Search

An analysis of covariance (ANCOVA) is done to correct for chance differences that occur when subjects are assigned randomly to treatment groups. When properly used, this correction results in adjustment of the group means for pre-existing differences caused by sampling error and reduction of the size of the error variance of the analysis. The adjustment of the means is done to reduce bias that may be caused by the differences. This hoped-for increase in power is a major advantage of ANCOVA. However, the inappropriate use of ANCOVA appears to be the rule rather than the exception. This paper explains the homogeneity of regression assumption and why it is so important to evaluate this assumption before conducting an ANCOVA. Small heuristic data sets (3 groups of 12 entries each for intelligence quotient and achievement) are used to make the discussion concrete. Three tables present the data sets, and five figures illustrate their application. A seven-item list of references is included. (SLD)

A study determined the effectiveness of computer-aided drafting (CAD) and standard drafting instruction on the ability of students with differing visual-haptic perceptual styles to solve orthographic projection problems. A 2x2x2 factorial design was chosen to compare two levels of perceptual style with two levels of treatment and two levels of prior drafting experience. Activities were administration of the Successive Perception I test to determine visual-haptic style; administration of a questionnaire regarding prior drafting experience; provision of CAD instruction to the experimental group and traditional instruction to the control group; administration of the direct measure of orthographic projection ability; and collection of plate grades. The sample consisted of 188 beginning drafting students at the University of Missouri-Columbia. Three-way univariate analyses of covariance were used to test the hypotheses. The lack of significant differences between student scores on plate grades or the unit exam grades indicated that either instructional method could be used with equal success, visual and nonvisual students benefited equally from drafting instruction, and prior drafting experience did not ensure better skills in solving orthographic projection problems. Findings implied that visual-haptic perceptual style may not affect ability to solve orthographic projection problems. (15 references) (YLB)

A study investigated: (1) whether children at risk for failure to learn to read and write increase their phonemic awareness while engaged in holistic language arts instruction; (2) if so, which language experiences and instructional practices are associated with the gains; and (3) the nature of the interaction between phonemic awareness development and reading and writing achievement among those children while in grade one. Four classroom instructional programs were selected on the basis of recommendations of supervisory personnel, teacher-given descriptions, and preliminary observations. Five children from each classroom were selected as focal children. Each classroom was observed for 10 full days and each focal child was assessed in October and April on 10 measures. Teachers and children were interviewed and parents completed questionnaires indicating their child's participation in a list of preschool language learning activities and experiences with print outside of school. Results indicated that children involved in holistic language arts programs improve their performance on tasks of phonemic awareness. The instructional practices that appear to contribute to this change are writing using invented spelling and reading connected text both chorally and independently. Results also indicated that phonemic awareness is necessary for reading and writing, but is not a sufficient condition for reading and writing success. Although holistic language arts instruction contributes to phonemic awareness, it is not sufficient training for all children. The contribution of home literacy to the development of phonemic awareness is apparent. (Seven tables of data are included and 21 references are attached.) (MG)

This paper discusses the use of the LISREL computer program to compare two groups of Ontario students who studied French as a second language in either a regular or an immersion program. Longitudinal data on growth in mathematical ability, from grade 4 to grade 6, are analyzed using seven different mathematical models. The paper illustrates the use of LISREL for models with structured means, and demonstrates its usefulness for investigating the structural relationships among measurements of one true variable taken at yearly intervals. (GDC)

The paper identifies and addresses four methodological weaknesses common to most previous studies that have used LISREL confirmatory factor analysis to test for the factorial validity and invariance of a single measuring instrument. Specifically, the paper demonstrates the steps involved in: (1) conducting sensitivity analyses to determine a statistically best-fitting, yet substantively most meaningful baseline model; (2) testing for partial measurement invariance; (3) testing for the invariance of factor variances and covariances, given partial measurement invariance; and (4) testing for the invariance of test item and subscale reliabilities. These procedures are illustrated with item response data from the Perceived Competence Scale for Children from 129 normal and 132 gifted students in grade 5 and 113 normal and 117 gifted students in grade 8 from two public school systems in Ottawa (Ontario). Seven tables present study data. (Author/SLD)

