<- file stat .html -> ********************** Items and scales *******************

Missing values. Myles.

=====================Andy Myles, 21 Sept 1995========ssc From: ajmy@festival.ed.ac.uk (A Myles) Subject: Re: Missing value imputation Message-ID: <43rbqn$au3@scotsman.ed.ac.uk> Dennis Roberts <DMR@PSUVM.PSU.EDU> writes: >seems to be most appropriate. Nevertheless, any inserted value is >clearly fraught with the potential of being nowhere close to what it >WOULD have been if legitimately gathered. Is anyone aware of any work on the use of robust regression methods to reduce the impact of badly imputed values? My first encounters with robust regression actually arose from wondering about just this problem, but I never eneded up going back to the imputation side of things... >The other concern is how many cases are missing? If a very small >percentage of the total data set, then using any of the methods probably >does not do much to add to the vitality of the data. But, if the >percent of data that are missing is rather large, then using any method >of replacement becomes highly problematical. Indeed. For the original poster, I've appended some references I collected at the time which will hopefully be useful. --------- %A M.J. Box %A N.R. Draper %A W.G. Hunter %T Missing Values in Multiresponse Nonlinear Model Fitting %J Technometrics %V 13 %N 3 %D 1970 %P 613-620 %K missing %A D.B. Rubin %T Comparing Regressions When Some Predictor Values Are Missing %J Technometrics %V 18 %N 2 %D 1976 %P 201-205 %K missing %A R.J.A. Little %T Regression With Missing X's: A Review %J Journal of the American Statistical Association %V 87 %N 420 %D 1992 %P 1227-1237 %K missing review %A A.A. Afifi %A R.M. Elashoff %T Missing Observations in Multivariate Statistics. I. Review of the Literature %J Journal of the American Statistical Association %V 61 %N 315 %D 1966 %P 595-604 %A S.F. Buck %T A Method of Estimation of Missing values in Multivariate Data suitable for use with an Electronic Computer %J Journal of the Royal Statistical Society (Series B) %V 22 %N 2 %D 1960 %P 302-306 %A R.J.A. Little %A D.B. Rubin %T Statistical Analysis with Missing Data %I John Wiley and Sons, Inc %D 1987 %O ISBN 0-471-80254-9 %K missing %A D.B. Rubin %T Multiple Imputation for Nonresponse in Surveys %I John Wiley and Sons, Inc %D 1987 %O ISBN 0-471-80705-X %K missing -------- Some variations on the use more ML methods here ------- %A S. Ahmad %A V. Tresp %T Some Solutions to the Missing Feature Problem in Vision %B Advances in Neural Information Processing Systems, 5 %E S.J. Hanson %E J.D. Cowan %E C.L. Giles %I Morgan Kaufmann Publishers, Inc %D 1993 %P 393-400 %O ISBN 1-55860-274-7 %K missing neural %A V. Tresp %A S. Ahmad %A R. Neureier %T Training Neural Networks with Deficient Data %B Advances in Neural Information Processing Systems, 6 %E J.D. Cowan %E G. Tesauro %E J. Alspector %I Morgan Kaufmann Publishers, Inc %D to be published %K missing neural %A S. Ahmad %T Feature Densities are Required for Computing Feature Correspondences %B Advances in Neural Information Processing Systems, 6 %E J.D. Cowan %E G. Tesauro %E J. Alspector %I Morgan Kaufmann Publishers, Inc %D to be published %K missing neural * * * * * * * * * * * * * * * * * * * * * * * * * * * * *

Cronbach's coefficient alpha

=======================Eugene Komaroff, 8 Feb 1996==========ssc Message-ID: <9601088238.AA823800221@mednet.med.miami.edu> From: Eugene Komaroff <ekomarof@MEDNET.MED.MIAMI.EDU> Some recent references to coefficient (Cronbach's) alpha: Cortina, J.M. (1993). What is coefficient alpha? An examination of theory and applications. Journal of Applied Psychology, 78, 98-104. Miller, M.B. (1995). Coefficient alpha: A basic introduction from the perspectives of classical test theory and structural equation modeling. Structural Equation Modeling, 2(3), 255-273. Reuterberg, S.E., & Gustafsson, J.E. (1992). Confirmatory factor analysis and reliability: Testing measurement model assumptions. Educational and Psychological Measurement, 52, 795-811. Zimmerman, D.W., Zumbo, B.D., & Lalonde, C. (1993). Coefficient alpha as an estimate of test reliability under violation of two assumptions. Educational and Psychological Measurement, 53, 33-49. Isn't it amazing that coefficient alpha has been around for about 60 years [Kuder & Richardson, 1937 (KR20)], and articles continue to be written explaining what it is, how it works and when to use it? ===================Paul Barrett, 10 Feb 1996==========ssc Message-ID: <Pine.SOL.3.90.960210143421.27106A-100000@cantua> From: Paul Barrett <psyc379@CANTUA.CANTERBURY.AC.NZ> Try these ... Cortina, J.M. (1993) What is coefficient alpha? An examination of theory and applications. Journal of Applied Psychology, 78, 1, 98-104 Feldt, L.S., Woodruffe, D.J., and Salih, F.A. (1987) Statisrical Inference for Coefficient Alpha. Applied Psychological Measurement, 11, 1, 93-103. Green, S.B., Lissitz, R.W., and Mulaik, S.A.(1977) Limitations of coefficient alpha as an index of test unidimensionality. Educational and Psychological Measurement, 37, 827-838. Hattie, J. (1985) Methodology Review: Assessing unidimensionality of tests and items. Applied Psychological measurement, 9, 2, 139-164. Miller, M.B. (1995) Coefficient Alpha: a basic introduction from the perspectives of classical test theory and structural equation modelling. Structural Equation Modelling,2, 3, 255-273. Regards ... Paul * * * * * * * * * * * * * * * * * * * * * * * * * * * * *

