Logistic Approximation to the Normal: The KL Rationale

A rationale is proposed for approximating the normal distribution with a logistic distribution using a scaling constant based on minimizing the Kullback–Leibler (KL) information, that is, the expected amount of information available in a sample to distinguish between two competing distributions using a likelihood ratio (LR) test, assuming one of them is true. The new constant 1.749, computed assuming the normal distribution is true, yields an approximation that is an improvement in fit of the tails of the distribution as compared to the minimax constant of 1.702, widely used in item response theory (IRT). The minimax constant is by definition marginally better in its overall maximal error. It is argued that the KL constant is more statistically appropriate for use in IRT. The author would like to thank Sebastian Schreiber for his generous assistance with this project.     

Table of mean deviates for various portions of the unit normal distribution

In the absence of sound units of measurement, the data obtained by use of a psychometric instrument may sometimes be so irregular that they cannot be accepted as a good representation of the true function of the trait measured. Frequently, ordered categorical or ranked data and other irregular distributions are obtained to represent traits that are obviously continuous and probably unskewed functions. In such cases, the data may be improved in the direction of a better approximation of the true function by converting them to normal distribution form. A table of mean deviates of various portions of the unit normal curve is presented to facilitate this conversion.The author wishes to acknowledge the kind counsel of Professor Truman L. Kelley in the preparation of this article.     

Calculation of the tetrachoric correlation coefficient

A new subroutine has been developed for calculating the terachoric correlation coefficient. Recent advances in computing inverse normal and bivariate normal distributions have been utilized. The iterative procedure is started with an approximation with an error less than±.0135.I am grateful to the Editor for valuable suggestions for improving the presentation.     

Higher-order approximations to the distributions of fit indexes under fixed alternatives in structural equation models

Higher-order approximations to the distributions of fit indexes for structural equation models under fixed alternative hypotheses are obtained in nonnormal samples as well as normal ones. The fit indexes include the normal-theory likelihood ratio chi-square statistic for a posited model, the corresponding statistic for the baseline model of uncorrelated observed variables, and various fit indexes as functions of these two statistics. The approximations are given by the Edgeworth expansions for the distributions of the fit indexes under arbitrary distributions. Numerical examples in normal and nonnormal samples with the asymptotic and simulated distributions of the fit indexes show the relative inappropriateness of the normal-theory approximation using noncentral chi-square distributions. A simulation for the confidence intervals of the fit indexes based on the normal-theory Studentized estimators under normality with a small sample size indicates an advantage for the approximation by the Cornish–Fisher expansion over those by the noncentral chi-square distribution and the asymptotic normality. The author is indebted to the reviewers for their comments and suggestions, which have led to the improvement of the previous versions of this paper. This work was partially supported by Grant-in-Aid for Scientific Research from the Japanese Ministry of Education, Culture, Sports, Science, and Technology.     

Asymptotic expansion of the distributions of the estimators in factor analysis under non‐normality

Equations for the Edgeworth expansion of the distributions of the estimators in exploratory factor analysis and structural equation modeling are given. The equations cover the cases of non‐normal data, as well as normal ones with and without known first‐order asymptotic standard errors. When the standard errors are unknown, the distributions of the Studentized statistics are expanded. Methods of constructing confidence intervals of population parameters with arbitrary asymptotic confidence coefficients are given using the Cornish‐Fisher expansion. Simulations are performed to see the usefulness of the asymptotic expansions in exploratory factor analysis with rotated solutions and confirmatory factor analysis. The results show that asymptotic expansion gives substantial improvement of approximation to the exact distribution constructed by simulations over the usual normal approximation.     