A study explored the effects of individualized, integrated language arts instruction on the attitudes of poor readers. A total of 63 elementary students participated in the study, 33 in the treatment group and 30 in the comparison group. Twice a week for a period of 12 weeks, the treatment group received reading instruction which incorporated listening, speaking, reading, and writing activities based on their interests and trade books; the comparison group received basal reading instruction. The Student Attitude Inventory was used as the pre- and post-measure. Results of groups-by-grade-level and groups-by-gender analyses of covariance using the attitude pre-measure as the covariate indicated a significant groups-by-grade-level interaction effect for self-perceptions as learners, a significant main effect for gender on the reading subarea, and a significant main effect for gender on total attitude scores. Treatment group students in the upper grades had higher self-perceptions as learners than did students in the lower grades, while comparison group students' self-perceptions were not affected. There were no significant effects on attitudes toward the subareas of listening, speaking, and writing. Findings suggest support for individualized instruction which includes the use of trade books related to students' interests in an integrated language arts approach. (Seventy references are attached.) (Author/RS)

The data analysis problem posed by a repeated measures design that includes a single observation on a covariate for each subject is considered. The current paper discusses how best to capture a possible dependence of the effect of the within-subject factor on the level of the covariate. Procedures originally explicated by Rogosa (1980) for dealing with heterogeneity of regression in the between-subjects case are extended to this repeated measures situation. We conclude that such pick-a-point and simultaneous inferential procedures not only provide more powerful overall tests of the within-subject effects but also permit a thorough analysis of the attribute-by-treatment interaction implied by a significant regression of the within effect on the covariate. (Author)

This study assesses the effectiveness of different types of Jewish schools in producing adult religious involvement, using a sample of 1,009 individuals from the Chicago area and an analysis of covariance design. Based on his analysis, the author concludes that the type of supplementary Jewish education received by over 80 percent of those Jews who have received any form of Jewish education has not had any lasting impact on their adult religious involvement. Statistical analysis showed that the relationship between Jewish schooling and adult religious involvement is not wholly linear. Schooling exhibits threshold, plateau, and ceiling effects; it begins to have an impact at 2,000 hours, reaches a plateau between 3,000 and 4,000 hours, and obtains a maximum effect at 10,000 hours when reinforced by other agents of religious socialization. (Author/JG)

Federal assistance for special educational programs makes necessary the regular study of evaluations of thousands of innovations in compensatory education, bilingual education, and reading programs. The results are reported to the President and to Congress. However, investigating organizations find only a few programs with adequate evidence and thousands with faulty evaluation designs. Some of the most common faults are discussed, with examples. There are other factors which lower hopes. If greater numbers with real evidence could be found, knowledge would increase even without an increase in the number of exemplary programs. (Author)

The problem of blocking on a status variable was investigated. The one-way fixed-effects analysis of variance, analysis of covariance, and generalized randomized block designs each treat the blocking problem in a different way. In order to compare these designs, it is necessary to restrict attention to experimental situations in which observations are collected within a fixed-effects generalized randomized block framework. An analytical technique was developed to aid a researcher in choosing one of these three designs. The technique is based on the criterion of maintaining power against particular treatment main effect non-nullities, at certain specified levels. (Author)

A Monte Carlo technique was used to investigate the small sample goodness of fit and statistical power of several nonparametric tests and their parametric analogues when applied to data which violate parametric assumptions. The motivation was to facilitate choice among three designs, simple random assignment with and without a concomitant variable and randomized blocks, and between nonparametric or parametric tests. The criteria for choice were power and robustness. The parameters of the Monte Carlo investigation were strength of relationship between the concomitant and dependent variables, number of levels of the independent variable, sample size, and location parameter. (Author)

The study investigated the visual discrimination abilities of children who varied in their ability to recognize words. Measures of word recognition and intelligence were obtained in 87 first-, second-, and third-grade subjects. All subjects then performed a visual discrimination task which uses artificial graphemes as stimuli. The task required subjects to match a standard grapheme with an identical form. Errors on this task were classified into six categories. A two-way multivariate analysis of covariance (grade level X word recognition skill) was performed. In the analysis intelligence test scores were covaried. The main effect for grade was significant (p. is less than .0004), while neither word recognition skill nor the grade X word recognition skill interaction approached significance. Results provide information concerning the possible need for a re-evaluation of visual training practices in schools. (Author)

The psychophysical technique of magnitude estimation was used to develop a ratio scale of subjective estimations of adult learning in various adult education activities. A rank order of 26 learning activities and the magnitude estimations in "units of learning" that are expected to occur in each activity were obtained from 146 adult education graduate and undergraduate students, 165 participants in a school board adult education evening program and 191 participants in credit courses offered by a community college. Arithmetic and geometric means were computed and plotted for each of the 26 learning activities. Analysis of the results indicates that the magnitudes do constitute a ratio scale. Analysis of covariance revealed age and years of schooling completed to be significant sources of variance on several scale items and sex was found to have a significant effect on one item. Test re-test reliability coefficients with a one month interval between tests, and a cross-modal matching technique with a one week interval between tests indicated the reliability of the instrument. The methodology of the study and the implications of the results for adult education research are discussed. A 14-item bibliography and a copy of the instrument are included. (Author/MS)

