Analyzing and Presenting Additive and Multiplicative Interaction Effects Between Categorical Variables in Social Work Research

Background and Purpose

The concept of two-way interaction focuses on the idea that the relationship between a predictor variable (X) and the outcome variable (Y) differs depending on the level of another predictor variable (M) controlling for covariates. One common approach to investigate two-way interaction effect is to add a multiplicative term between two predictor variables (X×M) in the regression model. While we can easily obtain the effect size of the multiplicative interaction term, the significance of the relationship between X and Y for non-reference levels of M is still unknown. To address this issue, numerous studies in social work research estimated a regression model for each subgroup sample stratified by the levels of M. However, this approach is problematic and unnecessary, as it assumes that the relationship between X and Y differs not only by the level of M but also by the levels of all covariates. In other words, the effect of covariates were not fully controlled for in such stratified analyses, leading to a loss of statistical power and a result deviant from what the original hypotheses intend to address. In addition, the additive interaction effect, despite its advantages in measuring changes of outcome frequencies, remained unexplored in social work research due to extra computations.

The purpose of this workshop is to teach attendants to systematically understand, analyze, interpret and present interaction effects between categorical variables in social work research. We will first discuss some basic concepts of interaction effects. We will then review the some common approaches in addressing interaction effects in the past and pinpoint their limitations. We will explain additive and multiplicative interaction effects and compare their strengths and context of use. In addition, we will introduce a four-step approach to present analyses of effect modification and interaction in research articles. Finally, we will demonstrate the relevant analytical procedures using Stata.

Topics    

• Overview of effect moderation and interaction.
• Review of past practice in presenting interaction effects
• Additive vs. multiplicative interaction
• Four steps for presenting effect modification and interaction
• Demonstration of additive and multiplicative interaction analyses using Stata.

Career Level and Prerequisites

Participants are expected to have knowledge in regression analysis. Experience with the Stata software is recommended, but not required. At the end of the workshop, participants will have a more in-depth knowledge in understanding interaction effects between categorical variables and presenting effect size and significance rigorously in social science research.  Hands-on learning materials will be available at the workshop.

Methods and Approach

This hands-on workshop will be conducted in an interactive manner between the attendants and the presenters. Powerpoint presentation and data analysis demonstration will be used with interpretations and discussions. Handouts will be provided. Attendants are welcome to ask questions during the session.

Recommended References

Jewell, N. P. (2003). Statistics for epidemiology. CRC Press.

Knol, M. J., & VanderWeele, T. J. (2012). Recommendations for presenting analyses of effect modification and interaction. International journal of epidemiology, 41(2), 514-520.

VanderWeele, T. J., & Knol, M. J. (2014). A tutorial on interaction.Epidemiologic Methods, 3(1), 33-72.