Reading SEM and HLM 2: Synthezing ES in HLM

New Developments and Techniques in Structural Equation Modeling (eBook) by Marcoulides, George A.; Schumacker, Randall E. Publication: Mahwah, N.J. Lawrence Erlbaum Associates, Inc., 2001. Chapter 4 Multilevel Modeling With SEM

Ronald H. Heck University of Hawaii at Manoa The author introduces the strength of multilevel modeling when it was used to analyze cluster data. Many references are listed which can put interested researchers up-to-date on how multilevel modeling developed. Specifically these references can illuminate why and how a family of multilevel models can flourish, in several fields of study in two decades, through powerfully handling the analysis units, providing coefficient estimation and variance component at individual level and organization level, in which individuals nested. This is helpful to students who is studying HLM, which is a very popular multilevel modeling method in social sciences, such as, education and psychology. [Multilevel modeling process may imply some clue for effect size Start model: Null-indicator ANOVA Following: one-factor ANOVA (indicator-added model) Random-effect added model . . . Final model Every two tandem models provide variance change and coefficient change. The inference test of coefficients and test of variance component may be linked to effect size computation for a interested factor. (How? Mixed model, i.e., combined equation , may not be a good way to go. Specifically, in each level, indicators could have corresponding individual effect sizes. . How to defined a effect size index in HLM? When synthesize effect sizes for an indicator across studies, what kind of rules should we follow? Strict rule: Model structure should be identical in previous studies. Loose rule: the target indicator should come same level in previous studies. ) ]

BOOK REVIEW (From SEM a Multidisciplinary Journal)

Testing Structural Equation Model. Kenneth A. Bollen and J. Scott Long (Eds. ) Newbury Park, CA: Sage, 1993, 320 page ( paperback) It is a volume can put interested scientists up-to-date on how apply SEM on problems. Broad ideas from researchers in the world about how SEM should be evaluated were introduced including Joreskog’s thought on then state of affaire in goodness-of –fit. Tanaka provides a alternative indices for goodness-of-fit. Bollen and Stine present their bootstrapping methodology for assessing goodness of fit. How Bayesian information criteria was used to select competing SEMs was talked by Rafferty. Bentler and Chou talked about new statistics to guide the respecification for a model. Joreskog, finally, outlined the problem of translating theory into a statistical model. A first course in factor analysis (2nd, ed) Andrew L. Comrey and Howard B. Lee, 1992, 430 pages. (SEM 1(3),279-282) Reviewed by Scott L. Hershberger “History tells us that methodological techniques, no matter their apparent sophistication, quick fall by the wayside if not useful for the analysis of human behavior.” Heather E. Bullock , Lisa L. Harlow and Stanley A. Mulaik Causation issue in structural equation modeling research How to understand “causal” in causal modeling? SEM generally was used to analyze the nonexperimental data that have traditional been analyzed using multivariate analysis of variance (MANOVA) and multivariate analysis of covariance. (p256) It is important to reiterate that no statistical routine (e.g. MANOVA, SEM) – by itself – can establish causation; causal potential is determined by the degree of control and validity built into the research design. (p257)

Greek Letters

HLM6 1: HLM 6 new features

• Graphical feature : box-plot, scatter-plot, cubic splines for original data can be plotted and grouped by higher-level factors. Also, the plots are color-coded. Also, residuals of level-1 can be color-plotted within higher level factors.
• Model presentation: Multiple-level model equations (several separate model equations) and mixed model equation (one combined equation) can be present with or without subscripts. Distribution assumptions and link functions can be presented in detail, in which even the factor’s centered-method (group or grand mean centered) can be distinguished.
• New algorithm method, such as Laplace approximation, is applied for stable convergence and accurate estimation in two-level hierarchical generalized linear models (HGLM).
• Data can be imported from different types of statistical package, such as, SAS, SPSS and Excel. And residuals can be easily saved as other data files, such as .sav and .dta etc..

Reading Kadell's 1: minor and cofactor

Minor of a matrix, D, is defined as the determinant of A, which includes some rows or columns of D. However, concept minor used in partial correlation computation is equivalent to cofactor.

Maurice Kendall's Advanced Theory of Statistics

Kendall’s advanced theory of statistics is a very good book, which includes three volumes, distribution theory, Bayesian inference and classical inference and linear model. Distribution theory now is on 6the edition. Volume 2b, Bayesian inference, is the most popular topic in statistical field. The classical inference and the linear model, volume 2a, talks about linear model, partial and multiple correlation and regression. The price for volume 1 6th edition is $104 though it is used. The latest set includes the brand new second edition of the popular "Volume 2B: Bayesian Inference," along with the sixth editions of "Volume I: Distribution Theory," and "Volume 2A: Classical Inference and the Linear Model." This set offers the complete, classic Kendall's Advanced Theory of Statistics in a single value-for-money pack (published in May, 2004, ISBN 0340814934). Total cost for three volumes is about $262. Another version of Kendall’s advanced theory of statistics arranged three books, distribution theory, inference and relationship, and Bayesian analysis, v1, v2 and v3, respectively, published in 1970s.

First picture. 

My trip to Hawaii