The multivariate general linear hypothesis (MGLH) has received relatively little utilization in educational research and evaluation. This is surprising in view of the fact that recent publications have made the MGLH tractable by practitioners. This paper seeks to stimulate interest in the MGLH by reviewing recent applications, emphasizing the advantages of the MGLH as a general data analytic tool, and making suggestions for conducting research within an MGLH framework. (Author)

This paper discusses four criteria for selecting covariates when utilizing the analysis of covariance in quasiexperimental research: completeness, causation, independence, and fallibility. Completeness is the extent to which the known or suspected sources of bias are accounted for by covariates used in an experiment. Causation is the requirement that covariates used in nonrandom research designs are causally related to the dependent variable of a study. Independence is the requirement that there is no treatment/covariate interaction. Fallibility refers to the reliability of the covariate measure. Ways of determining to what degree these criteria are met are suggested. (Author)

LISREL-type structural equation modeling is a powerful statistical technique that seems appropriate for social science variables which are complex and difficult to measure. The literature on the specification, estimation, and testing of such models is voluminous. The greatest proportion of this literature, however, focuses on the technical aspects of LISREL, and many are far advanced in statistical sophistication so as to be of little or no help to the novice modeler. This paper presents a "philosophical" approach to structural equation modeling and examines some of the basic concepts and assumptions underlying the formulation of these models. The rationales for the structural equation model and the measurement model are outlined, and the foundations of specification and interpretation clarified. As part of the SPSS-X package, LISREL will be readily available to many applied researchers. With this heightened availability of a complicated statistical procedure will undoubtedly come many misuses and abuses. Social scientists should become familiar with the theoretical premises of the method. (Author/JAZ)

A dichotomous dependent variable is used to determine a combination of variables that will predict group membership. Dichotomous variables are frequently encountered in multiple regression analysis. However, several textbooks question the appropriateness of using multiple regression analysis when analyzing dichotomous dependent variables. The critics state that in addition to the predictions made by the regression equation with a dichotomous dependent variable, the statistical tests are also invalid. This paper assesses the meaning of the charges against multiple regression and deals with some logical extensions of them. Data from the A3 data set in Gunst and Mason is used to examine the criticisms by comparing results of cases viewed as appropriate by critics with cases viewed as inappropriate for one predictor and two predictor examples. As the tests of significance are identical whether the dichotomous variable is independent or dependent, critics must deal with all significance testing with t-tests, analysis of variance, analysis of covariance, discriminant analysis, and any use of dummy variables in multiple regression if they are to be taken seriously. (BS)

Several multivariate statistical methodologies have been proposed to ensure objective and quantitative evaluation of the multitrait-multimethod matrix. The paper examines the performance of confirmatory factor analysis and covariance component models. It is shown, both empirically and formally, that confirmatory factor analysis is not a reliable method for simultaneous estimation of trait and method factors. The poor performance is due to an inherent rotational indeterminacy common to all factor analytic models of trait and method effects. Covariance component analysis, on the other hand, shows a more parsimonious parameterization of general, trait, and method variation in the multitrait-multimethod matrix and is therefore typically unaffected by rotational indeterminacies. The performance with 23 empirical multitrait-multimethod correlation matrices was also found satisfactory. (Author)

This study investigated whether first grade students could be taught metacognitive strategies to solve analogies. Subjects, 34 first grade students randomly assigned to two groups with adjustments made to make the groups equal in gender, ability, and urban or rural geographic location, were pretested for their ability to solve analogies. The experimental group spent 30 minutes per day for 10 days in direct instruction in analogy-solving strategies. An integrated curriculum (webbings) was used throughout the year. Students were taught to verbalize the general rule that "before we learn something new, we must find out what we already know." Various brainstorming techniques were used to accomplish this. The first day of the webbing students were asked, "What do we know about eggs?" or "What do we know about mice?" and responses were recorded on chart paper. These were then categorized according to levels of Bloom's taxonomy and saved to be compared with responses at the end of the webbing. Posttests were administered 2 weeks after the pretest. Results indicated that students in the experimental group were significantly more adept at finding the special relationships involved in the analogy as well as in utilizing a common language to help in its solution. Results also indicated an increased use of metacognitive strategies by students. (Two tables of data are included; journal samples, egg webbing, mice webbing, and 41 references are attached.) (RS)

