychology
Ottawa, Ontario. CANADA
INTERNET ADDRESS ----- rwhite@ccs.carleton.ca ------------------- E-MAIL
------------------------------------------------------------------------
Subject: Re: Cronbach alpha
From: Dennis Roberts
Date: Mon, 25 Nov 1996 23:38:36 -0500
Not to beat this to death but ... alpha may be one application of the intra
class correlation but ... correlations of this type are not restricted to
estimates of reliabilty ... so, in that sense, I think it is better not to
think of it AS a type of correlation ...
When we talk about "types" of rs ... we tend to think of Pearson r ... or
Spearman rho ... or a tetrachoric r ... etc. .... and Cronbach's alpha seens
not to be in the same spirit of "type" as these.
Another way to describe this is ... when we talk in a stat course in a
chapter on rs ... we would never bring up Cronbachs alpha ...
At 08:01 AM 11/26/96 -0500, you wrote:
>Dennis Roberts wrote:
>
>>Cronbachs alpha is not really a type of correlation ... but rather an index
>>of internal consistency reliability. One reference would be: Hoi Suen
>>(1990), Principle of test theories, Lawrence Erlbaum, Hillsdale: NJ ... see
>>pages 35, 57, and 61.
>
>Although alpha is used as an index of internal consistency it is a type of
>correlation. Specifically, it is an intra-class correlation. Shrout and
>Fleiss (1979) Psychological Bulletin and McGraw and Wong (1996)
>Psychological Methods highlight this interpretation.
>
>Rick
>
>
>Richard DeShon
>Dept. of Psychology
>Michigan State University
>East Lansing, MI 48824-1117
>E-mail: deshon@pilot.msu.edu
>Voice: (517) 353-4624
>Fax: (517) 353-4873
>
>
===========================
Dennis Roberts, Professor EdPsy !!! GO NITTANY LIONS !!!
208 Cedar, Penn State, University Park, PA 16802 AC 814-863-2401
WEB (personal) http://www2.ed.psu.edu/espse/staff/droberts/drober~1.htm
Subject: Seeking statistician for software development in Cambridge MA
From: Tom Lane
Date: Tue, 26 Nov 1996 11:15:40 -0500
Domain Solutions Corporation is seeking a statistician to help us
develop software for analyzing data in a manufacturing environment.
This statistician will
- Define the statistical content of software releases.
- Assist the quality assurance group in designing, carrying
out, and validating tests.
- Write software using C++ for Unix and Windows.
In addition, this person may occasionally be called upon to write
technical documentation, consult with software engineers, perform
competitive analysis, support the sales organization, and interact
with customers.
Qualifications:
Required:
- Masters in statistics.
- Knowledge of applied statistics areas such as regression,
quality control, design of experiments.
- Experience in software development.
Preferred:
- Ph.D. in statistics or a related field.
- Experience in software testing.
- Experience applying statistics in a manufacturing environment.
Domain Solutions Corporation, formerly a division of BBN, develops
and markets software to optimize manufacturing processes and to
manage clinical trials data. Our products include RS/1,
RS/Explore, RS/Discover, Cornerstone, Probe, Patterns, and Starfire.
You can apply for this position by sending a cover letter and resume
via email to tlane@domaincorp.com, by fax to 617-873-8199, or by
mail to Tom Lane, Chief Statistician, Domain Solutions Corporation,
150 CambridgePark Dr., Cambridge, MA 02140. Please send email in
the form of plain text, postscript, or Microsoft Word.
Subject: Can you please help?
From: Daniel Aufgang
Date: Tue, 26 Nov 1996 18:08:09 +0200
Please send answers to: aufgang@netmedia.net.il
Hello. First off, I am sorry about any cross posts and if this letter
is not relevant for this list. I am finishing up my thesis on "The
Subjective, Objective and Behavioral Environment in Urban Planning" and
am now stumped on the statistics part. In short, I know I have to use
multi dimensional scaling, but I do not know about all the sub-types
and their nuances. I feel, like factor analysis, that if you do not
know the methods inside and out, you are run into trouble when "those
who do" read over your =FDwork. I am using Statistica 5.0 and Excel for
the analysis.=20
Here is my problem. I conducted an open-ended survey which I am using
to conduct a close ended survey. I asked questions like "What words
would you use to describe the =FDlayout of the mall, the look of the
mall.." etc. In each question category, I compiled =FDabout 20-50 words.=
=20
Here are the responses to one of the questions; What is the main reason
you came here today. (translated from Hebrew for obvious reasons)
-to buy certain products
-movies
-window shopping
-general shopping
-eat
-drink coffee
-get out of the house
-center for entertainment
-walk around
-trip
-get services
-to do everything at once
-bank
-to enjoy the atmosphere
-meet friends
-pass the time
-to see things
-entertainment
-meet girls
-for no reason
-play
-rest
-for the children
-see what's new
-get bored
I put them into the following groups:
-meet friends
-> I am using the Kenyon as a place to meet my friends.
