Objective: To determine whether the newly derived interclass and intraclass correlation coefficients (ICCs)would overstate or understate the results from 2 previously published studies, which used better known ICCs that assume nested factors, and to determine mean absolute differences of observers' measurements for 3 previous studies.
Study design: Retrospective analysis of data from 2 blind studies with repeated-measure design. Two newly derived ICCs, appropriate to situations with 3 random factors (patients, examiners, and occasions) that bear a crossed (as opposed to nested) interrelationship, were applied to data from an experiment with random crossed factors.
Main outcome measures: Observer reliability is determined with ICCs, 95% CIs, and observer error analysis (mean absolute differences of observers' measurements) for angles and distances derived from Harrison's modified Risser-Ferguson line-drawing method on anteroposterior (AP) lumbar and AP cervical radiographic views. Observer error analysis for angles and distances derived from Harrison's posterior tangent method on lateral cervical views was also determined.
Results: The majority of ICCs for reliability of line drawing on both AP cervical and AP lumbar radiographs were in the high range; 13 of 16 ICCs were greater than 0.88. The other 3 ICC values (0.61, 0.76, 0.78) concerned determining the sacral base on AP lumbar views. The new ICCs underestimated observer reliability compared with previously published results (intraclass ICCs lower by 0.01-0.02 and interclass ICCs lower by 0.03-0.10). For an error analysis on data from both AP views, the mean absolute differences of observers' measurements were 1.1 degrees to 1.8 degrees for angles and 1.2 mm to 2.3 mm for distances. For the lateral cervical analysis, the observer error was in the interval 0.8 degrees to 3.2 degrees for angles and <1 mm for distances.
Conclusions: The ICCs assuming random crossed factors understate reliability compared with previously published ICC results assuming nested factors. Reliability of the Harrison modified Risser-Ferguson method of line-drawing analysis on AP views is in the high range, with the majority of ICCs >0.88. For both the Harrison modified Risser-Ferguson method on AP views and posterior tangent method on lateral cervical views, the mean absolute differences of observers' measurements are small.