Common Statistical Misconceptions in Biomedical Research

As a university student who studies Statistics, I often use many statistical methods to analyze various questions that I want to research in my daily study. While there are some statistical misconceptions and improper statistical analysis methods that I did know before but was used in my past statistical researches. From the second week's materials of my statistics course, I learned some misconceptions about data analysis, especially p-hacking is appropriate, p-values covey information about effect size and the standard error of the mean quantifies variability, which inspires and interests me a lot. Afterwards, I understand more statistical misconceptions, particularly in the field of biomedicine. In the following, I plan to share some common statistical misconceptions in biomedical scientific research that I learned recently. 

First of all, when analyzing data from a census procedure, many researchers may use statistical inference tools inappropriately, especially hypothesis testing and confidence intervals. These researchers rely on hypothesis testing to find their answer to the research question from data. For example, the observed effect size is 25% in a study that aims to research the effectiveness of an intervention. Then, the researchers want to analyze whether 25% is good enough and how the observed effect size compares to the alternative intervention according to the influence of the results. However, there is no p-value that can help the researchers to draw a conclusion to the research question and they would feel confused about the meaning of the results. Thus, if data are not sampling, then there is no uncertainty, so there will be no inference, confidence intervals and p-values which means you cannot make a reasonable conclusion from the results of hypothesis testing. 

Additionally, using separate p-values to compare groups is not proper. In statistical analysis, researchers may only observe the statistically significant change in X when group A is present but not observe the statistically significant change in X when group B is present. This error might lead researchers to mistakenly conclude the effects of group A and group B are distinct. However, the correct method is to observe the statistically significant change in X when group A and B are both present and compare the two groups directly. Moreover, it is better to avoid using a simple and cheap statistical design, called "before-after" study design, which can only measure the given variable on a single group of subjects rather than multiple groups of subjects. Since the study design does not comprise a control group, the results might contain many biases.

Furthermore, reporting baseline statistical comparisons in randomized trials is unnecessary. The first reason is that this analysis cannot help researchers to get the answer to the research question. Secondly, if randomization is used to assign specific treatments to participants in the biomedical study, then the null hypothesis is true for all baseline characteristics. 

These three statistical misconceptions mentioned above commonly exist in biomedical research. I hope more misconceptions can be avoided in future data analysis in any field. 

Reference:

Navarrete, M. S. (2019). Common statistical misconceptions-plainly explained. Medwave, 19(06). doi:10.5867/medwave.2019.06.7660