A comparison of linear and mixture models for discriminant analysis under nonnormality

Methods for discriminant analysis were compared with respect to classification accuracy under nonnormality through Monte Carlo simulation. The methods compared were linear discriminant analyses based both on raw scores and on ranks; linear logistic discrimination; and mixture discriminant analysis. Linear discriminant analysis and linear logistic discrimination were suboptimal in a number of scenarios with skewed predictors. Linear discriminant analysis based on ranks yielded the highest rates of classification accuracy in only a limited number of situations and did not produce a practically important advantage over competing methods. Mixture discriminant analysis, with a relatively small number of components in each group, attained relatively high rates of classification accuracy and was most useful for conditions in which skewed predictors had relatively small values of kurtosis.     

The Discrimination of Patient-Generated and Randomly Generated MMPIs

A discriminant analysis was applied to 34 randomly generated and 34 patient-generated Minnesota Multiphasic Personality inventory (MMPI) profiles. The discriminant was able to differentiate the two sets at greater than 97% accuracy. A cross validation based on derived unstandardized canonical discriminant functions was carried out on an additional 10 random and l0 patient MMPIs, It attained 95% accuracy. Eight experienced clinical psychologists attained tess than 70% accuracy–six of the eight did no better than chance. The importance of including actuarial data in validity judgments was discussed.     

The discrimination of patient-generated and randomly generated MMPIs

A discriminant analysis was applied to 34 randomly generated and 34 patient-generated Minnesota Multiphasic Personality Inventory (MMPI) profiles. The discriminant was able to differentiate the two sets at greater than 97% accuracy. A cross validation based on derived unstandardized canonical discriminant functions was carried out on an additional 10 random and 10 patient MMPIs. It attained 95% accuracy. Eight experienced clinical psychologists attained less than 70% accuracy--six of the eight did no better than chance. The importance of including actuarial data in validity judgments was discussed.     

Matching-to-sample and oddity-from-sample in goldfish.

Acquisition of three-alternative simultaneous matching-to-sample and oddity-from-sample was investigated. Five goldfish were trained on matching and five on oddity for a minimum of 70 days. Subsequently, six of the fish were trained for 70 days on the other task. Acquisition was similar for oddity and matching. Correct responding started at about chance level and slowly increased to about 75%, with some animals performing at levels of over 85%. Acquisition of oddity following matching and matching following oddity began below chance. Maximal level of performance on second-task oddity was comparable to that on first-task matching. By contrast, the maximal levels of performance when matching was the second task were not as high as that of the same subjects at the end of first-task oddity. All fish exhibited strong color preferences during matching acquisition but not during oddity acquisition. The data demonstrated that goldfish can acquire a discrimination in which the stimulus associated with reinforcement depends on the identity of a second stimulus.     

Pigeons may not remember the stimuli that reinforced their recent behavior

In two experiments the conditioned reinforcing and delayed discriminative stimulus functions of stimuli that signal delays to reinforcement were studied. Pigeons' pecks to a center key produced delayed-matching-to-sample trials according to a variable-interval 60-s (or 30-s in 1 pigeon) schedule (Experiment 1) or a multiple variable-interval 20-s variable-interval 120-s schedule (Experiment 2). The trials consisted of a 2-s illumination of one of two sample key colors followed by delays ranging across phases from 0.1 to 27.0 s followed in turn by the presentation of matching and nonmatching comparison stimuli on the side keys. Pecks to the key color that matched the sample were reinforced with 4-s access to grain. Under some conditions of Experiment 1, pecks to nonmatching comparison stimuli produced a 4-s blackout and the start of the next interval. Under other conditions of Experiment 1 and each condition of Experiment 2, pecks to nonmatching stimuli had no effect and trials ended only when pigeons pecked the other, matching stimulus and received food. The functions relating pretrial response rates to delays differed markedly from those relating matching-to-sample accuracy to delays. Specifically, response rates remained relatively high until the longest delays (15.0 to 27.0 s) were arranged, at which point they fell to low levels. Matching accuracy was high at short delays, but fell to chance at delays between 3.0 and 9.0 s. In Experiment 2, both matching accuracy and response rates remained high over a wider range of delays in the variable-interval 120-s component relative to the variable-interval 20-s component. The difference in matching accuracy between the components was not due to an increased tendency in the variable-interval 20-s component toward proactive interference following short intervals. Thus, under these experimental conditions the conditioned reinforcing and the delayed discriminative functions of the sample stimulus depended on the same variables (delay and variable-interval value), but were nevertheless dissociated.     

