Abstract:
Non-parametric and semi-parametric statistical approaches were developed to maximize accuracy of colorimetric data for monitoring biocolonization on stone surfaces, while simultaneously optimizing efficiency of data collection in the field. These approaches were applied to colorimetric data sets collected on three Kasota limestone capstones located at the National Museum of the American Indian in Washington, DC. Data was randomly resampled without replacement (the statistical “jackknife”), producing data subsets of diminishing resample sizes. This study assesses the extent to which reduced numbers of colorimetric measurements diverge from characterizations provided with greater numbers of measurements on the same stone surface. Resample subsets were compared against parent colorimetric data sets from each stone in terms of standardized effect size (Cohen’s d). A universal lowest limit on the number of colorimetric measurements necessary to accurately represent a stone surface cannot be established, as this will depend on properties intrinsic to the stone. However, it is shown that for levels of d = ±0.2, ±0.5, ±0.8 at 95 percent confidence, an homogeneously biocolonized dolomitic limestone requires approximately 35, 10, and 5 colorimetric measurements, respectively. By comparison, hypothesis testing following the Student’s t-test at d = ±0.2, ±0.5, ±0.8 showed that approximately 55, 20, and 10 measurements were required. Factors affecting the necessary minimum sample size for achieving pre-selected confidence levels and acceptable measurement error – including the impacts of a biocide treatment and heterogeneity of surface textures – were also investigated. Comparison of results for textured capstones suggests that rougher stones require greater numbers of measurements at identical d and confidence. Corresponding author: Paula DePriest.