Abstract: Comparing Semiparametric and Parametric Methods for Modeling Interactions Among Latent Variables

Many approaches have been proposed to estimate interactions among latent variables. These methods often assume a specific functional form for the interaction, such as a bilinear interaction. Theory is seldom specific enough to provide a functional form for an interaction, however, so a more exploratory, diagnostic approach may often be required. Bauer (2005) proposed a semiparametric approach that allows for the estimation of interaction effects of unknown functional form among latent variables. A structural equation mixture model (SEMM) is first fit to the data. Then an approximation of the interaction is obtained by aggregating over the mixing components. A simulation study is used to examine the performance of this semiparametric approach to two parametric approaches: the latent moderated structures approach (Klein & Moosbrugger, 2000) and the unconstrained product-indicator approach (Marsh, Wen, & Hau, 2004). Data were generated from four functional forms: main effects only, quadratic trend, bilinear interaction, and exponential interaction. Estimates of bias and root mean squared error of approximation were calculated by comparing the surface used to generate the data and the model-implied surface constructed from each approach. As expected, the parametric approaches were more efficient than the SEMM. For the main effects model, bias was similar for both the SEMM and parametric approaches. For the bilinear interaction, the parametric approaches provided nearly identical results, although the SEMM approach was slightly more biased. When the parametric approaches assumed a bilinear interaction and the data were generated from a quadratic trend or an exponential interaction, the parametric approaches generated biased estimates of the true surface. The SEMM approach approximated the true data generation surface with a similarly low level of bias for all the nonlinear surfaces. For example, Figure 1 shows the true surface for the bilinear interaction along with the SEMM estimated average surface. The results suggest that the SEMM approach can provide a relatively unbiased approximation to variety of nonlinear relationships among latent variables.     

Sequential Computerized Mastery Tests—Three Simulation Studies

A simulation study of a sequential computerized mastery test is carried out with items modeled with the 3 parameter logistic item response theory model. The examinees' responses are either identically distributed, not identically distributed, or not identically distributed together with estimation errors in the item characteristics. The simulations indicated that the observed results from the operating characteristic function differ significantly from the theoretical results, which is probably due to the use of an approximation formula. The mean number of items in a test, the distribution of test length, and the variance depend highly on how well we know the true values of the item characteristics and whether they are identically distributed or not.     

On the Mental Representation of Conditional Sentences

Four experiments are reported which attempt to externalize subjects' mental representation of conditional sentences, using novel research methods. In Experiment 1, subjects were shown arrays of coloured shapes and asked to rate the degree to which they appeared to be true of conditional statements such as 'If the figure is green then it is a triangle'. The arrays contained different distributions of the four logically possible cases in which the antecedent or consequent is true or false: TT, TF, FT, and FF. For example, a blue triangle would be FT for the conditional quoted above. In Experiments 2 to 4, subjects were able to construct their own arrays to make conditionals either true or false with any distribution of the four cases they wished to choose. The presence and absence of negative components was varied, as was the form of the conditional, being either 'if then' as above or 'only if': 'The figure is green only if it is a triangle'. The first finding was that subjects represent conditionals in fuzzy way: conditionals that include some counter-example TF cases (Experiment 1) may be rated as true, and such cases are often included when subjects construct an array to make the rule true (Experiments 2 to 4). Other findings included a strong tendency to include psychologically irrelevant FT and FF cases in constructed arrays, presumably to show that conditional statements only apply some of the time. A tendency to construct cases in line with the 'matching bias' reported on analogous tasks in the literature was found, but only in Experiment 4, where the number of symbols available to construct each case was controlled. The findings are discussed in relation to the major contemporary theories of conditional reasoning based upon inference rules and mental models, neither of which can account for all the results.     

Are True Scores and Construct Scores the Same? A critical examination of their substitutability and the implications for research results

Relations between constructs are estimated based on correlations between measures of constructs corrected for measurement error. This process assumes that the true scores on the measure are linearly related to construct scores, an assumption that may not hold. We examined the extent to which differences in distribution shape reduce the correlation between true scores on a measure and scores on the underlying construct they are intended to measure. We found, via a series of Monte Carlo simulations, that when the actual construct distribution is normal, nonnormal distributions of true scores caused this correlation to drop by an average of only .02 across 15 conditions. When both construct and true score distributions assumed different combinations of nonnormal distributions, the average correlation was reduced by .05 across 375 conditions. We conclude that theory‐based scales intended to measure constructs usually correlate highly with the constructs they are constructed to measure. We show that, as a result, in most cases true score correlations only modestly underestimate correlations between different constructs. However, in cases in which the two constructs are redundant, this underestimation can lead to the false conclusion that the constructs are ‘correlated but distinct constructs,’ resulting in construct proliferation.     