Function of intertrial interval in matching-to-sample

Twelve pigeons were trained on matching-to-sample using either a 0-, 5-, 15-, 25-, or 60-sec intertrial interval. Eight of these 12 pigeons were given one of the following intertrial interval changes: 0 to 60, 0 to 5, 5 to 0, 60 to 0, 15 to 25, 5 to 15, 60 to 5, 5 to 1, 1 to 5, 1 to 25, and 25 to 1 sec. Most intertrial interval changes were repeated at least once. The 0-sec intertrial interval subjects failed to match beyond chance levels, while other intertrial interval values resulted in matching acquisition. Changes from 0 sec to other intertrial interval values increased and changes to 0 sec decreased matching performance. Changes to intertrial interval values other than 0 sec resulted in little change in matching accuracy once stable performance had been attained.     

Multiple predictors of hypnotic susceptibility

In this article, we report two experiments in which various measures thought to be related to hypnotizability were analyzed by stepwise discriminant analysis techniques. Absorption (Tellegen, 1981, 1982; Tellegen & Atkinson, 1974) and preference for an imagic style of thinking (Isaacs, 1982) were robust predictors of hypnotizability; each variable accounted for significant variance in hypnotizability at their respective steps in two samples and correctly classified a significant proportion of low- and high-hypnotizable subjects in the discriminant analyses. The addition of two other variables in Experiment 2--a Sleep-Dream score derived from Evans's (1977) Cognitive Control of Sleep Mentation subscale and Gibson's (1985) Dream Questionnaire, and the Belief in the Supernatural subscale of the Taft (1969) Experience Questionnaire--increased the correct classification of the medium-hypnotizable subjects from chance levels to 74%. It is argued that a confirmatory and hierarchical approach is needed in future studies to explore correlates of hypnotizability more fully.     

Occupational attainment as a function of abilities and interests: a longitudinal analysis using project TALENT data

Ability, interest, gender, and family socioeconomic status of 13,248 tenth-grade participants in Project TALENT were studied as they relate to occupational attainment by using discriminant analysis. Individuals were classified into 12 broad categories reported 11 years after graduation. Accuracy analyses indicated correct classifications significantly above chance for all except the Technical and Sales categories. Within-category classification percentages were higher for all groups except Construction. Five canonical discriminant functions that jointly accounted for 96.8% of the between-groups variance were interpreted. The first 2 accounted for 81.9% of the variance. Function 1 was a general ability function; Function 2 differentiated the categories on the basis of mathematics ability and gender. Functions 3 through 5 accounted for 14.9% of the between-groups variance. Some implications of these findings are discussed.     

Transfer of matching performance in pigeons.

Three pigeons were given extensive training on three-key simultaneous matching problems using geometric-form and hue stimuli. After acquisition of matching, the birds were tested with pairs of stimuli involving one or both novel members. Matching during the test stimuli occurred less often than during the later stages of the acquisition phase, but more often than would occur if no transfer had taken place. Greater positive transfer was observed for problems that involved one, rather than two, novel stimuli. In the second phase of the experiment, previously trained birds were shifted to problems that required symbolic matching, i.e., the pigeons had to associate a particular center-key stimulus with a particular side-key stimulus. On each trial, one of two simuli was presented on the center key, and two other stimuli, different from those used on the center key, were displayed on the side keys. When the problem shift was introduced, correct responding was impaired, but remained considerably above chance level and quickly recovered in following sessions. The results were interpreted as favoring a stimulus-response-chaining account of matching behavior.     

A brief neuropsychological screening battery to detect brain damage in a psychiatric population

Seventy-five patients admitted to the psychiatric unit of a general hospital were administered a brief neuropsychological screening battery. Discharge diagnoses obtained from the medical records revealed that 20% had brainbased pathology. A discriminant function analysis resulted in a five-variable model (Symbol-Digit Modalities Test, educational level, Psychiatric-Organic subscale of the MMPI, and Shipley Vocabulary and Abstraction scores) which correctly classified 84% of the subjects, with 87% correct identification of the brain-damaged group. A jackknife cross-validation procedure resulted in comparable hit rates.     

