![School of Public Health](../../assets/md5images/35d233b2c9a8d4e92cd89657d26cb679.gif)
Dissertations & Theses (Open Access)
Date of Award
Summer 4-2019
Degree Name
Doctor of Philosophy (PhD)
Advisor(s)
Dejian Lai, Phd
Second Advisor
J. Michael Swint, Phd
Third Advisor
Momiao Xiong, Phd
Abstract
Kolmogorov-Smirnov test is a non-parametric hypothesis test that measures the probability of deviations, that the interested univariate random variable is drawn from a pre-specified distribution (one-sample KS) or has the same distribution as a second random variable (twosample KS). The test is based on the measure of the supremum (greatest) distance between an empirical distribution function (EDF) and a pre-specified cumulative distribution function (CDF) or the largest distance between two EDFs. KS test has been widely adopted in statistical analysis due to its virtue of more general assumptions compared to parametric test like t-test. In addition, the p-value derived from the KS test is more robust and distribution-free for a large class of random variables. However, the fundamental assumption of independence is usually overlooked and may potentially cause inaccurate inferences. The KS test in its original form assumes the interested random variable to be independently distributed while it’s not true in a lot of nature datasets, especially when we are dealing with more complicated situations like imgage analysis, geostatistical which may involve spatial dependence. I proposed a modified KS test with adjustment via spatial correlation. The dissertation concerns the following three aims. First, I conducted a systematical review on the KS test, the Cramer von Mise test, the Anderson-Darling test and the Chi-square test and evaluate their performance under normal distributions, Weibull distributions and multinomial distributions. In the review, I also studied how these tests perform when random variables are correlated. Second, I proposed a modified KS test that corrects the bias in estimating CDF/EDF when spatial dependence exists and calculate the informative sample size. Finally, I conducted a revisit analysis of coronary flow reserve and pixel distribution of coronary flow capacity by Kolmogorov-Smirnov with spatial correction to evaluate the efficiency of dipyridamole and regadenoson.
Recommended Citation
Zheng, Wenjun, "Kolmogorov-Smirnov Type Tests Under Spatial Correlations" (2019). Dissertations & Theses (Open Access). 97.
https://digitalcommons.library.tmc.edu/uthsph_dissertsopen/97