5 Common Statistical Mistakes and How to Avoid Them

By Timothy Achala | 16/02/2026

Mistake 1: Misusing Log Transformation to "Fix" NonNormality The Problem in Detail One of the most pervasive mistakes in statistical analysis is the automatic, reflexive application of logarithmic transformations (or BoxCox transformations) to skewed data, particularly the dependent variable in regression analysis. The typical reasoning goes something like this: "My residuals aren't normal, therefore I need to transform my data to make them normal, so I'll apply a log transformation and then everything will be fine." This approach reveals a fundamental misunderstanding of both what statistical methods require and what transformations actually do to your data and your scientific questions. When you read scientific articles, blogs, tutorials, and even textbooks, you'll quickly notice that transformation (especially logtransformation) of variables, particularly the response variable in regression analysis or hypothesis testing, is extremely common. Some researchers at least attempt to justify this step, some try the MaximumLikelihood approach through BoxCox, but many transform data almost automatically if the data are skewed or "nonnormal," completely ignoring basic questions: "But what for? What will I gain with that? Can I explain what happened after I did it?" What Transformations Actually Do When you transform your response variable, you fundamentally change the nature of your statistical model and the questions it can answer. You are not simply "fixing" a violation of assumptions while leaving everything else intact. Instead, you are: 1. Forcing your variables to