Search

The article presents a detailed description of the American Sociological Society along with information on its annual meeting to be held in September 1951 in the U.S. Article II of the society's constitution states that the objective of the society is to stimulate and improve research and encourage cooperative relations among people associated with the scientific study of the society. Article III states that any person interested in the society's objective is eligible for membership. According to Article VIII, the society can solicit and receive funds and endowments. Several papers will be presented at the society's annual meeting. Some of them include a paper on "Demography and Human Ecology," another on "Sociology of Religion," and another on "Sampling and Social Structure."

This paper describes two statistics which may be used to test hypotheses about the association between two variables when observations are cross-classified in a contingency table. The two sets of null hypotheses are described and differences between the sets of alternate hypotheses covered by each statistic are outlined. Responses from populations with underlying bivariate linear and monotonic association were computer-simulated to compare the relative control over Type I error and power for each statistic. Sample sizes of 50 and 100 and contingency tables of size 3 x 3 and 4 x 4 were employed in the simulation. For the selected parameter combinations and distribution characteristics, S yielded a proportion of Type I errors which was generally closer to the nominal a than did x² Further, tests based on S rejected proportionally more false α than did x² for nearly all parameter combinations. General experimental situations are described in which tests based on S would be preferred over those based on x². An example of such a situation is included. [ABSTRACT FROM AUTHOR]

This article discusses different sociological research methodologies by presenting three solutions to the problem of controlling for a large number of variables with a small sample size. It is suggested that a small sample size often does not permit the use of standard procedures for controlling two or more variables. The first method, described here, is to choose a few of the potential controls, basing the selection on the size of correlations between control and dependent variables or among the control variables, on substantive considerations, or on previous research findings. The second method is to form an index based on all of the control classifications. This is a widely applicable and useful method, because it can be used with an extremely small number and a large number of control classifications, but each control must be weighted so that an index can be derived for each individual. The third method presented, cell-ordering and combination, involves grouping the cells into broad classes to be used for control purposes. This method reportedly is not as widely applicable as index formation, for it requires a larger number of cases.

The article presents information on generalized rules for calculating the magnitude of an effect in factorial and ANOVA designs. Many present day researchers are using the phrases approaching significance, significant, highly significant, and dramatically significant in an attempt to describe the magnitude, and therefore the inferred importance, of effects in comparative experiments. The use of a particular phrase is dependent on the relative difference between the standard alpha probability value, .05 or .01, and the obtained or looked up probability value associated with the observed t or F for an effect. To correct this situation, researchers have shown renewed interest in correlation-like measures of association, which are not directly dependent on sample size but are meaningfully indicative of the size of the magnitude of the effect. Setting each obtained mean square equal to the expected mean square results in a series of equations which can be solved simultaneously to obtain estimates of each variance component. After each variance component has been obtained, the ratio of one variance component to the sum of all variances will represent the proportion of total variance due to that one source and will thus be an estimate of the magnitude of that effect.

This paper examines the range of samples used in recent English and American surveys and proposes an alternative to traditional techniques of sampling and significance testing which, it is suggested, may result in substantial savings in time and research resources. Three distinct applications of sequential sampling tests are illustrated by data derived from the Hull Family Survey. [ABSTRACT FROM AUTHOR]

This article presents a discussion on the measurement techniques in evaluation. Educational evaluation is broadening in scope and is truly becoming an activity concerned with value. The increased comprehensiveness of evaluation efforts and a recognition of what is being evaluated has required an expansion of the number of type of measurements included in the evaluation. Observation systems, interaction analysis, matrix sampling, generalizability theory, computer-controlled testing, and mastery testing are all relatively new developments that have important potential for improvement of evaluation. The objectives of an educational effort are still an important component of evaluation. Evaluation theorists indicate that evaluation should attend to outcomes other than specified objectives, to inputs or antecedent conditions, and to processes and transactions. Measurement strategies were also discussed in this article.

