Search

The union and disjunction models for combining individual test marks to yield final grade distributions were outlined. It was expected that these models would differentially favor students according to their characteristic methods of learning and studying. The models were examined empirically in two educational psychology classes; five performance assessments were available, as well as eleven dimensions of study behavior and academic attitudes. While there was a substantial correlation between the distributions derived from the two models, they were found significantly to favor different student characteristics. The union model, relative to the disjunction model, favored students who were dependent, who interrelated different aspects of their course work, who scheduled their work and who rote-learned material. These differences were interpreted in terms of the educational assumptions underlying the two models. Some implications of the study for grading practice were suggested. When an instructor chooses to use more than one instrument to assess student achievement in a class, he is faced with a problem. How does he derive a final distribution of marks that he can use for grading purposes? The progression from initial test distributions to the final grading distribution involves three steps. First, the original distributions have to be rendered equivalent, preferably by standardization; second, a weighting system has to be decided (e.g. a final exam is "worth more" than a term paper); and third, the distributions have to be combined in some way. In this paper, we are primarily concerned with the last question. Basically two models of combination are assumed in current practice. The most common is undoubtedly a form of the linear composite model (Cronbach 1960), which we call here the union model. Here, the scores are simply added (assuming some form of standardization and weighting). The instructor using this model is saying in effect: "To be a good student you need to be good at this and this and this ...." The union model essentially follows the subject-centered tradition in education (e.g. Biggs, 1971a) in which the nature of the content defines multiple goals, and individuals have to demonstrate competence with respect to each goal. The second is the disjunction model. Here, it is assumed that prowess may be demonstrated in one of several alternative ways. It implies an interaction between methods of evaluating performance and individual differences. This model is used, for instance, where instructors offer both objective and essay evaluations and allow students to choose their mode of evaluation, or where students write both kinds of tests and the highest score is taken as representing their "real" ability with respect to the course content. The disjunction model is less common than the union model

This paper is a presentation of the statistical sampling theory of stepped-up reliability coefficients when a test has been divided into any number of equivalent parts. Maximum-likelihood estimators of the reliability are obtained and shown to be biased. Their sampling distributions are derived and form the basis of the definition of new unbiased estimators with known sampling distributions. These unbiased estimators have a smaller sampling variance than the maximum-likelihood estimators and are, because of this and some other favorable properties, recommended for general use. On the basis of the variances of the unbiased estimators the gain in accuracy in estimating reliability connected with further division of a test can be expressed explicitly. The limits of these variances and thus the limits of accuracy of estimation are derived. Finally, statistical small sample tests of the reliability coefficient are outlined. This paper also covers the sampling distribution of Cronbach's coefficient alpha.

For the clinician, the validity of a test or assessment technique resides in the range and structural clarity of the information it provides him about the individual client with whom he works. 2 devices, for example, might be equal in forecasting a particular criterion, yet differ widely in their personological implicatons. Analysis of this problem permits specification of 3 levels of evaluation: primary, secondary, and tertiary. The conceptual model defined by these levels would appear to incorporate the kind of information which the diagnostician desires, and which indeed he must have if he is to function in an insightful and fully professional manner. The purpose of this paper is to offer a point of view concerning the meaning of measurement in psychology. Attention is centered on the use or application of such measures, and what they tell the interpreter about the individual who has been tested. Significant prior discussions of the validity issue in testing and diagnosis have sought to classify tests according to the criteria employed in their construction and evaluation (cf. Cronbach & Meehl, 1955), and to specify ways in which discriminations may be sharpened (cf. Campbell & Fiske, 1959). These emphases are important, but nevertheless do not touch on all of the significant facets of meaning subsumed under the concept of validity. The intention here is to present a different perspective, one which stresses the implications of any scale or variable when it is brought to bear upon the analysis of the individual case. From this perspective, the practitioner in testing seeks variables which permit individuated descriptions of the subject who has been tested, forecasts of what he will say or do

Theoretical Background In this paper, two quite different approaches to achievement testing converge. One of these is the strong form of educational behaviorism exemplified by B. F. Skinner's work in the area of programmed instruction (Skinner, 1954, 1958, 1961). The other is the positivistic approach to mental test theory exemplified by Lee J. Cronbach's work on "Generalizability" (Cronbach et al., 1963, Rajaratnam et al., 1965).

Cronbach (1955) has proposed that accuracy of person perception scores be divided into several components as follows: Total. Elevation, Differential Elevation, Stereotype Accuracy, and Differential Accuracy. In the present paper, hypothetical items are presented to illustrate the fact that the system used in assigning scores can artifactually reduce the Differential Elevation and Stereotype Accuracy components and in extreme cases cause them to disappear completely. An empirical example is presented in which these artifacts of the scoring system could result in an erroneous conclusion that there is no generality in accuracy of person perception.

