In psychological research, multilevel models are often used to analyze clustered data, where participants are clustered within groups or the same participants are repeatedly measured. The performance of multilevel models has been investigated in various simulation studies, most of which have assumed the equal size of clusters. However, in practice, often these clusters are unbalanced (unequal size). Previous research suggested that unbalanced cluster sizes could lead to low convergence rates (Van der Est et al., 2016) and inaccurate estimation of some parameters (Cools et al., 2009). However, these studies did not consider the effect of varying intraclass correlation coefficients (ICC) nor used a systematic method for controlling the degree of variation in cluster size. The purpose of the present study was to systematically examine the effects of unbalanced clusters as well as other factors on the performance of two-level models (with random intercept, random slope, and cross-level interaction; Arend & Schäfer, 2019).
For this purpose, a simulation study was conducted with total 108 conditions; three levels of ICC (0.1, 0.3, 0.5), three levels of the number of clusters (N = 50, 100, 200), three levels of mean cluster size (n = 5, 10, 20), and four levels of imbalance. For imbalance, negative binomial distribution was used to generate unbalanced cluster sizes and imbalance level was manipulated by changing the dispersion parameter of the cluster size distribution, cv (coefficient of variation; a higher value indicates a more variation in cluster size). Three levels of imbalance (cv = 0.5, 1, 1.5) as well as balanced data were considered.
Cho, Y. W. & Suk, H. W. (2019, October). Examining the Effects of Unbalanced Clusters on the Estimation Accuracy of Multilevel Modeling. Poster presented at the 2019 Fall Conference of Korean Society of Psychological Measurement and Assessment, Seoul, South Korea.
Abstract Perspective taking (PT) is often recommended as an effective way of resolving conflicts. Previous research has also shown some positive effects of PT in various social settings, such as reducing stereotypes and increasing prosocial behaviors. However, the effect of PT on direct conflicts might not be necessarily positive. Particularly, we expected the positive effects of PT would substantially diminish in conflicts over personally important issues since people tend to be biased and defensive regarding personally relevant matters. Thus, we conducted three studies to investigate 1) whether PT would have positive effects in situations involving conflicts and 2) how issue importance would moderate such effects. The results showed PT was generally associated with positive outcomes, even in conflict situations. As expected, however, such positive associations became substantially attenuated with the increased levels of issue importance. In addition, the feeling of broken reciprocity partially mediated the interaction between PT and issue importance.
Cho, Y. W. & Na, J. (under review: PSPB-19-560) The effects of perspective taking vary as a function of issue importance in conflict contexts. Personality and Social Psychology Bulletin.