The question of how to detect multivariate outliers has presented both philosophical and statistical problems. The method most widely used for the detection of multivariate outliers is Mahalanobis' D-Squared statistic (D$\sp2$), commonly viewed as analogous to a univariate standard score. D$\sp2$ is simple to calculate, and its asymptotic distribution is known to be the Chi-square distribution (Barnett, 1976, 1978a, 1978b, 1979; Beckman & Cook, 1983; Hawkins, 1974, 1980). Additionally, D$\sp2$ or D$\sp2$/df is available from major statistical packages such as BMDP and SPSSX. The distribution of D$\sp2$s calculated using the sample centroid and variance-covariance matrix is thought to be mathematically intractable (Barnett, 1984; Wilks, 1963). Some researchers have suggested the use of ordered Chi-square (Barnett, 1984; Beckman & Cook, 1983; Hawkins, 1980) or ordinary Chi-square (Comrey, 1984; Rasmussen, 1988; Tabachnick & Fidell, 1983) critical values for evalaution of D$\sp2$ in the detection of outliers. Tables of ordered Chi-square critical values are not available, thus, it was necessary to compute these values for the present study. This study examined the fit of D$\sp2$ with the Chi-square and ordered Chi-square distributions, via Monte Carlo Methods, and determined that neither provided accurate critical values. Consequently, critical values were generated empirically. The resulting tables of critical values cover the largest 25% of ordered D$\sp2$ for the conditions resulting from a full factorial cross of numbers of subjects (20, 30, 40, 50, 100, 200, 300, 500, 1000), and numbers of variables (2, 3, 4, 5, 6, 7, 8, 9, 10, 15, 20, 25, 30, 35, 50) / acase@tulane.edu

Sphericity is one of the assumptions underlying repeated measures analysis of variance. Box discovered a parameter called epsilon that reduces the degrees of freedom to adjust for violation of sphericity. Because the routine use of the epsilon-adjusted F-test does not guarantee the most efficient test for repeated-measures treatment effects, a preliminary test of the sphericity assumption appears to be prudent. After addressing the problems with the existing sphericity tests, the use of empirical critical values of Box's epsilon as criteria for testing sphericity is proposed. First, tables of empirical critical values of Box's epsilon were generated. Then, the Type I error rate of the proposed empirical epsilon test was examined. Finally, the relative power of the empirical epsilon test was compared with Mauchly's test, the most widely applied sphericity test, under normal population conditions with repeated measures designs / acase@tulane.edu

A covariance structure model consists of a measurement model, which is a collection of variables and their associated constructs or factors, and a structural model, which is a set of predictions about the causal relationships among the constructs. Data are collected on the variables and a covariance or correlation matrix is computed. The covariance structure modeling (CSM) procedure estimates the parameter values in the model that has been specified and assesses how well the model fits the data. Models failing to provide a good fit to the data are often modified. This process is known as a specification search The current Monte Carlo study investigated the effectiveness of the choice of one of five initial structural models, the method used to address the measurement model, and the search statistics used (Lagrange multiplier and expected parameter change) on the success rate of specification searches. 700 specification searches were conducted on 40 sample matrices (N = 300) generated from 4 population models. The effectiveness of the search statistics differed by the method used to address the measurement model with the expected parameter change having a higher success rate when the measurement model was assessed independently of the structural model and the Lagrange multiplier having a higher success rate when the measurement and structural models were assessed simultaneously. Successful model recovery did not differ significantly by the initial structural model used but the full initial model always performed equal to or superior to the other initial models. Successful model recovery differed markedly by population model with success rates ranging from 16% to 73%. The research concludes with a call for future research and a list of suggested research topics / acase@tulane.edu

A number of objective personality questionnaires have been published which aim to measure the six processes related to Acceptance and Commitment Therapy's model of treatment (acceptance, defusion, present moment awareness, self-as-context, values, and committed action). These measures operationally define these hypothesized processes in research settings. However, little research has been done to investigate whether these processes, as measured by these questionnaires, are differentiable from each other or from other, seemingly similar constructs such as distress tolerance and coping styles. Additionally, it is unclear whether these questionnaire measures have differing relationships with other potentially relevant constructs, such as psychopathology, functioning, and personality. The structure of these process measures was investigated across two participant samples. A multi-trait structure of ACT processes was found, with three higher order dimensions consisting of psychological inflexibility/cognitive fusion, mindfulness, and avoidance, as well as a number of distinguishable lower order traits. This structure was found across multiple samples, and measures of these factor analytically-derived traits were found to have incremental validity and to be distinguishable from other, superficially similar psychological processes. These results provide guidance for measurement selection and suggest future directions for scale development. Relevance to treatment outcome research is also discussed.

