Appelbaum MI. Bias in the analysis of repeated-measures designs some

Document Sample
Appelbaum MI. Bias in the analysis of repeated-measures designs some Powered By Docstoc
					                                                                                              CC/NUMBER 19

     This Week’s Citation Classic                                                              ~13,198~
F’McCall R B & Appelbaum M 1. Bias in the analysis of repeated-measures designs:
    some alternative approaches. Child Develop. 44:401-15. 1973.
      IFels Research Institute. Yellow Springs. OH and the Psychometric Laboratory. University of
      North Carolina. Chapel Hill. NC]

This paper reports that the commonly used repeat-        interpretation of each approach. The paper
ed-measures analysis of variance makes Important         was similar in style to a section from an àp-
assumptions regarding the variance-covariance            plied statistics text.
structure of the data (e.g.. correlations among             Getting it published, however, was not
repeated measures) that are often not met. Viola-        easy. I wanted it to appear where research-
tions of these assumptions bias the analysis to-
                                                         ers, especially developmentalists, who were
ward rejecting the null hypothesis. Several alter-       most likely to use such designs, would read
native approaches to analyzing such data are             it. Reviewers scratched their heads about
presented from an applied rather than a theoreti-        the appropriateness of such an article for
cal standpoint, and multivariate techniques that         Child Development. But the editor, Wendell
make no covariance assumptions and provide ex-
                                                         Jeffrey of UCLA, was encouraging and pUb-
act probability statements are featured, [The Sc,-
                    t     5                              lished it as a lead review article.
ence Citation lndes ’(SCl l and the Social Sciences          Reaction was immediate. Some people
Citation lndex~ (SSC!x) indicate that this paper has     asked, “You mean I have been analyzing my
been cited in over 170 publications, making it the
                                                         data wrong all these years?” Similarly, the
most-cited paper published in this lournal since
                                                         editor of another leading journal in develop-
1973.1                                                   mental psychology wrote to ask whether he
                                                         should make an editorial policy that all re-
                                                         peated-measures designs be analyzed by the
              Robert B. McCall                          alternative methods recommended in our
              Boys Town Center                           paper.
        Father Flanagan’s Boys’ Home                        The popularity of the paper is not because
                                                         it made a new contribution to knowl-
            Boys Town, NE 68010
                                                        edge—nothing stated in it was statistically
                                   April 15, 1985       new. Rather, it filled a knowledge gap be-
                                                        tween statisticians and researchers, and it
    Every behavioral researcher needs a                 communicated the problem and the solution
friend in    the applied statistics business.           to researchers in a nontechnical, here’s-why-
Mine was Mark Appelbaum. Mark taught                    and-how-you-do-it style. It was a timely con-
me that if the pair-wise correlations among             tinuing-education article. I also suspect that
repeated measures were not homogeneous                  its influence led to the invitation to Mark
(i.e., the assumption of compound symmetry              and me to write a comprehensive chapter on
was violated), traditional methods of analyz-           design and analysis for the recent Handbook
ing repeated-measures analysis-of-variance              of Child Psychology.
designs were biased toward rejecting the                    I hope this case study demonstrates to
null hypothesis. While all the statisticians I          editors of content journals that not every-
knew seemed to recognize this problem,                  thing they publish needs to be an original
none of my colleagues in developmental re-              empirical or theoretical article. Fields can
search was aware of this bias.
                  t                                     be moved forward by continuing-education
    Why this gap No one had told the re-                papers that are styled to be useful and Un-
searchers there was a problem. Most of my               derstandable to their readership. Similarly,
research colleagues took statistics before              journals having primarily an applied audi-
this issue was widely taught.                           ence might occasionally encourage articles
   Well, then, someone ought to tell them.              by researchers that review knowledge in a
But the “someone” should be one who could               scientific or scholarly domain that might be
communicate with researchers, and the                   useful to practitioners. Also, editors in one
“telling” should be in an applied rather than           discipline might encourage review articles
in a theoretical style. So I asked Mark to pro-         from scholars in other disciplines on topics
vide me with the statistical information on             that cut across disciplinary lines. Let us ad-
the nature of the problem and the advan-                mit that we cannot keep pace with develop-
tages and disadvantages, both practical and             ments in all disciplines or levels of knowl-
statistical, of various alternative strategies.         edge, and that most of us simply do not read
Then I wrote an article that attempted to               in disciplines or areas other than our
state this material in relatively nontechnical          specialties. Therefore, we need articles that
terms and to illustrate, complete with nu-              update our knowledge published in journals
merical examples, the computation and the               that we are likely to read.

t. App.lbau~M I & McCaU R B. Design and analysis in dexelopmental psychology. Kessen W, ed.) Handbook of
       child psychology. Vols,ne 1. History, theory, and ,,se,hods. New York: Wiley. 1983. p. 415.76.

22                                                   S&BS           ~ 1985 by SI® CURRENT CONTENTS®