Mixture model mapping of the brain activation in functional magnetic resonance images.

We report on a novel method of identifying brain regions activated by periodic experimental design in functional magnetic resonance imaging data. This involves fitting a mixture distribution with two components to a test statistic estimated at each voxel in an image. The two parameters of this distribution, the proportion of nonactivated voxels, and the effect size can be estimated using maximum likelihood methods. Standard errors of the parameters can also be estimated. The fitted distribution can be used to derive brain activation maps and two examples are described, one involving a visual stimulation task, the other an auditory stimulation task. The method appears to have some advantages over direct use of the P-values corresponding to each voxel's value of the test statistic.

The use of multivariate statistical methods in psychiatry.

Multivariate methods such as factor analysis and cluster analysis have been used for many years by psychiatric researchers primarily for investigating the problems of psychiatric diagnosis. In more recent times, the range of multivariate techniques used has been extended as has the number of areas in which they are being applied.

Analysis of longitudinal data. Beyond MANOVA.

BACKGROUND: Longitudinal data arise frequently in psychiatric investigations, and are most often analysed by multivariate analysis of variance (MANOVA) procedures. However, as routinely applied, the method is not satisfactory, particularly when the data are affected by subjects dropping-out of the study. More suitable methods are now available. METHOD: Problems with the MANOVA approach are discussed and the advantages of alternative procedures stressed. RESULTS: Using MANOVA on complete cases to analyse unbalanced longitudinal data can be seriously misleading. More recently developed methods are far more suitable, but only if the missing values are non-informative. CONCLUSIONS: Routine use of MANOVA for the analysis of longitudinal data, particularly when there is a substantial proportion of drop-outs, is ill advised. Statisticians have considerably enriched the available methodologies during the past decade, and psychiatric researchers dealing with such data should be aware of the advantages of the newer methods.

The comparison of latent variable models of non-psychotic psychiatric morbidity in four culturally diverse populations.

BACKGROUND: Factor analysis has been employed to identify latent variables that are unifying constructs and that parsimoniously describe correlations among a related group of variables. Confirmatory factor analysis is used to test hypothesized factor structures for a set of variables; it can also, as in this paper be used to model data from two or more groups simultaneously to determine whether they have the same factor structure. METHOD: Non-psychotic psychiatric morbidity, elicited by the Revised Clinical Interview Schedule (CIS-R), from four culturally diverse populations was compared. Confirmatory factor analysis was employed to compare the factor structures of CIS-R data sets from Santiago, Harare, Rotherhithe and Ealing. These structures were compared with hypothetical one and two factor (depression-anxiety) models. RESULTS: The models fitted well with the different data sets. The depression-anxiety model was marginally superior to the one factor model as judged by various statistical measures of fit. The two factors in depression-anxiety model were, however, highly correlated. CONCLUSIONS: The findings suggest that symptoms of emotional distress seem to have the same factor structure across cultures.

Annotation: correspondence analysis.

Correspondence analysis is a useful technique in the investigation of relationships between categorical variables. By displaying these relationships graphically, correspondence analysis solutions may allow investigators to obtain insights into the structure of their data more readily than would be obtained by simply studying tables of numerical values and examining associated chi-squared statistics. Some experience of using correspondence analysis solutions in practice is generally necessary to appreciate the method's usefulness fully. An interesting application of the method in a longitudinal study of cognitive development is given in Lautrey and Cibois (1992), and some suggestions as to how best to combine correspondence analysis with a more formal method such as log-linear modelling are given in Van der Heyden and de Leeuw (1985).

An introduction to finite mixture distributions.

Finite mixture densities can be used to model data from populations known or suspected to contain a number of separate subpopulations. Most commonly used are mixture densities with Gaussian (univariate or multivariate) components, but mixtures with other types of component are also increasingly used to model, for example, survival times. This paper gives a general introduction to the topic which should help when considering the other more specialized papers in this issue.

Contradictory correlations between derived scales.

In a clinical trial one scale of pain relief is scored backwards relative to another (high on one corresponding to low on the other), with a consequent large negative correlation. But two derived scales of total pain, obtained by multiplying average pain relief on each scale by duration of pain (common to both pain relief measurements) gave an almost zero correlation. This apparent contradiction is explained by the inverse relationship between the pain relief scales and the large differences in duration of pain experienced by the patients.

Scales for rating motor impairment in Parkinson's disease: studies of reliability and convergent validity.