An Overview of Possibilistic Handling of Default Reasoning, with Experimental Studies

This paper first provides a brief survey of a possibilistic handling of default rules. A set of default rules of the form, “generally, from α deduce β”, is viewed as the family of possibility distributions satisfying constraints expressing that the situation where α and β is true has a greater plausibility than the one where a and - β is true. When considering only the subset of linear possibility distributions, the well-known System P of postulates proposed by Kraus, Lehmann and Magidor, has been obtained. We also present two rational extensions: one based on the minimum specificity principle and the other is based on the lexicographic ordering. The second part of the paper presents an empirical study of three desirable properties for a consequence relation that capture default reasoning: Rationality, Property Inheritance and Ambiguity Preservation. An experiment is conducted to investigate 13 patterns of inference for the test of these properties. Our experimental apparatus confirms previous results on the relevance of System P, and enforces the psychological relevance of the studied properties.     

An approximation for the bias function of the maximum likelihood estimate of a latent variable for the general case where the item responses are discrete

Lord developed an approximation for the bias function for the maximum likelihood estimate in the context of the three-parameter logistic model. Using Taylor's expansion of the likelihood equation, he obtained an equation that includes the conditional expectation, given true ability, of the discrepancy between the maximum likelihood estimate and true ability. All terms of orders higher thann ^?1 are ignored wheren indicates the number of items. Lord assumed that all item and individual parameters are bounded, all item parameters are known or well-estimated, and the number of items is reasonably large. In the present paper, an approximation for the bias function of the maximum likelihood estimate of the latent trait, or ability, will be developed using the same assumptions for the more general case where item responses are discrete. This will include the dichotomous response level, for which the three-parameter logistic model has been discussed, the graded response level and the nominal response level. Some observations will be made for both dichotomous and graded response levels.     

Bayesian item selection criteria for adaptive testing

Owen (1975) proposed an approximate empirical Bayes procedure for item selection in computerized adaptive testing (CAT). The procedure replaces the true posterior by a normal approximation with closed-form expressions for its first two moments. This approximation was necessary to minimize the computational complexity involved in a fully Bayesian approach but is no longer necessary given the computational power currently available for adaptive testing. This paper suggests several item selection criteria for adaptive testing which are all based on the use of the true posterior. Some of the statistical properties of the ability estimator produced by these criteria are discussed and empirically characterized.Portions of this paper were presented at the 60th annual meeting of the Psychometric Society, Minneapolis, Minnesota, June, 1995. The author is indebted to Wim M. M. Tielen for his computational support.     

Asymptotic distributions of the estimators of communalities in factor analysis

Asymptotic distributions of the estimators of communalities are derived for the maximum likelihood method in factor analysis. It is shown that the common practice of equating the asymptotic standard error of the communality estimate to the unique variance estimate is correct for standardized communality but not correct for unstandardized communality. In a Monte Carlo simulation the accuracy of the normal approximation to the distributions of the estimators are assessed when the sample size is 150 or 300. This study was carried out in part under the ISM Cooperative Research Program (90-ISM-CRP-9).     

The variance of d' estimates obtained in yes-no and two-interval forced choice procedures

In studies of detection and discrimination, data are often obtained in the form of a 2 x 2 matrix and then converted to an estimate of d' based on the assumptions that the underlying decision distributions are Gaussian and equal in variance. The statistical properties of the estimate of d', d' are well understood for data obtained using the yes-no procedure, but less effort has been devoted to the more commonly used two-interval forced choice (2IFC) procedure. The variance associated with d' is a function of true d' in both procedures, but for small values of true d' the variance of d' obtained using the 2IFC procedure is predicted to be less than the variance of d' obtained using yes-no; for large values of true d', the variance of d' obtained using the 2IFC procedure is predicted to be greater than the variance of d' from yes-no. These results follow from standard assumptions about the relationship between the two procedures. The present paper reviews the statistical properties of d' obtained using the two standard procedures and compares estimates of the variance of d' as a function of true d' with the variance observed in values of d' obtained with a 2IFC procedure.     

