The reliability and validity of the thin slice technique: Observational research on video recorded medical interactions.

摘要

Introduction: Observational research using the thin slice technique has been routinely incorporated in observational research methods, however there is limited evidence supporting use of this technique compared to full interaction coding. The purpose of this study was to determine if this technique could be reliability coded, if ratings are consistent between the first, second and third slice, and if they are indeed representative of full interactions.;Methods: Three 30-second thin slices were sampled from the beginning, middle and end of a full-length video-recorded patient/physician interaction collected a part of a larger research study in a low income urban primary care clinic. Thin slice excerpts and full interactions were rated on five dimensions (liking, attention, coordination, trust and rapport) using a nine point Likert scale ranging from 'none' to 'high' by eight independent coders. Reliability was assessed using the intraclass correlation measure, validity of thin slices was assessed using the Friedman test (non-parametric equivalent of the Repeated measures ANOVA), and the comparison of thin slice coding to full interaction coding was assessed using the Wilcoxon Sign Ranks test (nonparametric version of the Paired t-test).;Results: Thin slice reliability on Likert scale items ranged from .762-.910 with an average IRR of .850. Friedman tests conducted on all five variables (liking, attention, coordination, trust and rapport) comparing the rating of the three slices of the interaction were non-significant. Results of the Wilcoxon Signed Ranks test indicated there was a significant difference between the composite thin slice rating (average across three slices) and the full interaction ratings with full interaction variables rated consistently higher than their respective thin slice composite.;Conclusion: Results indicate that thin slices can be reliability coded by independent coders with a high degree of agreement across coders. Observational ratings across thin slices sampled at the beginning middle and end of an interaction were not significantly different demonstrating convergent validity. However, there was a significant difference between ratings obtained from thin slices and ratings obtained from the interaction as a whole, indicating care should be taken when thin slices are used to represent the interaction as a whole.