摘要:

Recently, Valentine, Wilding&Mohindra (1984) argued that the negative correlation between slopes and intercepts sometimes found in memory‐scanning tasks can be explained on purely statistical grounds. Because theory underlying the memory‐scanning literature is so emphatic in stating that this should not be so, a more exhaustive examination of other possibilities was undertaken. In this paper we consider a more geometric approach and observe that the geometric properties of the transformationx→x + tdetermine the correlations obtainable from the variables ‘slope’ and ‘intercept’ associated with the kinds of data given in memory scan experiments. The family of lines produced are naturally negatively correlated for good geometric reasons. Even though our observations do not contradict the conclusions obtained by Valentineet al., our method is sufficiently different to be considered in its own right. It is hoped that the geometric considerations of the type presented here may find application in interpreting data obtained from families of regression lines representing runs o

ISSN:0007-1102    

DOI:10.1111/j.2044-8317.1991.tb00948.x

出版商:Blackwell Publishing Ltd    

年代:1991

数据来源: WILEY

摘要:

A bipartite graph, in which the nodes (or actors in a social network) are partitioned into two sets, can be studied using recent statistical models for dyadic interactions. These models, which are loglinear for the probabilities of dyadic choices or interactions, allow not only arcs or relationships to exist between the sets but also within the sets. Thus, the methods described here are applicable not only to bipartite graphs, consisting of arcs existing between nodes in different sets, but also to directed graphs that are defined within the two sets of nodes. Data on both types of graphs can be analysed simultaneously.A bipartite graph has an adjacency matrix (or sociomatrix) with two ‘modes’. The set of nodes in the row mode differs from the set of nodes in the column mode. For example, in marketing, one could study the dyadic relations in a ‘buyers‐by‐sellers’ network. Generally, the relations observed in a one‐mode network, which has a square sociomatrix (row mode = column mode) are bidirectional—the actors indexing the columns may also ‘relate to’ the actors indexing the rows. The relations observed in a two‐mode network are generally unidirectional—the row actors relate to or choose the column actors, but the column actors do not relate to the row actors. Referring to our example, a buyer might pay a seller for some item, but a seller would not pay a buyer.Statistical models for the separate analysis of these one‐mode and two‐mode matrices are extended in this paper to the simultaneous analysis of both types of networks. A superordinate one‐mode sociomatrix is created in which the rows and columns consist of all actors (that is, all buyers and sellers). This larger matrix contains both the one‐mode matrices and the two‐mode matrices. Multivariate analysis of unidirectional and bidirectional relations in social networks and complex directed graphs becomes possible with this simultaneous consid

ISSN:0007-1102    

DOI:10.1111/j.2044-8317.1991.tb00949.x

出版商:Blackwell Publishing Ltd    

年代:1991

数据来源: WILEY

摘要:

A new approach to the development of the item characteristic curve (ICC) is presented, in which knowledge states, decision processes and other circumstances underlying responding to objective tests receivea prioriconsideration. Earlier work on finite state models of objective test performance provides the basis for deriving expressions for ICCs that directly account for factors such as examinee willingness to guess, mode of test administration, number of options per item, and the response strategy of the examinees. This approach utilizes a parameterization of ability different from that used in conventional item response theory (IRT) and yields ICCs that are polynomial functions of ability. The degree and coefficients of these polynomials depend in part on psychological/circumstantial factors such as those just mentioned or others that may readily be introduced. Examples are provided to show how differing assumptions about objective test responding lead to variation in the shapes of the resulting ICCs. The advantages that IRT could gain from adoption of these ICCs are discussed, and the work that remains to be done before finite state polynomic ICCs can be used in practice is outlined. Some possible extensions to the finite state approach are also discussed.

ISSN:0007-1102    

DOI:10.1111/j.2044-8317.1991.tb00950.x

出版商:Blackwell Publishing Ltd    

年代:1991

数据来源: WILEY

摘要:

The polychotomous Rasch model is generalized to a mixture distribution model. It is assumed that the observed data are generated by two or more latent classes of individuals so that within each class the polychotomous Rasch model holds but with different parameters between the classes. Hence, the proposed model is also a generalization of latent class analysis which allows for quantitative individual differences within the classes. A parameter estimation procedure is outlined, employing conditional inference methods for the item parameters within classes and the EM‐algorithm for unmixing the data. The application of the model and control of model fit are illustrated by means of real data and simulated dat

ISSN:0007-1102    

DOI:10.1111/j.2044-8317.1991.tb00951.x

出版商:Blackwell Publishing Ltd    

年代:1991

数据来源: WILEY

摘要:

Researchers must be familiar with a variety of source literature to facilitate the informed use of structural equation modelling. Knowledge can be acquired through the study of an expanding literature found in a diverse set of publishing forums. We propose that structural equation modelling publications can be roughly classified into two groups: (a) technical and (b) substantive applications. Technical materials focus on the procedures rather than substantive conclusions derived from applications. The focus of this article is the former category; included are foundational/major contributions, minor contributions, critical and evaluative reviews, integrations, simulations and computer applications, precursor and historical material, and pedagogical textbooks. After a brief introduction, we annotate 294 articles in the technical category dating back to Sewall Wright (1921).

