In this article, the authors illustrate how random coefficient modeling can be used to develop growth models for the analysis of longitudinal data. In contrast to previous discussions of random coefficient models, this article provides step-by-step guidance using a model comparison framework. By approaching the modeling this way, the authors are able to build off a regression foundation and progressively estimate and evaluate more complex models. In the model comparison framework, the article illustrates the value of using likelihood tests to contrast alternative models (rather than the typical reliance on tests of significance involving individual parameters), and it provides code in the open-source language R to allow readers to replicate the results. The article concludes with practical guidelines for estimating growth models.
The purpose of the research was to assess the diagnostic efficiency of the Primary Care Posttraumatic Stress Disorder Screen (PC-PTSD) and the Posttraumatic Stress Disorder Checklist (PCL) as clinical screening tools for active duty soldiers recently returned from a combat deployment. A secondary goal was to examine the item-level characteristics of both the PC-PTSD and the PCL. A validation study conducted with a sample of 352 service members showed that both the PC-PTSD and PCL had good diagnostic efficiency. The overall diagnostic efficiency assessed by the area under the curve (AUC) was virtually the same for both the PC-PTSD and PCL. The most efficient cutoff values for the PC-PTSD were either 2 or 3 "yes" responses with the latter favoring specificity. For the PCL, the most efficient cutoff values were between 30 and 34, mirroring recommended PCL cutoff values from some studies in primary care settings. The examination of item characteristics suggested a 4-item PCL with an AUC virtually identical to that of the full PCL. Item analyses also identified that the most discriminate item in both scales pertained to symptoms of avoidance. Implications and limitations are discussed.
This study built on previous exploratory research (S. M. Jex & D. M. Gudanowski, 1992) that examined both self-efficacy and collective efficacy as moderators of stressor-strain relations. Based on survey data collected from 2,273 U.S. Army soldiers representing 36 companies, it was found that both self- and collective efficacy moderated the relationship between stressors and strains. Multilevel random coefficient model results revealed that respondents with strong self-efficacy reacted less negatively in terms of psychological and physical strain to long work hours and work overload than did those reporting low levels of efficacy. In addition, respondents with high levels of self-efficacy responded more positively in terms of job satisfaction to tasks with high significance than did those with low efficacy. The results also revealed that group-level collective efficacy moderated the relationship between work overload and job satisfaction and between task significance and organizational commitment. Limitations of the study and implications of these findings are discussed.
In the organizational literature, the impact of group size on the magnitude of the group-level correlation has not been explicitly delineated, despite the fact that group sizes vary considerably in organizational research. This article discusses the relationship between group size, ICC(J) values, and the magnitude of the group-level correlation, and shows that group size and ICC(I) values are important because they influence the reliability of the aggregate variables. Based on this discussion, a correction for attenuation formula is proposed that permits one to estimate the magnitude of the actual group-level correlation corrected for the reliability of the aggregate variables. A simulation study demonstrates that the correction for attenuation formula provides accurate estimates of the actual group-level correlation under a wide range of conditions. Implications for multilevel analyses are discussed.
Scholars have been interested in the extent to which organizational phenomena generalize across levels of analysis for quite some time. However, theoretical frameworks for developing homologous multilevel theories (i.e., theories involving parallel relationships between parallel constructs at different levels of analysis) have yet to be developed, and current analytical tools for testing such theories and models are limited and inflexible. In this article, the authors first propose a typology of multilevel theories of homology that considers different stages of theory development and different levels of similarity in relationships across levels. Building on cross-validation principles, the authors then delineate and demonstrate a comprehensive and flexible statistical procedure for testing different multilevel theories of homology. Finally, the authors discuss implications for theory, research, and practice, as well as potential caveats of the new statistical tests.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.