A study investigated differences in reading rate and comprehension scores of 8th grade students on a cloze reading exercise involving the reading of a cloze passage from 10 different modes of text presentation (8 different computer screen formats and 2 different paper formats). Subjects, 300 students in computer literacy classes from 7 schools in a large suburban school system, randomly placed into 1 of 10 treatment groups, read a passage with 50 missing words and wrote the missing words on an answer sheet. The time taken to complete the exercise was considered to be the subject's reading rate. The passages were scored to determine the subject's comprehension measure. Results indicated no significant differences in reading rate or comprehension for any of the 10 groups; thus the ability of individuals to comprehend text does not appear to be impaired by the decreased legibility of computer screens. (Four tables of data are included; 24 references are attached.) (RS)

This study examines the relative effects of information on problem solving. An introductory section discusses theoretical approaches to problem solving. The major portion of the paper describes the author's research. In her study, subjects were asked to read a court transcript from a criminal case, to give their verdict in the case, and to recommend an appropriate sentence if necessary. Finally, subjects were asked which aspects of the transcript they considered most important in making their decisions. Each group of subjects received the same basic account of an assault and battery incident, with variables being defendent's race, presence of defense witness, and sex of plaintiff. Results indicated that subjects were more likely to favor a feamle plaintiff's case and less likely to favor the plaintiff when a witness testified. However, if a male was charged with assautling a female, the presence of a witness had no effect. White assailants were more likely to be found guilty than blacks when the plaintiff was a male. Most subjects reached what was considered the correct verdict. However, subjects who did not list material facts as being most important were three times more likely to err in judgement. The experimental treatment demonstrates that the characteristics of information sources interfere with problem solving. Source credibility is particularly important. Subjects often find it difficult to disregard their feelings about individuals when making decisions. (BC)

This study evaluated the implementation of and the outcomes from a local child parent center compensatory education program. It intended to contribute to a better understanding of one phase of compensatory education; and, in the process, it proposed to develop a relatively simple and practical evaluation strategy which would verify the implementation of a program and relate the implementation measures to the outcome measures. The evaluation strategy included practical methods for modeling the program, gathering data, and analyzing data. Causal modeling techniques were used to relate implementation measures to outcome measures. The strategy was than analyzed regarding its usefulness as an evaluation design which would measure program implementation and provide explanatory power. (Author/PN)

Joreskog has developed a very general and powerful model and computer program for analyzing relationships among variables, called LISREL. Structural analysis of covariance matrices by Joreskog's LISREL method is proposed as an alternative to regression methodology in the analysis of aptitude-treatment interaction (ATI) data. LISREL resolves some of the basic problems in regression analysis--such as biased estimates caused by unreliability of measures, and interpretative difficulties when there are multiple aptitudes and outcomes. LISREL models range in complexity: from regression models to models involving complex structural relations between latent aptitude and latent outcome variables. (Consideration is also given to the formulation of LISREL models when there are observational units at different levels of aggregation). (Author/JKS)

Four types of control groups are commonly used in cognitive manipulation studies: (1) no-treatment; (2) practice with own methods; (3) practice and training with competing treatments; and (4) practice and training with irrelevant treatments. There are problems associated with the use of each group as a baseline for identifying the "true" treatment effect. A general, two-step control procedure was advocated in which the researcher first identifies the most appropriate control group for a given context and then supplements this experimental control with statistical control. In reference to the latter controls, procedures were presented for developing a task-specific index as a covariable to be employed with a suitable measure of verbal aptitude in a multiple analysis of covariance. The procedure was demonstrated on data from a quasi-experimental design in which it was shown that three different conclusions were possible, depending upon the control procedures employed. The converging operations obtained from the multivariate approach to the data analyses supported the results from the two-step control procedure. (Author/RD)

The conceptualization of analysis of covariance (ANCOVA), as an analysis of variance (ANOVA) on the residual scores that are obtained when the dependent variable is regressed on the covariate, is mathematically incorrect. If residuals are obtained from the pooled within-groups regression coefficient, ANOVA on the residuals results in an inflated alpha-level. If the regression coefficient for the total sample combined into one group is used, ANOVA on the residuals yields an inappropriately conservative test. In either case, ANOVA of residuals fails to provide a correct test, because the significance test in ANCOVA requires consideration of both the pooled within-groups regression coefficient and the regression coefficient for the total sample combined into one group. It is recommended that the significance test of treatment effects in ANCOVA be conceptualized as a comparison of models whose parameters are estimated by the principle of least squares. (Author/RL)

