Measures of success: characterizing teaching and teaching change with segmented and holistic observation data

Abstract

Background: Numerous studies show that active and engaging classrooms help students learn and persist in college, but adoption of new teaching practices has been slow. Professional development programs encourage instructors to implement new teaching methods and change the status quo in STEM undergraduate teaching, and structured observations of classrooms can be used in multiple ways to describe and assess this instruction. We addressed the challenge of measuring instructional change with observational protocols, data that often do not lend themselves easily to statistical comparisons. Challenges using observational data in comparative research designs include lack of descriptive utility for holistic measures and problems related to construct representation, non-normal distributions and Type-I error inflation for segmented measures. Results: We grouped 790 mathematics classes from 74 instructors using Latent Profile Analysis (a statistical clustering technique) and found four reliable categories of classes. Based on this grouping we proposed a simple proportional measure we called Proportion Non-Didactic Lecture (PND). The measure aggregated the proportions of interactive to lecture classes for each instructor. We tested the PND and a measure derived from the Reformed Teaching Observation Protocol (RTOP) with data from a professional development study. The PND worked in simple hypothesis tests but lacked some statistical power due to possible ceiling effects. However, the PND provided effective descriptions of changes in instructional approaches from pre to post. In tandem with examining the proportional measure, we also examined the RTOP-Sum, an existing outcome measure used in comparison studies. The measure is based on the aggregated items in a holistic observational protocol. As an aggregate measure we found it to be highly reliable, correlated highly with the PND, and had more statistical power than the PND. However, the RTOP measure did not provide the thick descriptions of teaching afforded by the PND. Conclusions: Findings suggest that useful dependent measures can be derived from both segmented and holistic observational measures. Both have strengths and weaknesses: measures from segmented data are best at describing changes in teaching, while measures derived from the RTOP have more statistical power. Determining the validity of these measures is important for future use of observational data in comparative studies.