Testing Rausch assumptions

=====================Jean-Luc Kop, 30 May 1996======(spss) Message-ID: <199605301643.SAA12981@clsh.u-nancy.fr> Date: Thu, 30 May 1996 18:43:10 +0200 From: Jean-Luc KOP <kop@CLSH.U-NANCY.FR> >>Is anyone aware of a SPSS Module or previously written program that >>allows a series of test or battery items to be subjected to Rausch >>Analysis? >> >>We also use SAS if anyone knows of an applicable PROC or program. > >I'm not aware of anything in either package, or macros written for >either package, to do Rasch analyses. I have seen papers that cast >the Rasch model in terms of a loglinear model, and of course both >packages will handle loglinear modeling. This would require a fair >amount of sophistication on the part of the user though. > >-- >----------------------------------------------------------------------------- >David Nichols Senior Support Statistician SPSS, Inc. >Phone: (312) 329-3684 Internet: nichols@spss.com Fax: (312) 329-3668 >----------------------------------------------------------------------------- See perhaps: Ten Vegert, E., Gillespie, M. & Kingma, J. (1993). Testing the asumptions and interpreting the results of the Rasch model using log-linear procedures in SPSS. Behavior Research Methods, Instruments, and Computers, 25, 350-359. Good luck, * * * * * * * * * * * * * * * * * * * * * * * * * * * * *

Likert scaling.

=====================Don Johnson, 2 Mar 1995===========??? From: "Don Johnson" <DJOHNSON@WTAMU-EDUCATION.WTAMU.EDU> Subject: Likert Scaling There is a long and venerable history of using ordinal items, summing them to produce a scaler value, and then treating that value as if it were really an interval score. That is essentially what the likert scoring system is all about. For example, the Rosenberg self-esteem scale was originally designed as a Guttman scale, but it has become increasingly popular to score with the Likert type system. For a review and evaluation of this, check the following reference. Wallace, G.R. (1988). RSE-40, An alternative scoring system for the Rosenberg Self-esteem Scale (RSE). (Report, ERIC Document Reproduction Service No. ED 298 154). Some other references you might look up are: Likert, R. (1932). A technique for the measurement of attitudes. Archives of Psychology, 140. Stouffer, S.A., Borgotta, E.F., Hays, D.G. & Henry, A.F. (1952). A technique for improving cumulative scales. Public Opinion Quarterly, 16, 273-291. Torgerson, W. (1958). Theory and methods of scaling. NY: Wiley. * * * * * * * * * * * * * * * * * * * * * * * * * * * * *

... compute an ICC with unequal N? (intraclass correlation)

=====================Michael Bailey, 11 Oct 1995========ssc From: jm-bailey@nwu.edu (Michael Bailey) Subject: Re: reliability query Message-ID: <jm-bailey-1110951430340001@mac189.psych.nwu.edu> In article <bermanjs.1163706761A@msuvx2.memphis.edu>, Jeffrey Berman <bermanjs@MSUVX2.MEMPHIS.EDU> wrote: <snip> > The F ratio from this ANOVA would be the test of the intraclass correlation. > The only complication in calculating the intraclass correlation itself is > that you appear to have unequal numbers of estimates within each company. A > method for computing the intraclass correlation in the case of unequal class > membership is outlined in the classic reference by Haggard (1958, chap. 2). > I would be interested if others on the list have more recent references for > calculating the intraclass correlation in the case of unequal class membership. Thanks Jeff, you're right, and the correct way to do it is on page 14 of Haggard. Let, R=intraclass correlation, BCMS=between classes mean square, WMS=mean square within, c=number of classes, and ki=number in the ith group. THen: R = BCMS - WMS __________ BCMS + (k~-1)WMS where k~= 1 _____ * (sigma ki - (sigma (ki**2))/sigma ki) (sigma means summation) c-1 * * * * * * * * * * * * * * * * * * * * * * * * * * * * *

Whatever happened to Guttman analyses?