The problem of determining whether or not a particular individual has deteriorated intellectually or with respect to other skills is one which has exercised the ingenuity of clinical psychologists for a long time. Clearly the most satisfactory method from the point of view of reliability and validity is by means of the test-retest method over a given period of time, using tests for which adequate statistical information is available. However, since a direct estimate of the individual's previous level of intelligence is usually not available and since the psychologist often cannot wait six months or longer to obtain a direct retest, some method had to be devised which would overcome this difficulty. Babcock (1930) hit on the method of using the Vocabulary test as an estimate of previous level of intellectual ability and contrasting it with performance on other tests supposedly sensitive to deterioration in order to measure the amount of decline which had taken place.Some aspects of the use of vocabulary in clinical psychological testing have been reviewed by a number of writers (Cronbach, 1942; Feifel, 1949; Hunt, 1936; Jastak, 1949; Lewinski, 1948; Raven and Walshaw, 1944; Shakow, 1946), and the immense interest shown in vocabulary as a test may be inferred from the 101-page bibliography of vocabulary studies published by Dale (1949). However, previous reviews suffer from two defects in that(a)they do not attempt to cover the ground thoroughly or systematically, being usually introductory to an experiment, or not specifically concerned with the subject matter of this paper; and(b)they naturally give no information about the important research which has taken place in this field within the last few years. The present review has been undertaken for these reasons, and because of the intrinsic importance of the problem.

Summary. This paper describes ten attitude scales for use with pupils aged 9 to 11 years. These are: attitude to school; interest in school work; importance of doing well; attitude to class; ‘other’ image of class; conforming versus non-conforming pupil; relationship with teacher; anxiety about school work; social adjustment; self-image. The scales were derived empirically and each is made up of a number of statements made by children during group discussions and selected after factor analyses and scalogram analyses. For each scale the aim has been to achieve homogeneity of content and where possible a cumulative structure. The internal consistency of each scale is expressed in terms of Cronbach's alpha coefficient. The intercorrelations of the scales with each other and also with external data are given. Information for the latter came from several sources. These were (a) personality and ability ratings of the pupil by the teacher; (b) parents' ability ratings and educational aspirations for the child; (c) sociometric data; (d) achievement test scores; and (e) interest scores.

The operation of reliable response sets or stylistic consistencies has been frequently noted on personality and attitude scales with a true-false or agree-disagree format (cf. Cronbach, 1946, 1950; Fricke, 1956; Messick & Jackson, 1958). It has recently been conjectured (Jackson & Messick, 1958) that the major common factors in personality inventories of this type are interpretable primarily in terms of such stylistic consistencies rather than in terms of specific item content. The present paper attempts to annotate the influence of two response styles, the tendency to agree or acquiesce and the tendency to respond in a desirable way, using the Minnesota Multiphasic Personality Inventory (MMPI) as an example of inventories with this general response form. In particular, a high correlation will be noted between factor loadings on the largest factor, as obtained in several published factor analyses of the MMPI, and certain indices of acquiescence. Barnes (1956b), in evaluating the Berg (1955) deviation hypothesis on the MMPI, found that the tendency to answer atypically or deviantly "true" was highly correlated with

As part of the development of a test battery to determine proficiency in Black standard and nonstandard speech, a test was devised consisting of a repetition task. Fifteen sentences in Black standard and fifteen sentences in Black nonstandard English were to be repeated. The sentences were contained within two similar stories recorded on tape by a bidialectal speaker. A Black experimenter administered the test to 35 Black kindergarten children (18 male, 17 female) in the Spring of 1972. Tests were administered individually. The experimenter stopped the tape after each sentence containing a test item and asked the child to repeat the sentence. The response was scored as correct if the child repeated the test item exactly as modeled on the tape. Mean scores were 10.9 on the nonstandard and 11.3 on the standard section of the test, indicating a general balance between standard and nonstandard. The reliability of Section A (nonstandard) of the test was 0.49 (Cronbach a); for Section B (standard) it was 0.43. Subjects were also assigned a balance score (Section A minus Section B), which measured the dominance of nonstandard over standard. Scores of the same students on the Stanford Achievement Test and its subsection on letters and sounds correlated positively and significantly with the standard section of the test. Where there was an imbalance in favor of nonstandard there was a significant negative correlation with the Stanford Achievement Test and its subsection on letters and sounds. This paper reports on the experimental development of a test to measure the language proficiency of children who are speakers of Black English. The test is designed to measure the ability to speak both nonstandard and standard English. Black standard English has been defined as English that follows most of the grammatical rules of standard English but is

This paper is a critical survey of methods for analyzing the similarity between profiles. Among the measures of similarity examined are intraclass correlation, Cattell's rp, Cronbach and Gleser's distance measure, analysis of variance, Hotelling's T2, discriminant function, and Mahalanobis' D. An evaluation is made of each, some illustrative examples are given, and the relationships among several of them are presented.

The paper is intended as a contribution to the sampling theory of reliability estimation when a test has been divided into two, not necessarily parallel, parts. Under normality assumptions, a strict t -test of a point hypothesis about the coefficient alpha parameter is derived. The test is then converted to yield confidence intervals for alpha. These techniques remain applicable even when the initial distributional assumption is considerably relaxed. The methods developed here are complementary to certain large sample techniques of the same intent. Worked examples are appended by way of illustration. (Author)