Analysis strategies are discussed for the nonequivalent control group design when three models of continuous natural growth are known. For Model I type natural growth it was shown that the fan spread hypothesis always holds, and Analysis of Covariance (ANCOVA), Analysis of Variance (ANOVA) of Residualized Gain Scores, and ANOVA of Standardized Change Scores all are potentially correct analysis strategies. For Model II and Model III type natural growth it was shown that the fan spread hypothesis holds and the ANCOVA, ANOVA of Residualized Gain Scores, and ANOVA of Standardized Change Scores are potentially correct analysis strategies only when Model II and Model III type natural growth reduce to Model I type natural growth. Further, it was shown that given any natural growth situation, there is a value of K for which ANOVA of Index of Response is a potentially correct analysis strategy. In order that an index of response works, the exact form of natural growth must be known. Most growth models, and in particular most linear growth models, do not conform to the fan spread hypothesis nor are the usual analysis strategies correct for these models. (Author/RL)

Since measurement errors exist in panel surveys, LISREL (a procedure for controlling measurement error by the analysis of covariance structures) was used in this investigation to determine the stability of the social-psychological concept of locus of control expectancy change with the aquisition of post secondary education. The National Longitudinal Study of the High School Class of 1972 made a survey of 22,652 seniors in 1,318 public and private high schools. A subset of the data on 8,650 white males was used for this study. The measures of the locus of control, based on Rotters' Scale, were used twice, once in the students' senior-year and again four years later. The variables of ability and socio-economic status were also incorporated into the study. Apparently the acquisition of further education beyond high school does not substantially increase one's internal control expectancies. Parallel results are to be found in a study on self-esteem made by Bachman, O'Malley and Johnston (1978). Data are presented in tabular form throughout the study. (Author/AEF)

Comparisons between whites and blacks in models of educational achievement were found to be suspect when based solely on least-square estimates, since the estimates are biased by measurement error varying by race. In this study, white high school seniors were shown to report their parents' status characteristics more reliably than black high school seniors. Data were drawn from the National Longitudinal Study of the High School Class of 1972 as the seniors moved into early adult years. Variables included father's and mother's educational attainment in the base year and first followup year, and father's occupation measured by Duncan's socioeconomic index adjusted to the 1970 census occupation classification. Simultaneous covariance structure analysis methods developed by Joreskog were used in estimating measurement parameters for both groups. Explicit statistical tests of group differences in measurement error patterns are described. (Author/CM)

Religiosity is a personality attribute that affects social attitudes and behavior. To examine whether religiosity also affects television use, a study administered questionnaires to 346 Christian conservatives, moderates, and liberals in six different churches in northeast Ohio during November and December 1987. The questionnaire measured religiosity, watching programs containing sex and violence, viewing motivations, and television attitudes. Findings showed that compared to nonconservatives, religious conservatives were less motivated to watch television because of the sexual appeal of characters and programs, watched fewer programs containing sexual content, found television to be less realistic, and felt television was less important to them. Findings also showed no religious group differences for reactionary motivation or watching violent programs. (Three tables of data are included, and 35 references are attached.) (Author/MS)

Hypothetical data sets are used to demonstrate how canonical correlation methods subsume other commonly utilized parametric methods. Analysis of variance, analysis of covariance, multiple analysis of variance, and multiple analysis of covariance are heavily used by educational researchers. It is concluded that researchers would do well to consider using general linear model (GLM) techniques as opposed to analysis of variance and its analogs. GLM and analysis of variance are equivalent when the ways in an analysis of variance design each have exactly two levels. In other cases, the GLM analytic approach yields more specific information regarding effects, and greater power against Type II error, because the comparisons are planned a priori rather than post hoc. Multivariate canonical correlation methods are superior to univariate techniques in providing important information on the interrelationships among variables. Furthermore, multiple univariate procedures tend to inflate the probability of a Type I error. (GDC)

