Quantitative Brown Bag Series
Location:  Young Hall 166 (Unless specified otherwise)
Thursdays from 1:35 PM  2:35 PM (Unless specified otherwise) 
Calendar Administrator:
ACADEMIC YEAR: 2014  2015 Print Page  

Fall 2014  
10/02/2014 
Bayesian versus Frequentist Estimation of MultitratMultimethod Confirmatory Factor Models
SPEAKER: Dr. Jonathan Helm Campbell and Fiske's (1959) separation of trait, method, and unique variance across a set of multitraitmultimethod (MTMM) manifest variables directly translates to a confirmatory factor model, and several reports support this approach for partitioning variance (Cole, 1987; Widaman, 1985; Schmitt & Stults, 1986). However, researchers selecting this approach often encounter estimation problems (i.e., failed convergence or solutions with outofbounds estimates; Widaman, 1985). Mathematical investigations have identified several potential sources of these problems (Kenny & Kashy, 1992; Grayson & Marsh, 1994), forcing applied researchers to face an analytic conundrum when performing MTMM data analysis. The advent of Bayesian estimation for structural models offers many new opportunities, including the ability to fit models that would fail to converge when estimated within a frequentist framework (Scheines, Hoijtink, & Boomsma, 1999; Asparouhov & Muthén, 2010). Based on the nonidentification problems that typically arise when fitting the CTCM model to MTMM data (Kenny & Kashy, 1992; Grayson & Marsh, 1994), and extra modeling flexibility provided by Bayesian estimation, the current paper examines the differences between maximumlikelihood (ML) and Bayesian estimation of the CTCM model. *Prior to Dr. Helm's presentation Dr. Emilio Ferrer will be introducing our area's new graduate students. He'll also be giving updates on the area search for a new faculty member, with emphasis on opportunities for us to attend prospective candidates' job talks later this quarter and/or early winter. The meeting will conclude with asking for volunteers to present at fall brown bag meetings. 

10/09/2014 
Modeling Timevarying Interdependence
SPEAKER: Dr. Jonathan Helm Interdependence between two individuals may be estimated using a variety of statistical techniques, but these models typically assume that dependence remains constant across repeated measures. To the extent that theory predicts different patterns of dependence as a function of time, current analytic approaches may not adequately test relevant hypotheses. This dissertation focuses on the development of a novel method for analyzing changes in dependence, as it unfolds over time for a sample of dyads. The first chapter gives detail of a specific theory that may benefit from the new method, explains why several common methods cannot investigate change in dependence, and outlines criteria for an approach to summarize change in dependence appropriately. The second chapter introduces an approach that satisfies the criteria, describes the technique analytically, and tests its mathematical properties via simulation. The third chapter describes an extension of the method that accounts for measurement error, and, via simulation, summarizes situations when the extension outperforms the simpler version. The fourth chapter provides an application of the method to empirical data to inform the theory outlined in the first chapter. The fifth chapter summarizes the benefits gained from the method, describes the limitations and assumptions, and suggests future steps for further innovation to the proposed method. 

10/16/2014 
Fit Index Sensitivity to Restricted Factor Analysis Model Misspecification
SPEAKER: Joseph E. Gonzales While there is contention over the use of fit indices (Barret, 2007; Bentler, 2007), they are often used to assess whether models are tenable representations of data, often utilizing rules of thumb (Hu & Bentler, 1999) that, despite being overgeneralized (Marsh, Hau, & Wen, 2004), have been widely adopted. Previous work has shown that interindividual models of individual data are relatively insensitive to heterogeneity in intraindividual factor structures with respect to factor loadings (Molenaar, 2004). In the present study I expand on this finding using simulated data to evaluate if fit indices are sensitive to heterogeneity in factor structure when cases generated using either a one or twofactor model are randomly mixed and fit with a one or twofactor CFA model. Results suggest that AIC and BIC were able to discriminate between one and twofactor models with as little as ~10% heterogeneity (9:1 onefactor to twofactor cases), but CFI, TLI, and RMSEA all failed to reject the onefactor model until heterogeneity reached ~20%. SRMR performed the worse; failing to reject the onefactor model until ~55% heterogeneity was present. The most efficient measure of fit was the chisquare, which rejected the onefactor model with the smallest proportion of heterogeneity considered (5%). Note that the twofactor model was not closely evaluated for rejection, as its fit to data was consistently excellent. In the case of the twofactor model, inspection of the correlation between the two factors could be used to justify appropriateness of a onefactor model. However, with greater heterogeneity of one to twofactor model generated data, estimated correlations will be reduced depending on the strength of the correlation in the generating twofactor model. Consequently, inspection of the estimated factor correlations, an obvious indicator for the onefactor model, would likely be less informative, or possibly masked depending on the amount of heterogeneity in the sample. 

10/23/2014 
Brown Bag Meeting Cancelled
No Quantitative Brown Bag Today 

10/30/2014 
The madgenius paradox: Can creative people be more mentally healthy but highly creative people more mentally ill?
SPEAKER: Dr. Dean K. Simonton The persistent madgenius controversy concerns whether creativity and psychopathology are positively or negatively correlated. Remarkably, the answer can be "both"! The debate has unfortunately overlooked the fact that the creativitypsychopathology correlation can be expressed as two independent propositions: (a) among all creative individuals, the most creative are at higher risk for mental illness than are the less creative and (b) among all people, creative individuals exhibit better mental health than do noncreative individuals. In both propositions, creativity is defined by the production of one or more creative products that contribute to an established domain of achievement. Yet when the typical crosssectional distribution of creative productivity is taken into account, these two statements can both be true. This potential compatibility is here christened the madgenius paradox. This paradox can follow logically from the assumption that the distribution of creative productivity is approximated by an inverse power function called Lotka's Law. Even if psychopathology is specified to correlate positively with creative productivity, creators as a whole can still display appreciably less psychopathology than in the general population because the more at risk creative geniuses represent an extremely tiny proportion of those contributing to the domain. The hypothesized paradox has important scientific ramifications. 

11/06/2014 
TBA
SPEAKER: Dr. Paul Hastings & Jonas Miller TBA 

11/13/2014 
TBA


11/20/2014 
TBA
SPEAKER: Nathan Smith TBA 

11/27/2014 
Thanksgiving Holiday
Agenda No brown bag this week. Happy Thanksgiving and safe travels. 

12/04/2014 
TBA
SPEAKER: Melissa McTernan 

12/11/2014 
TBA
SPEAKER: Marilu Isiordia TBA 

12/18/2014 
Finals Week
Agenda No brown bag this week. Good luck with finals, whether taking or grading them. 