Effect of sample presentation time on long-delay matching in the pigeon

The effect of sample stimulus presentation time on long-delay matching in highly practiced pigeons was investigated. The birds were found capable of above chance matching performance at a delay of 60 sec provided the sample stimulus was presented for 4 sec or longer. Matching accuracy increased as a negatively accelerated function of sample stimulus presentation time and decreased as a negatively accelerated function of time since the termination of the sample. The rate of forgetting was found to be independent of sample stimulus presentation time. The data were inconsistent with a temporal discrimination interpretation of the effect of presentation time on delayed matching. The data were interpreted as supporting a simple trace strength and decay model of pigeon delayed matching.     

Predicting suicide using the Rorschach Inkblot Test

Three hundred and seventy-five Rorschach protocols were scored blind and then divided into three groups: psychiatric controls, suicide attempts, and suicide effectors. Using the stepwise procedure of multiple discriminant analysis a trio of formulas, composed of six weighted variables each, and a constant, were constructed to apply to each of the three groups based on a test sample of 100 subjects in each group. Cross validation results on a new sample of 25 subjects in each group predicted classification with 52% overall accuracy at step six (chance .33). Internal reliability tests showed all values significant beyond .001. Discussion integrates the test data into the literature on suicide, and attends to the problem of overlap of predictability which occurs with most prediction scales.     

Tool selectivity in a non-primate,the New Caledonian crow (Corvus moneduloides)

We present an experiment showing that New Caledonian crows are able to choose tools of the appropriate size for a novel task, without trial-and-error learning. This species is almost unique amongst all animal species (together with a few primates) in the degree of use and manufacture of polymorphic tools in the wild. However, until now, the flexibility of their tool use has not been tested. Flexibility, including the ability to select an appropriate tool for a task, is considered to be a hallmark of complex cognitive adaptations for tool use. In experiment 1, we tested the ability of two captive birds (one male, one female), to select a stick (from a range of lengths provided) matching the distance to food placed in a horizontal transparent pipe. Both birds chose tools matching the distance to their target significantly more often than would be expected by chance. In experiment 2, we used a similar task, but with the tools placed out of sight of the food pipe, such that the birds had to remember the distance of the food before selecting a tool. The task was completed only by the male, who chose a tool of sufficient length significantly more often than chance but did not show a preference for a matching length. Electronic Publication     

Patterns of comprehension performance in agrammatic Broca's aphasia: a test of the Trace Deletion Hypothesis

We tested the core prediction of the Trace Deletion Hypothesis (TDH) of agrammatic Broca's aphasia, which contends that such patients' comprehension performance is normal for active reversible sentences but at chance level for passive reversible sentences. We analyzed the comprehension performance of 38 Italian Broca's aphasics with verified damage to Broca's area, who completed sentence-to-picture matching tasks using active and passive reversible sentences as stimuli. The results failed to confirm the predictions made by TDH: only a small minority (15%) performed at chance on passive sentences and better than chance on active sentences. Furthermore, the distribution of the 38 subjects' performance on passive sentences differed from that predicted by the TDH since many more subjects performed at better-than-chance levels than expected. We discuss the implication of these results for claims about the distribution of language processing mechanisms in the brain.     

A Cluster of Suicides at a London Psychiatric Unit

In an epidemiologically based study, a spate of 14 suicides among current patients of a London psychiatric unit was investigated. Statistical analysis showed it to be a discrete cluster of suicides, rather than a chance occurrence. There was no evidence of direct linkage or “contagion” between the suicides. Thirteen of the patients suffered from severe, chronic mental illness and all but 2 had been known to the psychiatric unit for at least a year. Twelve used violent methods, in 8 cases jumping from a high place. Although no definite cause for the cluster could be established, it coincided with a period of uncertainty concerning the future of the hospital and with changes in and absenteeism of senior medical staff.     

The Influence of Child-Directed Speech on Word Learning and Comprehension

This paper describes an investigation into the function of child-directed speech (CDS) across development. In the first experiment, 10–21-month-olds were presented with familiar words in CDS and trained on novel words in CDS or adult-directed speech (ADS). All children preferred the matching display for familiar words. However, only older toddlers in the CDS condition preferred the matching display for novel words. In Experiment 2, children 3–6 years of age were presented with a sentence comprehension task in CDS or ADS. Older children performed better overall than younger children with 5- and 6-year-olds performing above chance regardless of speech condition, while 3- and 4-year-olds only performed above chance when the sentences were presented in CDS. These findings provide support for the theory that CDS is most effective at the beginning of acquisition for particular constructions (e.g. vocabulary acquisition, syntactic comprehension) rather than at a particular age or for a particular task.     