In evaluating the importance of sampling definition in relation to consistency of epidemiological research results in schizophrenia, four samples of patients from various combinations of inpatient and outpatient facilities in Monroe County, New York, were selected for study. The purpose of the study was to provide empirical data which would specify degrees of representativeness and, hence, the limits of generalization that would probably be associated with specific sampling strategies. The results of distributions across and analyses within samples indicated that variations in sampling source can produce highly divergent research findings. [ABSTRACT FROM AUTHOR]

Researcher Herbert Blumer argues in favor of a particular definition of the term public opinion. In effect Blumer takes the pollsters to task for not formally defining their field of study. He states that the implicit definition on which they have been working is sociologically incorrect because it assumes that public opinion is adequately measured by an unweighted count of individual opinions and he suggests that pollsters should be studying a kind of public opinion that weights individual opinion in terms of their actual impact on the people who make social decisions. Blumer is saying that public opinion survey people have failed to define what they are after. Blumer is criticizing as a sociologist so he is naturally concerned with the possibility of sociological generalizations that may come out of polling. He goes much too far, however, in implying as strongly as he does that the polls should be conducted so as to provide those generalizations. Some polls run by universities may be operated in the interests of pure science but the ones to whom Blumer refers are primarily public service institutions. As such they operate in the field of applied science, and pure scientific generalizations are incidental by-products, not the basic reasons for their existence as Blumer apparently would like them to be. It is not fair, therefore, to criticize the polls for something they have no business to be doing.

The article presents information on a study which empirically compared the psychometric assessment of correlation matrices. The initial procedure involved the application of Bartlett's test of sphericity. The Bartlett test led to a clear rejection of the hypothesis of interest for three of the four matrices. Three tests for psychometric adequacy have been applied to four well-known correlation matrices from the literature. Those matrices represented somewhat disparate conditions under which factor analysis has been utilized. Sample sizes ranged from 50 to over three thousand. The results of the tests have been cited and some speculative conclusions have been drawn. Initially, the power of the Bartlett test seems quite sensitive to the sample size. It seems possible that erroneous results might still be obtained by routinely using the Bartlett test. The anti-image covariance matrix allows comparable decisions. One of the advantages of measures of sampling adequacy lies in the fact that it provides specific assessment of individual variables.

By both random and nonrandom assignments of numbers to rank orders (which are consistent with the monotonic nature of the categories), it is shown that ordinal variables can be treated as if they conform to interval scales. The scoring systems, of which 18 were randomly generated by a computer, resulted in negligible error when comparing any assigned scoring system with any selected "true" scoring system. Errors are determined by the Pearsonian correlation coefficient (r) and r². The advantages of treating ordinal variables as interval are demonstrated with regard to the relation between occupational prestige and suicide. These advantages include: (1) the use of more powerful, sensitive, better developed and interpretable statistics with known sampling error, (2) the retention of more knowledge about the characteristics of the data, and (3) greater versatility in statistical manipulation (e.g., partial and multiple correlation and regression, analysis of variance and covariance, and most pictorial presentations). The computer approach to this problem does not exhaust all possibilities for assigning numbers, which partially limits the generality of the findings. [ABSTRACT FROM AUTHOR]

Standard statistical literature has been developed almost entirely in terms of independent observations obtained by simple random sampling (SRS). The statistical tests and confidence intervals one finds in textbooks and in journal articles are based on the assumption that the sample observations were selected independently, at random. Although this assumption either appears in "fine print" or not at all, it constitutes the basis of all confidence intervals and tests of hypotheses, their validity is based on it. The ubiquity of the SRS assumption in statistical theory can be explained by the basic nature of SRS and by its facilitation of interesting theoretical results. On the other hand, most social research, especially survey work, is in fact carried out by means of complex sample designs. Simple random selection of human populations is a rare phenomenon, limited to small and confined populations. Socially important human populations are usually large and widely scattered, so that a simple random selection of them would prove in most cases to be uneconomical and impractical.

This article illustrates the point that sample results are both a function of the way in which the universe is defined by the investigator and the manner in which he chooses items from the field of content that defines the topic. An implication is that conclusions concerning the scalability of a given topic must be restricted to some one's version or conception of that topic, assuming a sampling method free of bias. If sampling is selective, then inferences can hardly be extended to any totality of content, no matter how conceived. Evidently, an objective method of listing content is needed in order to lift the foregoing restriction. Two procedures suggest themselves. First, construct the universe operationally in terms of a content analysis of some body of material that lends itself to analysis by this technique. The list of items yielded by this procedure could then be considered as the universe of content from which samples of items could be randomly drawn. An important limitation of this method is that the body of material selected for content analysis may not coincide with whatever was originally thought of as defining the topic — attitude, opinion, value, concept.

