Some metrical issues with meta-analysis of therapy effects

Two metrical issues, both concerning the index of effect size used in meta-analyses of therapy effects, are discussed with reference to such analyses. The first issue refers to use of the standard deviation of change scores as a metrical unit, and the conclusion is that the standard deviation of unadjusted post-test scores should be used instead. The second issue concerns measurement errors, and it is argued that the metrical unit—irrespective of which kind of variability metric is chosen—should be corrected for such errors, hence expressing each therapy effect relative to an estimated true rather than observed unit. It is proposed that a set of metrical alternatives should be used in a meta-analysis, so that the analysis is performed for each alternative and the results are compared across the alternatives.     

Meehl and the ex post facto design

Paul E. Meehl's problem of systematic unmatching is discussed, and it is argued that his reasoning and conclusion are incorrect. Unmatching will probably occur seldom in empirical cases, and if it does take place, it will constitute some sort of undercorrection. Moreover, his attempt at interpretation of statistical control in terms of so-called "counterfactuals" is found to be misleading.     

Schooling and intelligence: A Methodological and logitudinal study

A general account of the adjustment problem is given within the frame of pretest-posttest design, and analysis of covariance, change score analysis and standardized change score analysis are discussed with reference to models of selection and growth. A Norwegian longitudinal investigation of the relationship between schooling and intelligence is described, and the results, generated by several adjustment techniques, showed substantial schooling effects. Thus the difference in estimated effect between a higher and a lower educational group varied from 49 to 73 per cent of the pretest standard deviation. In addition, considerable changes from pretest to posttest were observed, the mean gain amounting to 114 per cent of the pretest standard deviation. A comparison with two Swedish studies yielded similar trends in estimated schooling effect in that one study had higher and the other lower estimates than the present one.     

Multidimensional scaling of political parties A methodological study

Abstract.— With 7 Norwegian parties as stimuli, comparisons were made of various content and distance approaches to scaling. As expected, the former gave two factors, socialistic and nonsocialistic, and the latter gave one bipolar dimension, interpreted as socialistic-non-socialistic. Individual differences in structure were not found. Among several principles of similarity, Ekman's original (1961) gave the best fit. It is maintained, however, that the fundamental vector model of the content approach is inadequate, both for theoretical reasons and because subjects are unable to operate in accordance with it.     

A comparison between the Grid Test of Schizophrenic Thought Disorder and diagnostic psychologicaI testing

Abstract.— Bannister and Fransella's Grid Test of Schizophrenic Thought Disorder based on Personal Construct Theory and the concept of "loosened construing", as measurable in the scores of Intensity and Consistency between intercorrelations, has been applied to a Scandinavian sample of psychiatric patients. The validity of the test was illustrated, and a significant difference was found between a group of schizophrenics and schizophrenic borderline states and a group without schizophrenic thought disorders. The Grid Test scores were found to agree with thought disorder manifestations as evaluated in qualitative terms on the basis of cognitive and projective tests, but they did not differentiate between developmental levels of thinking corresponding to con-creteness in organic impairment versus diffuseness in schizophrenia. Thus, the concept of "loosened construing" as applied in this thought disorder test seems too unspecific.     

Causal relations in field research as a basis for practical control

The utility of causal relations in field research for practical control is critically examined. It is argued that generalizations involving hypothetical intervention or change, formulated in terms of the parameters of causal regression functions for natural variation, are incompatible with the formal selection requirements for use of these functions. Practical results may thus deviate markedly from those predicted by the functions. Consequently, the causal relations commonly measured in field research within psychology and related disciplines cannot be used in any strict sense for the purpose of practical control. This conclusion, however, applies to generalizations involving intervention which are based on natural variation, not necessarily to those based on manipulated variation. The account demonstrates the importance of adopting a concept of causality which makes a clear distinction between causal measurements, on the one hand, and generalizations from such results, on the other.     

A methodological note concerning long-term effect of treatment

The long-term effect of treatment is defined as the score change produced directly and indirectly by the treatment during and after the treatment interval. Various instances of combinations of short-term effects, attained during the treatment interval, and of long-term effects are presented. Designs for measuring the short-term and long-term effects are sketched, and suggestions are given for distinguishing between these effects in six representative cases. A multiple-occasion measurement of the long-term effect is recommended.     

The statistical regression phenomenon: I. A metamodel

A metamodel for the traditional regression concept—consisting of two general parts—is proposed. The first part contains the basic and general definition of the concept, together with the central assumptions. The phenomenon is essentially defined as a conditional trend, predicted within a linear and bivariate regression system for a specified population, and as a function of the imperfectness of the correlation and of the extremeness of the selected score. The second part of the metamodel consists of special cases of the regression phenomenon, which are defined with respect to various psychometrical/distributional aspects, to various "sources" which reduce the correlation coefficient, and to observed and true scores. Also, the general features of the metamodel are elaborated, and correct and incorrect interpretations are pointed out.