| abstract |
This study investigates current controversies surrounding the use of p-values in statistics
and related fields. The definition of statistical terms, such as statistical significance, are
investigated through their contributions to p-hacking, publication bias, and misconceptions in
statistics education. Past datasets are utilized to analyze the effectiveness and correctness of
p-values and other statistical analysis methods. Further, a series of statistical studies are
conducted to conclude that p-values have a few limitations when compared to alternative
statistical measures, such as Bayesian statistics, through the use of statistical modeling. These
studies prompt the discussion that the understanding around p-values requires clarification and
modification for some. Thus, I clarify specific cautions on the use of p-values and discuss
alternate methods of analysis.
Introduced as an alternative means of analysis, Bayesian statistics involves a distinct
school of thought that is not solely based on the data itself. For this reason, I pursue both classic
non-Bayesian and Bayesian methods further to gauge their respective strengths and weaknesses
through the implementation of a simple statistical task of estimating the probability of a
potentially biased coin. Modeling and simulation results yield Bayesian statistics as the more
adaptable analysis method due to its probabilistic reasoning and incorporation of prior
knowledge. The advantage of the Bayesian model over the usage of p-values is further discussed
using modern day applications with severe acute respiratory syndrome coronavirus 2
(SARS-CoV-2).
|
