On the Application of Statistical Analysis for Interpretation of Experimental Results in Environmental Microbiology

Abstract

Statistical analysis is an integral part of every experiment, since it helps researchers to make conclusions out of the work done. Textbooks with detailed description of the algorithms for statistical analysis exist, as well as diverse software packages. Statistical methods are, however, often applied incorrectly, which leads to erroneous and inadequate conclusions. The present article is an attempt to determine the main problems emerging in the course of statistical analysis of experimental results in the field of environmental microbiology. For instance, the classical parametric tests, which are most often used in experimental articles (t-test, analysis of variance, Pearson correlation coefficient, etc.), are applicable only when a random variable of n independent observations has normal distribution. Importantly, the normality of the distribution must be proved, and it is possible to do only for minimal sample size of 20‒30 independent observations per group. The latter is crucial for the incubation, isotope, and molecular biological experiments. The family of normal distributions is not the only family of parameter-dependent (parametric) distributions of random variables. Moreover, real-life distributions usually differ from the normal ones, and normal distribution may be considered only as a certain approximation. High diversity of parametric families complicates the choice of criteria and statistical tests for analysis of a set of data collected from n independent observations. Nonparametric statistics may be of help, since it is free of sample size and distribution requirements, although, as any method of analysis, it also has limitations. An increasing number of experimental data, including those in the field of environmental microbiology, are nowadays analyzed by using nonparametric statistics, which indicates a certain tendency for substitution of parametric methods by nonparametric ones.