There Is a World Outside of Experimental Designs: Using Twins to Investigate Causation

Abstract

This study introduces a co-twin control method commonly used in the medical literature but not often within educational research. This method allows for a comparison of twins discordant for an "exposure," approximating alternative outcomes in the counterfactual model. Example analyses use data drawn from the Florida Twin Project on Reading to determine whether exposure to "teacher quality," measured by growth in oral reading fluency (ORF) scores of classmates, causally affects ORF performance of twins in the subsequent years. The analysis highlights PROC MIXED in SAS, including a novel expansion to allow for the nested data. Results from 2,788 twins suggested that being in classrooms with lower teacher quality in first grade leads to lower ORF scores in second and third grade with little indication of possible genetic or environmental confounding.

Keywords: causation; classroom effects; methodology; oral reading fluency; twins.

Figure 1. Hypothetical scenarios representing potential results from the co-twin control design

Note. DZ = dizygotic; MZ = monozygotic. Scenario A represents a similar effect of exposure no what matter the relatedness of the individuals, suggesting a possible causal effect. The remaining scenarios represent noncausal situations. Scenario B represents genetic confounding, whereas Scenario C represents shared environmental confounding. Scenario D represents genetic and shared environmental confounding.

Figure 2. Results from the co-twin control analyses exploring the effect of teacher quality on reading performance

Note. DZ = dizygotic; MZ = monozygotic; ORF = oral reading fluency.