-general shopping
-to buy certain products
-get services
-to do everything at once
-bank
->I came here to do my shopping
-entertainment
-pass the time
-to see things
-window shopping
-get out of the house
-center for entertainment
-see what's new
-trip
-for the children
-to enjoy the atmosphere
-rest
-play
-meet girls
->I came here to have an enjoyable time.
-drink coffee
-I came here to eat
->I came here for the resturaunts and cafes.
-movies
->I came here to to go a movie
I then asked 20 people: "These are reasons why people claimed they were
in the Kenyon HaArem the day of our survey. Please group the reasons
into logical groups. Try to create between 3 to groups." The groups
differed from what I chose. For example, one student put "To get
services" & "bank" in their own category of SERVICES and "to buy certain
products" "general shopping" "drink coffee" "eat" and "to do everything
at once" under the category "To perform specific functions"
I then made a matrix which shows how often each word appears with each
other. (triangle matrix)
Now what? Is my question worded ok? What is the best way of creating
the matrix? I did it in Excel and it took my a very long time. I will
soon have MANY MANY more questionnaires to enter. What is the most
valid way of grouping this data? In the end, I want groups like I
created. For the other questions, these will be turned into Personal
Constructs.
Subject: Re: Adequacy of sample size
From: Dennis Roberts
Date: Mon, 25 Nov 1996 23:33:51 -0500
The fact is ... this IS complicated. In your case, to make it as simple as
possible .. you are looking for a correlation between some predictor and
some criterion. Well ... how BIG should that be .... for you to think the
relationship is strong enough for you to conclude that there is an IMPORTANT
relationship between X and Y? This is the first and most fundamental
question that has to be answered ... and the answer is NOT ... big enough to
be significant. Then ... how relatively critical would it be to make errors
of the type I or type II variety ... ie, is it relatively more serious (or
equal) to conclude that there is a relationship when perhaps there is not
... or to conclude there is NO relationship when in fact there is?
Only after you answer a series of questions like the above ... can you then
start to attempt to answer the question: how large of a sample do I need to
find a relationship as big as I have said is important ... given that I am
willing to tolerate some percent of making each of the errors that can be
made ...
As I said ... it ain't simple.
At 10:14 PM 11/25/96 GMT, you wrote:
>I've asked in another post how one tells how large a sample size to
>use to get significant results. The responses I've gotten include:
>(1) ask a statistician, and (2) it's too complicated to tell without
>knowing what you're trying to prove. So here's my research question
>and how I intend to go about answering it. If you have the time, I
>welcome any suggestions--especially with regard to how many patients
>I'll need to include in the study to get significant results. Thanks.
>
>At my hospital (where I'm a resident) we admit patients through the
>emergency room to the family practice unit. Some of those patients
>come in with respiratory diagnoses (such as pneumonia and asthma).
>Occasionally, a patient's condition will deteriorate, and he'll have
>to be transferred to the medical intensive care unit (MICU) for more
>intensive care (such as being placed on a ventilator). This is true
>for patients with all conditions, but my study only deals with
>respiratory conditions.
>
>It would be in the interests of better patient care if we could
>identify in the emergency room (before the patient is ever admitted to
>the hospital) which respiratory patients are likely to be transferred
>to the MICU. That way, we could admit them directly to the MICU or,
>more commonly, place them in an intermediate-care unit (what we call a
>"step-down" unit).
>
>So the research question is: which respiratory patients admitted to
>the family practice unit are likely to be get worse rather than
>better, to the point of needing transfer to the MICU?
>
>We have identified about 20 variables which we would like to test.
>These include age, gender, blood oxygenation (how much oxygen is
>actually getting into the body through the diseased lungs),
>co-existing medical conditions, previous admissions to the hospital
>for respiratory problems, previous admissions to the MICU for
>respiratory problems, and many other factors.
>
>The data will be gathered as follows: When the patients are admitted
>to the family practice unit, a researcher will gather the values of
>each of the twenty variables. These values will be obtained from the
>chart and, if necessary, from the patient. The researcher will not be
>anyone involved in the patient's care.
>
>After the patient has been discharged from the hospital, the
>researcher will pull the patient's chart and find out whether the
>patient went to the MICU. This will be the outcome variable.