The instructional components of mediation, a teaching technique used in dynamic assessment, were evaluated with 100 four-year-old children attending daycare or preschools in London, Ontario. The effects of familiarization with task materials, task-specific rule teaching, and elaborated feedback were assessed using a pretest-posttest design. Children trained on the Stencil Design Test using various combinations of the mediation components were compared to each other and to a non-instructed control group. Analyses of covariance on posttest measures with type of instruction as the independent variable and pretest scores as the covariate revealed that feedback was the most effective component of the mediation procedure. Performance on the Stencil Design Test and near transfer performance on the Animal Stencil Test was significantly better than performance of non-instructed control children whenever feedback was included in instruction on the Stencil Design Test. There was no transfer to the Animal House Coding Task of the Wechsler Preschool and Primary Scales of Intelligence, or to the Conceptual Grouping subtest of the McCarthy Scales of Children's Abilities. Results were discussed in terms of the importance of understanding the nature of the instruction provided in dynamic assessment. (Author/JAZ)

Parametric analysis of covariance was compared to analysis of covariance with data transformed using ranks. Using a computer simulation approach the two strategies were compared in terms of the proportion of Type I errors made and statistical power when the conditional distribution of errors were: (1) normal and homoscedastic, (2) normal and heteroscedastic, (3) non-normal and homoscedastic, and (4) non-normal and heteroscedastic. The results indicated that parametric ANCOVA was robust to violations of either normality or homoscedasticity. However when both assumptions were violated the observed alpha levels underestimated the nominal alpha level when sample sizes were small and alpha=.05. Rank ANCOVA led to a slightly liberal test of the hypothesis when the covariate was non-normal and the errors were heteroscedastic. Practical significant power differences favoring the rank ANCOVA procedure were observed with moderate sample sizes and skewed conditional error distributions. (Author)

A study was conducted to assess the possible effects of headings on the search and recall performance of high- and low-knowledge subjects, 153 college students in an introductory psychology course. Seventy-five of the subjects read a 1,760-word adaptation of the passage on organizational processes in memory--the low-knowledge passage. The other 78 subjects read a 1,760-word passage on human sexuality--the high-knowledge passage. Each passage was divided into eight sections by the inclusion of embedded headings, each consisting of a word or short phrase. For each section of each passage, one short-answer question quizzing high-level, main-idea information and one short-answer question quizzing low-level, detail information were generated. Approximately half of the subjects in each passage condition read the passage in booklet format with the headings present, and the remainder of the subjects read the passage with headings absent. Subjects answered questions on one test by consulting the text, and then in a second test based only on memory. Results indicated that headings can positively affect the performance of low-knowledge subjects as they search text for answers to questions. The score in the headings-present condition was significantly higher than the score in the headings-absent condition for the low-knowledge passage, but the score in the headings-absent condition was significantly higher than the score in the headings-present condition for the high-knowledge passage. (Three tables of data are included.) (MG)

A study investigated the interrelationships among warrants, argument dimensions, and perceived argument effectiveness. Warrants were operationalized as analogies, examples, and causal warrants, while argument dimensions were defined as degree of logicality, emotionality, and ethicality. Based upon prior research, weak and strong arguments were developed using each of the three types of warrants for each of three attitudinally-different topics. Further, the influence of critical thinking instruction was examined. A total of 393 students from two major western universities were surveyed. Analyses revealed topic, warrant, and instruction effects. The research suggests implications for bases of argument perceptions and pedagogical approaches toward critical thinking instruction. One appendix containing arguments used in the study, and 35 references are attached. (Author/SG)

This paper presents two extensions of the full-rank multivariate linear model that are particularly useful in multivariate analysis of covariance (MANCOVA) and repeated measurements designs. After a review of the basic full-rank model, an extension is described which allows restrictions of a more general nature. This model is useful in the analysis of repeated measurements designs when the researcher assumes certain interactions do not exist and wishes to pool sources of variation, or in the analysis of repeated measurements designs with covariates. A further extension of the basic model allows an even more general class of restrictions. This extended model allows testing hypothesis on the between groups portion of a repeated measures design with covariates, subject to a restriction of equality of slopes over measures. (Author/BW)

Analysis of covariance is examined for its usefulness in attribute-by-treatment interaction (ATI) research, where subjects are randomly assigned to treatments but not to levels of the attribute factor. Contrary to what might be expected, the use of a covariate does not confound the interpretation of the interaction between attribute and treatment, because the expected value of the interaction effect is independent of the particular covariate chosen. Independence occurs regardless of whether the attribute variable is discrete or continuous, and regardless of whether the covariate is measured with perfect or less than perfect reliability. The primary implication of these results is that analysis of covariance provides a much needed means for researchers to achieve powerful designs in assessing attribute-by-treatment interactions. (Author/GDC)