=====================David Nolle, 29 Feb 1996========(spss) Message-ID: <Pine.SOL.3.91.960229061057.6287A-100000@clark.net> From: "David B. Nolle" <dbnolle@CLARK.NET> Subject: Re: Guttman Scalogram > Subject: Guttman Scalogram > > Does anyone have a brief explanation of why Guttman scalograms have been > dropped from recent versions of SAS and SPSS? One of my dissertation The deterministic version of Guttman scaling has been replaced with various probabilistic generalizations of the Guttman model. A good outline of the replacements can be found in the following article: C. C. Clogg and D. O. Sawyer, A Comparison of Alternative Models for Analyzing the Scalability of Response Patterns, Pages 240-280 in S. Leinhardt (Editor) Sociological Methodology 1981. San Francisco: Josey-Bass. Professor Clogg is deceased, but one of his former students, Dr. Scott Eliason, has incorporated Clogg's MLLSA into an excellent PC-based suite of computer programs. MLLSA is a computer program which can be used to estimate the various genralizations of the Guttman model. My guess is that you might be able to reach Dr. Eliason at the following address: seliason@blue.weeg.uiowa.edu If you are really pressured for immediate results, you might want to consider using loglinear analysis within SPSS to get results on Leo Goodman's 1975 version of the Guttman model described in volume 70 of the Journal of the American Statistical Association. Good Luck, David ============Paul Barrett, 01 Mar 1996========ssc Message-ID: <Pine.SOL.3.90.960301190909.19804B-100000@cantua> From: Paul Barrett <psyc379@CANTUA.CANTERBURY.AC.NZ> Subject: Re: Guttman Scalograms > For that matter, does anyone know where one can > obtain software for Guttman's later MSP > (Multidimensional Scalogram) analyses? > Sean Hammond has posted 3 programs bound up in a facet analysis suite of programs - Smallest Space Analysis, Multidimensional Scalogram Analysis, and Partial Order Scalogram Analysis. The programs are on the idanet (Individual Differences and Assessment Net) executable program filestore at... http://www.canterbury.ac.nz/psyc/barrett/programs.htm under the heading FACET analysis. A full description of each program is given - with references and other info. The file can be downloaded via your web browser. Regards .. Paul Barrett ===============David Nolle, 02 Mar 1996========(spss) Message-ID: <Pine.SOL.3.91.960302084827.25769B-100000@clark.net> From: "David B. Nolle" <dbnolle@CLARK.NET> Subject: Re: Guttman Scalogram (More) Jae, The deterministic version of the Guttman model has indeed fallen from favor. However, if a probabilistic version of the Guttman model seems to be a reasonable approach to your data, then you might want to know that some probabilistic extensions of L. Guttman's deterministic (scalogram) model have been generalized to the multiple-group case. In your situation, multiple groups might be different age groups and/or different time periods. Some multiple-group generalizations of probabilistic extensions of the Guttman model are carefully developed in the following article: C. C. Clogg and L. A. Goodman, On Scaling Models Applied to Data From Several Groups, Psychometrika, 51, (March, 1986), 123-135. Scott Eliason's PC version of C. C. Clogg's Maximum Likelihood Latent Structure Analysis (MLLSA) computer program ought to estimate the parameters which you need. Allan McCutcheon's 1987 Sage publication (titled Latent Class Analysis) provides a very readable introduction to some of the probabilistic extensions to the Guttman scale model. Good Luck, David * * * * * * * * * * * * * * * * * * * * * * * * * * * * *

Should I use kappa, other than nominal data?

========================Rich Ulrich, 02 Sep 1996======ssc From: wpilib+@pitt.edu (Richard F Ulrich) Subject: Re: kappa Message-ID: <50fa45$2t8@usenet.srv.cis.pitt.edu> Robin (RLHEAT00@UKCC.UKY.EDU) wrote: : Does anyone have opinions on kappa being used on data other than nominal data? In short, Don't do it. In fact, if you are trying to learn something, don't use it on nominal data if there are more than two categories. You will find more information by looking at the categories in pairs. (It might be okay for summarizing what you find, if there is nothing else of interest.) : Someone has suggested that I use kappa for interrater reliability on a series o : f Likert scales. Thanks in advance for any advice. I would rather look at paired t-tests, along with Pearson correlations, in this situation. The t tells me whether raters score at the same level, while the correlation (which SPSS handily provides) shows reliability. If there are 3 or more raters, it is also more informative to look at the pairs of raters than to look at the intraclass correlation (which may be useful to summarize). * * * * * * * * * * * * * * * * * * * * * * * * * * * * *

Document by Rich Ulrich. E-mail to wpilib+@pitt.edu

FAQ top.

Ulrich home page.

Ulrich FAQ. http://www.pitt.edu/~wpilib/stats99.html