Test selection in psychological assessment is guided, both explicitly and implicitly, by how informative tests are with regard to a trait of interest. Most existing formulations of test information are sensitive to subpopulation variation, with the result that test information will vary from sample to sample. Recently, measures of test information have been developed that quantify the potential informativeness of the test. These indices are defined by the properties of the test, as distinct from the properties of the sample or examinee. As of yet, however, measures of potential information have been developed only for unidimensional tests. In practice, psychological tests are often multidimensional. Furthermore, multidimensional tests are often used to estimate one specific trait among many. This study develops measures of potential test information for multidimensional tests, as well as measures of marginal potential test information---test information with regard to one trait within a multidimensional test. In Study 1, the performance of the metrics was tested in data simulated from unidimensional, first-order multidimensional, second-order, and bifactor models. In Study 2, measures of marginal and multidimensional potential test information are applied to a set of neuropsychological data collected as part of Rush University's Memory and Aging Project. In simulated data, marginal and multidimensional potential test information were sensitive to the changing dimensionality of the test. In observed neuropsychological data, five traits were identified. Verbal abilities were most closely correlated with probable dementia. Both indices of marginal potential test information identify the Mini Mental Status Exam as the best measure of that trait. More broadly, greater marginal potential test information calculated with regard to verbal abilities was associated with greater criterion validity. These measures allow for the direct comparison of two multidimensional tests that assess the same trait, facilitating test selection and improving the precision and validity of psychological assessment.

Exercise rooted in changing one’s appearance is associated with increased disordered eating and body image pathology. There are a limited number of scales assessing appearance-based exercise, and those that do are methodologically flawed. The aim of the current work was to develop a psychometrically sound measure of appearance-based exercise (Exercise Appearance Motivations Scale (EAMS)). Female undergraduate students (N = 650) completed an online survey designed to assess the EAMS’ psychometric properties. Factor analysis and hierarchical regressions were used for measure development and validation. Five factors of the EAMS were identified through factor analysis: muscularity, appearance, societal pressures, shape/weight, and avoidance/shame. Pearson product moment correlations were used to examine the associations between the EAMS and scales assessing convergent validity (appearance comparison, disordered eating, appearance evaluation, internalization of body ideals) and discriminant validity (belief in a just world). Results indicated that Cronbach’s alpha (α = .94) and test-retest reliability coefficients (r = .77) were adequate. The EAMS demonstrated adequate construct and incremental validity. These results provide preliminary evidence that the EAMS scale is a reliable and valid measure of appearance-based motives of exercise behavior when used with undergraduate women. Implications, limitations, and future research ideas are discussed.

There is a recent rekindled interest concerning the stability, reliability, and predictive validity of skilled performance across repeated measurements. One common phenomenon resulting from the repeated measures of subjects' performance across trials during skills acquisition is the superdiagonal correlation matrix, also known as the simplex matrix, in which correlations of performance decrease as a function of the separation of trials (or as a function of time). The present study collected fifty such correlation matrices from both published and unpublished sources. Next, the standardized item coefficient alphas were calculated from correlations for all possible combinations of adjacent trials to identify rules for which trials should be used to maximize reliability. When early or late adjacent trials showing low correlations were dropped from the computation of the standardized item coefficient alpha, reliability sometimes increased, although not dramatically. The rows of correlations above the diagonal of a superdiagonal matrix were plotted across trials and it was found that the resulting graphs could be used in deciding which adjacent early and/or late trials to drop to maximize the reliability. Seven figures, representative of the different sorts of published matrices, provide graphical decision rules for determining which trials to average to maximize reliability. The standardized coefficient alpha for the entire matrix should also be computed as a benchmark reliability / acase@tulane.edu

Identification of interaction effects is of increasing importance to the social sciences; however, interaction (or moderator) effects have often been difficult to detect with continuous data. Structural equation modeling (SEM) methods have been touted as a solution to the problem of detecting moderators with continuous data because they are thought to account for the presence of measurement error. Also some of the optional fitting algorithms are thought to be less sensitive to non-normality, a common characteristic of the cross-product terms used in evaluation of interaction effects. Although much of the literature to date describes SEM methods to detect interactions among latent variables, the current study contrasts well known moderated multiple regression (MMR) as compared to various analogous SEM models for estimating moderation among manifest variables. While some SEM estimation methods were found inferior, no clear advantage of any SEM method over MMR was observed in the detection of interaction effects. Furthermore, SEM models, with stable Type I error rates, either failed to converge or reported errors about 10% of the time while MMR always yielded a solution / acase@tulane.edu

Effects of a computer assessor were compared to those of human assessors on two types of social desirability scales: Impression Management and Self-Deception. Effects of the independent variable on reactions to the assessment and assessor and on reactions to computers in general were also examined. One hundred and twenty undergraduate psychology students were assessed for social desirability response bias by either a computer or human assessor, following which their reactions were assessed by paper-pencil. / No differences were found as a function of Type of Assessor on either social desirability scale. No differences between groups were found in reactions to the assessment and assessor, although respondents in both groups were generally positive. / Examining reactions to computers in general demonstrated that respondents assessed by computer evidenced less computer aversion (p =.015) and preferred a computer assessor in the future more often (p $<$.0001) than did those in the human assessor group. This is consistent with previous research which indicated that subjects who had physically touched a computer had more positive attitudes towards computers. It suggests that when using a computer assessor, aversion to computers by respondents may be diminished by actual exposure to the assessor. / Exploratory analyses suggest that both human and computer assessors have aspects which may be important to respondents' comfort and should be considered in any assessment procedure. / Source: Dissertation Abstracts International, Volume: 52-08, Section: B, page: 4481. / Major Professor: Mark H. Licht. / Thesis (Ph.D.)--The Florida State University, 1991.

Abstract not available