The author feels researcher Herbert Blumer must have not tried to direct his criticism toward all the people who use polling as a device, all at once. Blumer should have aimed at a sharply defined segment instead of the entire spectrum. Some of those who use polling devices are preoccupied with techniques. A good many of them reveal an absence of using specific studies to test a general proposition. Blumer's major target could have been a real one instead of an illusory one. Public opinion is the pattern of the diverse views and positions on the issue that come to the individuals who have to act in response to that pattern. This is one of many possible identifications as a class of phenomena. There have been many other identifications of the class which goes by the same name in various times and in various places. But it is not the kind of generic object which the U.S. pollers have in mind during the 1940s. The author thinks that when Blumer judges that polling is unable to isolate public opinion as an abstract or generic concept, he really means that he prefers his own formulation of the concept to that of others. The inherent deficiency of public opinion polling, he says, is contained in its sampling procedure. And yet his comments have largely to do with the kind of questions asked by those who use sampling procedures. The author feels Blumer could have shown some inherent connection between sampling and failure to obtain adequate information.

This article presents some observations on public opinion and on public opinion polling. The observations are designed to invite attention to whether public opinion polling actually deals with public opinion. In this article the author has presented criticisms on public opinion polling as a method for the recording and measurement of public opinion. These criticisms have centered around the distortion that stems from the use of a sample in the form of an aggregation of disparate individuals having equal weight. The criticism applies when such a sampling procedure is used to study a matter whose composition is an organization of interacting parts instead of being merely an aggregation of individuals. Many actions of human beings in a society are mass actions such as casting ballots, purchasing tooth paste, going to motion picture shows, and reading newspapers. Such actions lend themselves readily to the type of sampling that are prevalent in the current public opinion polling. In fact, it is the existence of such mass actions of individuals which explains, the successful use in consumer research of sampling such as is employed in public opinion polling. This article criticizes the use of such sampling with its implicit imagery and logic in the study of a matter which, like the process of public opinion, functions as a moving organization of interconnected parts.

Field interviews of individuals in their homes and in the community setting reveal problems not present in experimental designs that are set up in the comparatively simple school-room situation, familiar in educational research. The purpose of experimental designs applied to the study of social factors is to measure the effect of one social factor at a time, all other factors in the situation being held constant. Randomization is theoretically the solution to obtaining control of these unknown factors. A random sample from the population is to be selected, which may then be divided at random into the experimental group which is to receive the social program. Each group is then to be measured simultaneously on some criteria designed to evaluate the effect of the program. The practical difficulty to be overcome is that of cost. More serious than this difficulty is whether directors of a social program will be willing to give their treatment to a randomly selected group and exclude from treatment another identical group to provide a theoretically sound experimental design.

Using the effect size definitions and metric-free values proposed by Cohen (1969), power calculations were made for every standard statistical test re-porting significance in AERJ for the period November 1969 to May 1971. There were 154 F ratio, 51 Student's t and 168 correlational tests for which power was calculated and reported. The power values were found to be quite low with combined averages of .14, .58 and .78 for small, medium and large effect sizes respectively. In an effort to minimize the ambiguities of research reports, several reporting schemes were proposed which involve a reporting of power, a, effect size and sample size for all research reports utilizing classical hypothesis testing. [ABSTRACT FROM AUTHOR]

The present study investigated the efficacy of prediction of Horst's Absolute and Differential Prediction Methods when applied to within-groups correlation matrices. Two treatments of missing data, employing increases of total group intercorrelation matrices and subgroup intercorrelation matrices were incorporated in both the Absolute and Differential Prediction Methods. The synthesis of these methods generated four composite weighting systems, thus permitting a comparison of the accuracy of predictions resulting from the four systems on the basis of weight validities (cross-validation correlation coefficients). [ABSTRACT FROM AUTHOR]

This article focuses on the review of methods and empirical studies in education. This article is devoted to a systematic study of differential weighting. In the first section, different types of weights and methods of deriving weights are discussed, and the mathematical restrictions which limit the effectiveness of many sets of weight are pointed out. The second section contains a summary of empirical investigations of weighting in each of the typical situations where weighting has been considered potentially useful. In the final section, consideration is given to response-determined scoring, a term which implies that a range of scores is possible for a given test item and that it is the subject's response pattern which determines his scores.