The uses of data on siblings for educational and psychological research purposes are explained. Examples are based on empirical data. Five types of research problems, using sibling data, are explored: (1) statistical control of family background variance in psychological or educational studies which use analysis of covariance (ANCOVA) or multiple regression analysis; (2) testing the adequacy of age-standardized scores; (3) testing the interval scale property of age standardized scores, such as intelligence or achievement quotients; (4) analysis of the correlations among variables into components which can be attributed to correlation among variables between families or to correlation among variables within families; and (5) assessment of the cultural bias in test scores or of the common factors measured by tests. Between-family-variance is defined as variance due to genetic and environmental influences common to all siblings in a family, but which differ between families. Variance within families is described as genetic and environmental influences which cause siblings to differ from one another. Some types of variables can have both between-family and within-family differences. (MH)

Latent Variables Path (LVP) models have been found to be a helpful device for the formulation and validation of complex theories in educational sciences. The estimation of parameters for LVP models can use the Maximum Likelihood (ML) or the Partial Least Squares (PLS) method. This paper focusses on the PLS method introduced by H. Wold, and compares it to the LISREL method introduced by K.G. Joereskog. The differences in the results are explained by the differences in the statistical assumptions, and some clues to the choice of ML or PLS are given. Several select applications show the range where only PLS estimation is applicable. (Author)

The central purpose of the present study was to evaluate several statistical techniques (analysis of variance (ANOVA) on raw gains, standardized ANOVA, standard analysis of covariance (ANCOVA), and z-score ANOVA) under various quasi-experimental conditions (three levels of reliability, three levels of sample size, three levels of gain, and the equality of pretest means). One thousand sets of data were generated for each combination of statistical technique/conditions. Based on the results of the study, none of the statistical techniques were found to appropriately model reality. There was no statistically significant difference between the results produced by all four statistical techniques when there was no gain in either group and the pretest means of both groups were equal. ANOVA on raw gains and z-score ANOVA produced similar results when there was no gain in either group and the pretest means were different for both groups. The idea of developing a multiple analysis scheme was not supported for any of the statistical techniques used in this study. Finally, the study did not support the use of the statistical techniques investigated under the quasi-conditions constructed for this study. (BW)

The study examined associated treatment effects on State Board Examination (SBE) scores of two predominantly black nursing classes who had been exposed to different intervention methods: (1) the teaching of effective test-taking skills, and (2) a combination of that instruction with learning team methods. The SBE performance of the two groups who received intervention was compared with each other and with a control group who received no intervention instruction. Both groups receiving intervention attained significantly higher scores than the control group. However, the group exposed to both effective test taking instruction and learning team methods had a mean SBE score significantly higher than the other two groups. Results indicated that cooperative learning, coupled with effective test taking techniques, assisted students in acquiring, processing, and applying information on comprehensive standardized tests. (Author/DWH)

Nuclear data uncertainties are provided as covariance matrices in standard nuclear data libraries and propagating them trough neutronics simulations helps quantify the associated uncertainties on the final result. However, processing these matrices often poses challenges. Currently, the IRSN nuclear data processing code GAIA processes cross sections via several modules like DOP (Reconstruction and Doppler), TOP (URR), and SAB (TSL), but lacks the capability to process covariances. This paper introduces a new module named COP (COvariance Processing). The COP module aims to process covariance matrices comprehensively, including cross section (File 33), angular distribution (File 34), and resonance parameters (File 32). The preliminary results obtained using the COP module of GAIA in comparison with the ERRORR module of NJOY and PUFF module of AMPX are presented. [ABSTRACT FROM AUTHOR]

The characterization of the uncertainty in radiation damage metrics presents many challenges. This paper examines the current approaches to characterizing radiation damage metrics such as hydrogen and helium gas production, material heating, trapped charge in microelectronics, and lattice displacement damage. Critical uncertainty aspects go beyond just the material cross sections and involve the consideration of energy-dependent cross reaction correlations, the recoil ion energy spectrum, and models used for the partitioning of the recoil ion energy into various forms of energy deposition. This paper starts with a review of terminology and then examines the current approaches in the characterization of uncertainty in radiation damage metrics for several applications. The major deficiencies in the uncertainty of the damage metric characterization are also identified. [ABSTRACT FROM AUTHOR]

The mathematical derivation of the statistics used for inference in some linear models assumes that the values of the independent variables are measured without error. This assumption is often disregarded when these models are utilized in research. This study is an investigation of the consequences of the violation of this assumption for one family of linear models on the magnitude of these statistics and the frequency of Type I error in these models. The results of this study indicate that although the frequency of Type I error in these models is relatively unaffected, the values of the statistics are seriously affected by the concomitant variable being other than error-free. (Author)