Study 1 examined the reliability of the ratings assigned to the performance of five sign-and-symptom items drawn from tests of motor impairment in Parkinson's disease. Patients with Parkinson's disease of varying severity performed gait, rising from chair, and hand function items. Video recordings of these performances were rated by a large sample of experienced and inexperienced neurologists and by psychology undergraduates, using a four point scale. Inter-rater reliability was moderately high, being higher for gait than hand function items. Clinical experience proved to have no systematic effect on ratings or their reliability. The idiosyncrasy of particular performances was a major source of unreliable ratings. Study 2 examined the intercorrelation of several standard rating scales, comprised of sign-and-symptom items as well as activities of daily living. The correlation between scales was high, ranging from 0.70 to 0.83, despite considerable differences in item composition. Inter-item correlations showed that the internal cohesion of the tests was high, especially for the self-care scale. Regression analysis showed that the relationship between the scales could be efficiently captured by a small selection of test items, allowing the construction of a much briefer test.

The characteristics of delusions: a cluster analysis of deluded subjects.

Using a group of 55 psychotic subjects, 11 belief characteristics of delusions were recently assessed. The present study describes a cluster analysis of these data to investigate whether subjects showing characteristic patterns of delusional experience form relatively distinct groups. Three sub-groups were generated, each of which was given a descriptive label: hebephrenic, paranoid and depressed. The characteristics of these groups are described, in terms of associations with measures of mental state and belief content.

A Monte Carlo Investigation of the Likelihood Ratio Test for Number of Classes in Latent Class Analysis.

When using latent class analysis to explore multivariate categorical data an important question is -- how many classes are appropriate for this data? An obvious candidate to answer the question is the likelihood ratio test of c[SUB0] against c[SUB1] classes. In this paper this test is investigated by Monte Carlo methods; results confirm that the usually assumed null distribution is inappropriate.

A Note on Parameter Estimation for Lazarsfeld's Latent Class Model using the EM Algorithm.

Latent class analysis is formulated as a problem of estimating parameters in a finite mixture distribution. The EM algorithm is used to find the maximum likelihood estimates, and the case of categorical variables with more than two categories is considered.

The McGill Pain Questionnaire: a replication of its construction.

The McGill Pain Questionnaire is in widespread use as a means of understanding the pain patient and monitoring treatment response. The current study consists of a replication of the construction of the questionnaire through the use of maximally dissimilar methodology and statistical techniques. The study comprised two stages: (a) an attempt to replicate the grouping of words within the questionnaire; and (b) an investigation of the intensity relationships of words within each subgroup. A direct grouping technique was used for stage (a), whereby 90 subjects sorted the words into semantically similar groups. A similarity matrix was constructed in terms of the number of times each word was associated with each of the other 78 words and subjected to cluster analysis. Inspection of the 20-group solution revealed considerable similarity with the original questionnaire. The intensity relationships were examined by asking a further group of 20 subjects to rate words on analogue scales. The results suggest a unidimensional solution to be inappropriate for a proportion of the subgroups. While there was a close resemblance with the MPQ, there was evidence for reducing the number of subgroups, as a 16-group solution offered a sensible and statistically parsimonious amalgamation. The implications of this work for the use of questionnaire methods are discussed.

Bimodality and the nature of depression.

Arguments concerning the nature of depressive disorders have involved as a central issue the question of the bimodality or otherwise of the distribution of some variable expressing variation in symptomatology. The implications of a particular type of frequency distribution along this dimension, whether uni- or bi-modal, have been misunderstood by a number of workers, and an attempt is made to clarify the situation.

A Monte Carlo Investigation Of The Likelihood Ratio Test For The Number Of Components In A Mixture Of Normal Distributions.

A likelihood ratio test to determine whether data arises from a single or a mixture of two normal distributions is investigated by Monte Carlo methods. The results show that the proposed sampling distribution of the test appears to be appropriate only for sample sizes above fifty, and for data where the sample size is ten times the number of variables. For such cases the power of the test is considered and found to be fairly low unless the generalized distance between the components is greater than 2.0.

Interactions in contingency tables: a brief discussion of alternative definitions.

The purpose of this note is to indicate how the disagreement between Tennant & Bebbington and Brown & Harris has arisen. The difference between the 2 pairs of authors is due, at least in part, to their use of different models for the analysis of the data in question. There can be no final answer as to which model is correct - the data are simply open to more than one interpretation. Nevertheless, it is interesting that the use of a multiplicative model, seemingly ignored by Brown & Harris, leads to a simple description of the data in which the 'vulnerability factor' and the 'provoking agent' may be considered to act independently on the response. Consequently, in stating that their data contain an interaction so obvious that it can be detected by 'visual inspection', Brown & Harris may have been somewhat rash.