>
>After data on a sufficient number of patients is gathered, the SAS
>program will be used to figure out which if any of the variables are
>predictors of decline in respiratory patients. Those variables which
>are predictors will then be incorporated into a formula whose result
>will be an "index of decline in respiratory patients admitted to the
>family practice unit through the emergency room".
>
>In the second phase of the study, data will be gathered in a similar
>manner, with the exception that only those variables incorporated into
>the index will be needed. When data on a sufficient number of
>patients is gathered, the SAS program will be used to look for a
>correlation between the index and the outcome.
>
>Thanks for evaluating my study design. Your suggestions are welcome
>and appreciated.
>
>Also, I'd be most interested in any explanations you care to offer
>regarding, in a general way, how one goes about determining adequacy
>of sample size.
>
>Matt Beckwith, M.D.
>Jacksonville, Florida
>
>
===========================
Dennis Roberts, Professor EdPsy !!! GO NITTANY LIONS !!!
208 Cedar, Penn State, University Park, PA 16802 AC 814-863-2401
WEB (personal) http://www2.ed.psu.edu/espse/staff/droberts/drober~1.htm
Subject: Help with Probability on Temperture variations
From: Jon Sauvageau
Date: Tue, 26 Nov 1996 11:10:22 -0800
In article <329A5D82.2741@gsfc.nasa.gov> you write:
>R wrote:
>>
>> In article <3291F61B.E13@gsfc.nasa.gov>, Jon Sauvageau says:
>> >
>> >Currently I am doing a small school project that deals with temperature variences inside a building vs what the outside temp is doing in relation to the cost of energy (electricity). Has anyone done something similar and would share your data/process. I need help in setting up tests such as Chi Square or some other hypothesis test. If you can help in anyway, point me to a BB or maybe suggest a book/artical. Thanks in advance.
>> >
>> >Jon Sauvageau
>>
>> If you can find the book "Facts From Figures" by M.J. Moroney, which is a paper back and quite cheap, I think this will help enormously.
>> R.
>Can you help me with this, I have looked in several libraries and book
>stores to no avail. Do you have a source for this book?
>
>Please let me know ASAP.
>
>Jon Sauvageau
Last year someone in sci.stat.consult posted a similar request.
If you post your request there, you might get some real help.
Tony Rizzo
If you know of this book or know where I can locate more info please let
me know. Jon.M.Sauvageau.1@gsfc.nasa.gov
Subject: Re: E[ X | X => X^*] = ?
From: Hans-Peter Piepho
Date: Tue, 26 Nov 1996 15:30:52 +0100
>Tatsuo asked:
>
>>Suppose X ~ N(mu, sigma=AC2). Then, what is the formula for
>> E[X|X=3D>X=AC*] : Expected value of X given X is greater than or
>>equal to some fixed number X=AC*
>
>you can find the expected value by evaluating the integral of x*p(x) from
> -mu/sigma=AC2 to oo and multiplying the result with sigma=AC2
>p(x) is the pdf of N(0,1) and is equal to exp(-x*x/2)*(2*pi)=AC-0.5
>
>BTW the indefinite integral of x*p(x) in this case evaluates to -p(x)
>
>The book that I found most thorough in explaining Expected values and
>introductory statistics in general was :
>"Statistical Theory and Methodology in Science and Engineering"
>by K.A.Brownlee, 1960
>In my opinion this book beats any modern intro statistics book hands-down.
>Page 38 is on expected values.
>
>I had asked a similar question in this newsgroup and it took me a good
>week to justify the answer I got but I finally did. Thanks again to Aaron
>Brown who gave me the initial help.
>
>Dimitri
>
>
This is the expected value of a truncated normal with truncation from below=
=20
at X. The truncated normqal is discussed in: Johnson and Kotz 1970=20
Continuous univariate distributions I. p.81
Let f(z) denote the p.d.f of the standard normal and F(z) its c.d.f.
E(X|X>X*) =3D mu + sigmaµ2*f(Z*)/(1-F(Z*))
where Z* =3D (X*-mu)/sigma
_______________________________________________________________________
Hans-Peter Piepho
Institut f. Nutzpflanzenkunde WWW: http://www.wiz.uni-kassel.de/fts/
Universitaet Kassel Mail: piepho@wiz.uni-kassel.de
Steinstrasse 19 Fax: +49 5542 98 1230
37213 Witzenhausen, Germany Phone: +49 5542 98 1248
=20
=20
Subject: Re: Power?