A tutorial on approximate Bayesian computation

This tutorial explains the foundation of approximate Bayesian computation (ABC), an approach to Bayesian inference that does not require the specification of a likelihood function, and hence that can be used to estimate posterior distributions of parameters for simulation-based models. We discuss briefly the philosophy of Bayesian inference and then present several algorithms for ABC. We then apply these algorithms in a number of examples. For most of these examples, the posterior distributions are known, and so we can compare the estimated posteriors derived from ABC to the true posteriors and verify that the algorithms recover the true posteriors accurately. We also consider a popular simulation-based model of recognition memory (REM) for which the true posteriors are unknown. We conclude with a number of recommendations for applying ABC methods to solve real-world problems.     

Suppression of valid inferences and knowledge structures: the curious effect of producing alternative antecedents on reasoning with causal conditionals

These studies looked at the difficulty that reasoners have in accepting conditional ("If P then Q") major premises that are not necessarily true empirically, as a basis for deductive reasoning. Preliminary results have shown that when reasoners are asked to produce possible alternate antecedents to the major premise ("If A then Q"), they paradoxically tend to deny the modus ponens (MP) inference ("If P is true, then Q is true"). Three studies further explored these results. The first study gave university students paper-and-pencil tests in which instructions to "suppose that the major premise is true" was followed by a request to determine the next number in a sequence, to retrieve information unrelated to the premises, or to retrieve a possible case of "If A then Q." Relative to a control group, reasoners asked to produce an alternative antecedent showed a significant tendency to deny the MP inference, whereas no such tendency was observed for the two other tasks used. A second study compared performance on a condition in which reasoners were asked to produce an alternative antecedent with that when they were given an explicit alternative. Premises used in this study were such that the latter alternative antecedent was also spontaneously produced by over 70% of reasoners. Results showed that the tendency to refuse the MP premise could not be accounted for by the specific nature of the alternative produced. A third study found that the tendency to refuse the MP inference after producing an alternative antecedent was affected by the number of "disabling conditions" (i.e., conditions that allow "P to be true" and "Q to be false") available for the major premise. These results are interpreted as being consistent with a model that supposes that logical reasoning requires selective inhibition of real-world knowledge.     

Proper Names and the Necessity of Identity Statements

An identity statement flanked on both sides with proper names is necessarily true, Saul Kripke thinks, if it's true at all. Thus, contrary to the received view – or at least what was, prior to Kripke, the received view – a statement like(A) Hesperus is Phosphorusis necessarily true if, as certainly seems to the the case, it's true at all. The received view is that (A) is true but only contingently true, while(B) Hesperus is Hesperuswhich is also true, of course, is necesarily true. Epistemologically, however, both the tradition and Kripke have it that (A) is a posteriori and (B) a priori.There are tensions in Kripke's views concerning (A), though, and ultimately in the views of anyone who holds that (A) is necessary. In this paper I draw attention to some of them and advance an argument for thinking that (A) is contingent.     

Approximation,Idealization, and Laws of Nature

Traditional theories construe approximate truth or truthlikeness as a measure of closeness to facts, singular facts, and idealization as an act of either assuming zero of otherwise very small differences from facts or imagining ideal conditions under which scientific laws are either approximately true or will be so when the conditions are relaxed. I first explain the serious but not insurmountable difficulties for the theories of approximation, and then argue that more serious and perhaps insurmountable difficulties for the theory of idealization force us to sever its close tie to approximation. This leads to an appreciation of lawlikeness as a measure of closeness to laws, which I argue is the real measure of idealization whose main purpose is to carve nature at its joints. This revised version was published online in June 2006 with corrections to the Cover Date.