This Chapter represents a departure in two ways from past statistical reviews published in the Review of Educational Research. First, rather than presenting a review of the literature, the authors were encouraged to evaluate current research and to give their own viewpoints as to the past and future practice of statistics. Second, it was apparent that no one person could accomplish this task because of the specialized nature of the field of Statistics today. The chapter is divided into four sections. The topics presented are Bayesian statistics, multivariate statistics, non-parametric statistics and Monte Carlo methods. Each section was independently authored with no effort made towards integration. The authors do not claim comprehensive coverage of even the areas of statistics represented here. [ABSTRACT FROM AUTHOR]

This article discusses sample surveys in education. In a general sense, sample surveys are a special application of the central problem of theory and methods of statistical inference, namely, regulating the degree of uncertainty. The idea of randomization, originally introduced by Fisher, by which statistician introduces an element of chance into the selection of observations and thus creates a random experiment to be a link between several branches of design for collecting and analyzing such observations as experiment. In education, the use of sample surveys is growing. Mathematical statisticians have been employed by the U.S. Office of Education for assistance in sample design. Sample surveys published by the office dealt with teacher turnover, the status of beginning teachers, programs in the schools for home economics, the organization of elementary schools in large cities, and the status of adult education. A list of limitations of educational applications is also presented and discussed in this article.

The article presents the author's comments on three recent notes on N versus N-1 that have dealt with the pedagogical problem of how best to define the sample variance. The primary consideration is that this particular unbiased sample function has minimum variance in the class of all possible estimators. The mean square error for a good estimator also has to be at a minimum. Even though this second inequality is weaker than the first, estimators with a minimum mean square error usually do not exist. The final comment about the correlation coefficient is in order. The authors referenced in the first paragraph discuss the problem of using either a biased or an unbiased estimate of the population variance in calculating the sample correlation. Nothing as simple as changing a constant, however, will be of much help in providing an optimal estimate of the population correlation coefficient. A minimum variance unbiased estimator has been obtained but it requires a complicated representation as an infinite series.

In replying to the statements of researcher Julian Woodward and researcher Theodore M. Newcomb the author wishes to confine to a single point. The crucial difference between their views and the author's lies in a methodological matter whether a procedure like public opinion polling is to gain its justification through its adherence to a technique or, instead, through a realistic analysis of what it proposes to study, Woodward and Newcomb regard current public opinion polling as operating with a conception of public opinion that is as tenable as any alternative conception. Thus, they seem to regard the author's criticism as having no value in that it is merely the application of a different conception of public opinion, a conception which has no preferential status. The problem is precisely one of whether one proposes to study public opinion with a conception that is true to its empirical character or whether one proposes to study it with a conception which is patently unrealistic. The author submits that current public opinion polling necessarily operates with a conception of public opinion that is a gross distortion. By virtue of its sampling procedure, current public opinion polling is forced to regard public opinion as an aggregate of equally weighted opinions of disparate individuals. If it be granted that public opinion polling as it functions in society is organic, then public opinion polling as currently practiced cannot yield a faithful or even a meaningful account of public opinion. In the author's judgment there is no scientific virtue in a study which blinds itself to the obvious character of what it proposes to study.

The use of a test proposed in the article "Interstate Migration Differentials," by James D. Tarver raises some important considerations. With the corrected error term entered into the test formula, all of the differences between age categories greater than 0.56 in the author's table as mentioned in the article, attain significance or, in other words, 18 of the 21 comparisons. This article examines the three differences that failed to reach significance, taking into account that the migration rate data are proportions based on sample sizes in the neighborhood of 100,000. A test between proportions based on this sample size yields a normal deviate greater than 6.0 for even the small difference of 0.3 between age categories seven and six. Obviously, the difference between age categories one and is also highly significant. That there is great regularity in these data is indicated by the fact that sign tests of 20 of the 21 differences, based on 26 observations. With the aid of computers, it is frequently feasible to present simultaneously means or other statistics for samples from more than two different populations.