Objectives: Many authors argue that inconsistencies between studies of skeletal markers are based on different data collection protocols, especially when comparing age-related markers such as osteoarthritis. Less attention is given to the choice of statistical techniques that are used to test the hypotheses associated with the data. This paper addresses how different statistical techniques compare the prevalence of age-related skeletal indicators, specifically osteoarthritis.
Materials and methods: Osteoarthritis prevalence was scored in eight postcranial joints in 243 adult individuals from seven prehistoric archaeological sites in Central California, and data was compared between three time periods [Early (4800-2800 BP), Middle (2800-1200 BP), and Late (1200-250 BP)] using commonly used statistical tests: chi-square, Fisher's exact, and odds ratios. In addition, we analyzed the data with tests that are able to take into consideration the effect of age on osteoarthritis prevalence: ANCOVA and Factorial ANOVA. Finally, bootstraps were applied to the data to investigate how fluctuating frequencies, sample size, and age-at-death distributions affected the interpretations resulting from each test.
Results: The results demonstrate that the tests that consider age as a covariate (ANCOVA and Factorial ANOVA) are more efficient in rejecting the null hypothesis when smaller magnitudes of difference are observed between samples, irrespective of sample size, even though osteoarthritis prevalence fails to meet assumptions of normal distribution and homoscedasticity.
Discussion: ANCOVAs or Factorial ANOVAs that incorporate age as a covariate should be considered more often in studies that test different prevalences of age-related osteological markers among past populations.
Keywords: ANCOVA; osteoarthritis; quantitative methods.
© 2017 Wiley Periodicals, Inc.