Mediation analysis is a popular statistical method throughout psychology, communication, business and other behavioral sciences. However, previous research has demonstrated that that typical sample sizes are too small to have high power to detect indirect effects in between-subject designs; however, some inferential methods have higher power than others. Montoya and Hayes (2017) developed new methods for estimating and conducting inference on indirect effects in within-subject designs. For tests of means, within-subject designs boast power advantages over between-subject designs; however, this advantage has not been demonstrated for mediation analysis. Monte Carlo simulation is used to compare six inferential methods, two designs, and a broad range of sample sizes, effect sizes, and correlations among repeated measurements. The results suggest within-subject designs require about half the sample size of between-subject designs to detect indirect effects of the same size, and the effect of design is much greater than the effect of inferential method. Factors which can impact power (e.g. highly correlated measurements) are discussed, and how to determine if a within-subject design is appropriate for a given research question. MEMORE is an easy to use tool for estimating and conducting inference on indirect effects in two-condition within-subject designs.
Mediation analysis is commonly used in social-personality psychology to evaluate potential mechanisms of effects. With the recent replicability crisis, researchers are turning to power analysis to help plan studies; however, power analysis for mediation is not implemented in popular software (e.g., G*Power). Our symposium includes two presentations focusing on implementation of power analysis for mediation: (1) describing easy-to-use tools for implementing power analysis (e.g., pwr2ppl R package), and (2) evaluating whether different inferential methods result in similar recommended sample sizes and the role of assumption violations in these differences. Two presenters focus on study characteristics which can affect power: (1) use of the bias-corrected confidence interval and alternatives which better balance power and type I error, and (2) how measurement error on the mediator can impact power and how to correct this issue with latent variable models. Presentations will include applied examples, aimed at a social-personality audience, and provide concrete steps for increasing the validity and replicability of mediation analyses conducted in social-personality research. (Symposium Presented at SPSP 2021)
Machine learning methods are being increasingly adopted in psychological research. Lasso performs variable selection and regularization, and is particularly appealing to psychology researchers because of its connection to linear regression. Researchers conflate properties of linear regression with properties of lasso; however, we demonstrate that this is not the case for models with categorical predictors. Specifically, the coding strategy used for categorical predictors impacts lasso’s performance but not linear regression. Group lasso is an alternative to lasso for models with categorical predictors. We demonstrate the inconsistency of lasso and group lasso models using a real data set: lasso performs different variable selection and has different prediction accuracy depending on the coding strategy, and group lasso performs consistent variable selection but has different prediction accuracy. Additionally, group lasso may include many predictors when very few are needed, leading to overfitting. Using Monte Carlo simulation, we show that categorical variables with one group mean differing from all others (one dominant group) are more likely to be included in the model by group lasso than lasso, leading to overfitting. This effect is strongest when the mean difference is large and there are many categories. Researchers primarily focus on the similarity between linear regression and lasso, but pay little attention to their different properties. This project demonstrates that when using lasso and group lasso, the effect of coding strategies should be considered. We conclude with recommended solutions to this issue and future directions of exploration to improve implementation of machine learning approaches in psychological science.
Women’s underrepresentation in science, technology, engineering, and math (STEM) fields is well established; however, there is much variation in women’s among STEM fields. Women received 60% of bachelor’s degrees in biology but only 18% of computer science degrees in 2009. One explanation for this difference may be social stereotypes of the fields. Male-dominated STEM fields are stereotyped as asocial, which may lead women to think that these fields will not give them opportunities to work with and help others. Three studies examined how communal goals, goals related to creating and maintaining interpersonal relationships, relate to interest in STEM fields. Study 1 (N = 120) found that women were more likely than men to endorse communal goals. Additionally, women were more interested in STEM classes they thought would fulfill their communal goals. Participants’ communal goal endorsement was negatively correlated with their interest in male-dominated STEM fields, like computer science. Study 2 (N = 296) examined group work as a potential factor that could affect perceptions of communal fulfillment in science classes. Classes with group work were perceived to be higher in communal goal fulfillment. Male-dominated STEM fields were perceived to have less group work than female-dominated STEM fields. Additionally, preference for group work was positively correlated with communal goals. Study 3 (N = 91) experimentally manipulated the collaboration policy of a computer science class syllabus, testing for changes in communal goal fulfillment and interest. The class with a pro-collaboration policy was perceived as having the most communal goal fulfillment, but these perceptions did not carry over to perceptions of the field of computer science as a whole. These studies suggest that increasing opportunities for communal experiences in computer science may help women to feel that computer science would fulfill their goals, increasing their interest in the field.
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.
hi@scite.ai
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.