From: wpilib+@pitt.edu (Richard F Ulrich)
Date: 26 Nov 1996 22:18:36 GMT
I will not let certain mis-statements go unmentioned --
Russ Boucher (at764@FreeNet.Carleton.CA) wrote:
<< ... deleted, citation; and some comment, leading to a couple of
CORRECT statements, and then an error. >>
RB: The problem with low power is that you run the risk of indecisive
: results, even if the phenomenon you're investigating is real. Stated
: differently, the effect may well be there, but without adequate power, you
: won't find it.
RB: Also, let's distinguish the difference between a Type I error and
: a Type II error. A Type I error is rejecting the null when it is true. A
: Type II error is failing to reject the null when it is true. Power is
: related to the latter, not the former.
-- By ordinary and by mathematical definition, "failing to reject
the null when it is true" is NOT an error, assuming "it" refers to
the null. "Power is related to the latter..." reflects confusion
which is re-iterated even more clearly in the next Answer:
<< Q. deleted, about power greater of test is for p < .2 compared
to p < .05>>
RB: Alpha is the probability of making a Type I error. Beta is the
: probability of making a Type II error. Thus, power is equal to 1 minus beta.
: In other words, power isn't really related to alpha at all.
-- True, power= (1 minus beta), which is a SIMPLER formula than
the relation of power to alpha -- but it is really WRONG to say
that power isn't related to alpha at all. For a given test, there
is a strict relationship among sample size and effect size,
alpha and beta. There are books that can show you tables, and
computer programs that will plot you curves. And there is an
interesting equivalence which exists by definition: for the
effect size nearing 0, for any N, power = alpha.
Rich Ulrich, biostatistician wpilib+@pitt.edu
Western Psychiatric Inst. and Clinic Univ. of Pittsburgh
Subject: Re: Tricky sampling question
From: T.Moore@massey.ac.nz (Terry Moore)
Date: 26 Nov 1996 21:31:38 GMT
In article <961125095604.69@IASX02.LARC.NASA.GOV>, Ken Rutledge
wrote:
>
> w.r.t. the following discussion "Tricky sampling question"
>
> wrote:
>
> >My question is best presented by an example. Say a runner wants to test
> >whether he can run faster with a particular pair of running shoes than
> >with another pair. He does 10 runs with one pair of shoes and 10 runs
> >with the other pair.
The answer is clear. Use a paired design, one type of shoe on each
foot. If one is really better, the runner will run in circles :-)
Terry Moore, Statistics Department, Massey University, New Zealand.
Imagine a person with a gift of ridicule [He might say] First that a
negative quantity has no logarithm; secondly that a negative quantity has
no square root; thirdly that the first non-existent is to the second as the
circumference of a circle is to the diameter. Augustus de Morgan
Subject: Re: Tricky sampling question
From: mdupin@kodak.com (Mike)
Date: Tue, 26 Nov 1996 11:30:02 -0600
Two references:
In Journal of Quality Technology, January 1996, p. 123, Lloyd Nelson has
an interesting article "Notes on the Use of Randomization in
Experimentation", and in that article he references an article by WJ
Youden that appeared in Technometrics February 1972, "Randomization and
Experimentation".
Quoting the abstract of Youden's article: "Randomization, often specified
as an indispensable requirement in experimental design, is required only
when the order or position of the experimental unit influences the order
of the unit."
For your example, if you believe the day-to-day effects on running
performance (wind, etc.) occur independently, then you might argue that
there is no need to randomize, or that lack of randomization does not
render the t-test invalid.
My interpretation, anyway .....
-----------------
> On Wed, 20 Nov 1996 13:25:27 -0500, "Michael Kleiman (SOC)"
> wrote:
>
> >My question is best presented by an example. Say a runner wants to test
> >whether he can run faster with a particular pair of running shoes than
> >with another pair. He does 10 runs with one pair of shoes and 10 runs
> >with the other pair.
Instead the t-test is being performed on
> >the first 10 runs with each shoe, not on 10 runs randomly selected from
> >all runs. The reason I think I'm still OK though is that while it's not
> >technically a random sample, everything that could affect the running
> >times other than the shoes is uncorrelated with the shoes. For instance
> >the wind could be blowing harder for one of the runs, but this is
> >uncorrelated with the shoes because the wind doesn't decide to blow based
> >on my shoes. If anyone could answer this question for me I'd greatly
> >appreciate it, the more so if you can also give me a reference I can
> >consult. Of course, if doing a significance test in such a case is *not*
> >OK then a lot of real-world testing is going to have to be trashed. After
> >all, it'd be idiotic, not to mention impractical, for the runner to run
> >with the 2 shoes for the rest of his life and then randomly sample 10 runs
> >from the resulting population!