A linear classification rule (used with equal covariance matrices) was contrasted with a quadratic rule (used with unequal covariance matrices) for accuracy of internal and external classification. The comparisons were made for seven situations which resulted from combining three data conditions (equal and unequal covariance matrices, minimal and nonminimal group centroid separation, and two and three criterion groups) for different sets of data. For the internal analysis the quadratic rule was superior in all seven situations. For the external analysis the linear rule was superior in nearly all of the situations. (Author)

Title I evaluations using the RMC Model C design depend for their interpretation on the assumption that the regression of posttest on pretest is linear across the cut score level when there is no treatment; but there are many instances where nonlinearities may occur. If one applies the analysis of covariance, or model C analysis, large errors may result. Various methods are suggested for model C users to deal with non-linearities. One method is Mosteller and Tukey's re-expression of posttest scores. Another method depends upon differential weighting of scores in different parts of the pretest distribution. A third method uses quadratic regression lines and extrapolation to obtain a "no-treatment expectation." Still another method is to assume parallel regression lines for the treatment and control groups. A number of different logistic curves may be tried if both floor and ceiling effects are present. Graphing the scatterplot of pretest versus posttest scores is also recommended. (Author/CTM)

The problems inherent in analyzing data in which subjects are "nested" within hierarchical units (such as classrooms or schools), and thus are not independent of one another, are addressed through the use of Joreskog's LISREL model for analyzing covariance matrices. A solution to the problem is proposed and illustrated using data from the National Longitudinal Study of the High School Class of 1972. (JKS)

Linear classification functions (LCFs) arise in a predictive discriminant analysis for the purpose of classifying experimental units into criterion groups. The relative contribution of the response variables to classification accuracy may be based on LCF-variable correlations for each group. It is proved that, if the raw response measures are "standardized," the ordering of the variables can be accomplished by ordering the means of the standardized measures within each group. A computational procedure is presented for obtaining these means as well as for obtaining the LCF-variable within-groups correlations. These correlations may be used for two purposes: (a) to characterize each group with respect to variable scores, and (b) to determine the relative contribution of the variables to separate group classification accuracy. It is proved that the LCF-variable correlations for a group are proportional to standardized variable means for that group. The proof enables a researcher to quite readily accomplish the two interpretation purposes. Two data sets are used to illustrate the interpretation of LCFs. (Author/RL)

The assertion that Model C is feasible for Elementary Secondary Education Act Title I evaluation, why it is feasible, and reasons why it is so seldom used are explained. Two assumptions must be made to use the special regression model. First, a strict cut-off must be used on the pretest to assign students to Title I and comparison groups. Second, the functional relationship between pretest and posttest must be assumed in order to fit a regression line for use in determining the performance level of students without Title I. A single composite measure that combines test scores, teacher ratings, and past academic performance can be used to select Title I students, and as the pretest. Model C needs only a district-wide assessment of all students and access to a computer to match the data. Six reasons given for not choosing Model C include: (1) Model C requires more work than does Model A; (2) local programmers lack expertise in manipulating scientific data; (3) large districts lack experience in matching pretest and posttest scores; (4) data analysis is necessary; (5) selection process must be open; and (6) normal curve equivalent gains are lower than when using Model A. (RL)

The multivariate analysis of variance approach to the analysis of data from repeated measures designs that include covariates is reviewed and contrasted with univariate methods. Assumptions necessary for meaningful interpretation of the results of the multivariate analysis are discussed and contrasted to those of the univariate approaches. An extension of the full-rank multivariate linear model approach and its use in testing various hypotheses and assumptions is also presented. Questions of model fitting versus hypothesis testing in exploratory research are also included. Examples illustrating the various methods are provided. (Author/JKS)

The presence of artifactual bias in analysis of covariance (ANCOVA) and in matching nonequivalent control group (NECG) designs was empirically investigated. The data set was obtained from a study of the effects of a television program on children from three day care centers in Mexico in which the subjects had been randomly selected within centers. However, differences in the scores of the children between centers were sufficient to provide a more able and a less able group for subsamples in an artificial quasiexperimental study of the same variables. With these subsamples, the same treatment effects were estimated using each NECG design. Unbiased and NECG estimators were then compared for each design, using the one-tailed binomial test to seek the presence of a systematic artifactual bias. A significant bias underrating the group initially lower on the pretest measures was found in both designs. The use of modern multivariate analysis designs was recommended as a better alternative, but experimental studies of these designs were also recommended. The advantage of randomization and true experimental design was pointed out. (Author/CTM)