ISSN:0007-1102    

DOI:10.1111/j.2044-8317.1991.tb00952.x

出版商:Blackwell Publishing Ltd    

年代:1991

数据来源: WILEY

摘要:

Two new indices each based on two different methods of estimating the population covariance matrix Σ are derived. Certain algebraic results on these indices are also reported. A case study has been presented to show how the proposed indices in comparison to Jöreskog‐Sörbom's GFI are more efficient in detecting the ‘best’ structure out of 13 competing structures hypothesized for the Graduate Record Examination data. Algebraically it has also been shown that the limiting value of GFI fails to reach the desired value of zero under model misspecification. A Monte Carlo study also shows that with increasing sample size the proposed indices as compared to others show more and more sensitivity to model misspecification. Other advantages of the indices as measures of structural closeness ofZto S are d

ISSN:0007-1102    

DOI:10.1111/j.2044-8317.1991.tb00953.x

出版商:Blackwell Publishing Ltd    

年代:1991

数据来源: WILEY

摘要:

This paper argues that Ross's (1939) method of obtaining presentation sequences for use in paired comparisons deserves to be better known. The method is reformulated for computer implementation and extended to deal with even numbers of items. A program for generating presentation sequences is included.

ISSN:0007-1102    

DOI:10.1111/j.2044-8317.1991.tb00954.x

出版商:Blackwell Publishing Ltd    

年代:1991

数据来源: WILEY

摘要:

Wide use has been made of multidimensional scaling (MDS) techniques since the pioneering papers of Shepard and Kruskal. In the main, dissimilarities used in the various MDS techniques are derived for pairs of objects or stimuli. This is termed 2‐way, 1‐mode data, meaning pairs of objects within a single set are considered. Some MDS techniques are designed for 3‐way or even higher, and for 2‐mode, 3‐mode or more. One such example is the CANDECOMP model which can deal withn‐way,m‐mode data where 3≤n≤7 and 2≤m≤7. This model considersn‐tuples of objects at a time, selecting these from m different sets.To date there are no models which considern‐way, 1‐mode data, where n≥3. An attempt is made in this paper to cater for this situation by an extension of the Shepard‐Kruskal approach to non‐metric multidimensional scaling which deals with dissimilarities defined for three or more objects. A computer program has been written using the new model to produce a configuration in Euclidean space to represent the objects. Some historical voting data and some

ISSN:0007-1102    

DOI:10.1111/j.2044-8317.1991.tb00955.x

出版商:Blackwell Publishing Ltd    

年代:1991

数据来源: WILEY

摘要:

It is often assumed that balanced factorial designs are desirable as they result in hypotheses to be tested in an analysis of variance which are independent of each other. It is also a consequence of balance that, with the exception of the highest‐order interaction, significant analysis of variance effects derived from balanced designs generalize only to differences between linear functions of cell means with fixed weights. Such generalizations are therefore restricted to replications of complete studies including all the factors not involved in the significant effects. When dealing with non‐experimental factors, e.g. sex, age, etc., there are distinct advantages to designs in which the Observations are Sampled Completely At Random (OSCAR designs). Though this leads to unbalanced designs, analyses can be performed which treat each observation equally. In this way any significant effect generalizes both to the population sampled and to partial replications of the study which omit factors not involved in the particular significant effect. These unconstrained generalizations require that each effect is tested against its within‐effect variance, thus separate error terms are required for each test. Unweighted (equally weighted) means analyses when applied to OSCAR designs suffer from the same demerits as balanced designs and lower‐order effects do not generalize to differences between means in naturally occurring popu

ISSN:0007-1102    

DOI:10.1111/j.2044-8317.1991.tb00956.x

出版商:Blackwell Publishing Ltd    

年代:1991

数据来源: WILEY

摘要:

A common problem is comparing two independent treatment groups in terms of some random variableYwhen there is some covariateX. Typically the comparison is made in terms ofE(Y|X). However, highly skewed distributions occur in psychology (Micceri, 1989), and so a better method might be to compare the two groups in terms of the median ofYgivenX, sayM(Y|X). For thejth group, assume thatM(Y|X) = βjX +αj. Theβsand αs can be estimated with the Brown‐Mood procedure, but there are no results on how one might testH0:β1= β2orH0:α1= α2. Let and be the estimates of α and β, respectively. One of the more obvious approaches is to use a jackknife estimate of the variance of and then assume that the resulting test statistics have a standard normal distribution. This approach was found to be unsatisfactory. Some alternative procedures were considered, some of which gave good results when testingH0:α1= α2, but they were too conservative, in terms of Type I errors, when testingH0:β1= β2. Still another procedure was considered and found to be substantially better than all others. The new procedure is based on a modification of the Brown‐Mood procedure and a bootstrap estimate of the standard errors. Some limitations of the new

ISSN:0007-1102    

DOI:10.1111/j.2044-8317.1991.tb00957.x

出版商:Blackwell Publishing Ltd    

年代:1991

数据来源: WILEY