班级朋友对抑郁的影响:一项社会网络分析

采用朋友提名和Beck抑郁问卷,以73个班级3034名大学生为被试,考察班级朋友网络中不同连接步数朋友的抑郁水平对个体抑郁的影响。结果发现:(1)互选朋友数量对抑郁存在显著的负向影响;(2)在控制了互选朋友数量的影响后,互选朋友中是否存在抑郁者,对个体自身的抑郁不具有显著的预测作用;(3)提名与被提名的直接朋友和两度距离朋友中,抑郁朋友数量对个体的抑郁具有显著的正向预测作用;(4)同性抑郁朋友的数量对个体的抑郁具有显著的正向预测作用,而异性抑郁朋友的数量对个体的抑郁则不具有显著的预测作用。综合上述结果可以得出以下结论:抑郁朋友在班级朋友网络中的影响范围仅限两度距离以内(包括两度),且这种影响不受朋友关系方向的限制。随着抑郁朋友的数量增多,个体抑郁的风险也显著增加,然而抑郁朋友对个体抑郁的影响,可能更多的是受到同性抑郁朋友的影响。     

Beliefs about surgery: Development and validation of an instrument to assess cognitive representations about surgery,in the context of a chronic illness

Objectives: Within a trial of medical and surgical treatments for gastro-esophageal reflux disease (GORD), involving randomised arms and preference arms, we tested the applicability of the Beliefs about Medicines Questionnaire (BMQ) and developed and tested the validity of a new Beliefs about Surgery Questionnaire (BSQ). Methods: Patients with GORD (N = 43) were interviewed to elicit their beliefs about medical and surgical treatments. These contributed to the development of BSQ items. The BMQ and BSQ were completed by trial participants at baseline (randomised trial: N = 325; preference trial: N = 414). Factor analysis and discriminant function analysis were used to assess validity. Results: Principal components analysis (PCA) largely replicated the four-factor BMQ structure. PCA of the combined BMQ/BSQ yielded six factors explaining 54.5% variance. BSQ items loaded onto distinct factors, demonstrating divergence from BMQ. As predicted, BMQ/BSQ scores enabled correct classification of 78.5% of participants to medication and surgery groups in the preference trial (χ²(6) = 205.9, p < 0.001) but only 54.5% (no better than chance) in the randomised trial (χ²(6) = 9.4, p = 0.154). Conclusions: The BSQ is a valid measure of perceptions about surgical treatments for GORD. With the BMQ, it provides information that may guide patients’ choices about treatment. This measure may be applicable to other conditions.     

Self-perception and the career aspirations of mathematically talented female adolescents

The role of self-perceptions of career-relevant abilities in understanding the career aspirations of mathematically talented adolescent females is studied. Female high school sophomores rated themselves on the 12 self-estimates of ability in Holland's Self-Directed Search. Based on her ideal career aspiration, each student was assigned to one of six career-aspiration groups. The results of the multiple discriminant function analysis indicated that nontraditional math career aspirants could be discriminated from the remaining five career-aspiration groups on the basis of seven self-estimates of ability. Further discriminations were less distinct although classification accuracy for all six career groups surpassed that expected by chance alone.     

Comparison of three multiscale inventories in identifying the presence of psychopathological symptoms

Two new multiscale inventories, the Basic Personality Inventory (BPI) and Millon Clinical Multiaxial Inventory (MCMI), assess the domain of psychopathology. Both use modern principles of test construction that arose over the years since the introduction of the MMPI. These three tests were compared in terms of their ability to discriminate between the presence and absence of 11 symptoms of psychopathology. Symptoms used in the analysis were extracted from medical records and showed interrater reliabilities ranging from kappa = .412 to 1.00. All symptoms had incidences of at least 15% in the sample of 107 psychiatric inpatients used. The MMPI was scored for both clinical and Wiggins content scales. Parallel discriminant analyses were performed for each set of test scores for the presence or absence of each symptom. A conservative alpha level was adopted, and results favored the Basic Personality Inventory.