Effect sizes, like relative risks and Cohen’s d, are shown to be important, rather than mere statistical significance.

Bias means systematic error. Its most common form is confounding bias, where various factors in the context of treatment influence the results, without the awareness of clinician or patient. Incorrect claims are made when these confounding factors are ignored. Randomization is the best solution to confounding bias. Clinical examples are provided for antidepressant discontinuation in bipolar depression and for the relationship between substance abuse and antidepressant-related mania. Other forms of bias are discussed, such as measurement bias.

Statistics have a philosophy, which reflects philosophy of science. Different aspects of knowledge in science, as expressed in statistics, are discussed.

The Bayesian approach is a way to interpret a study within the context of the entire literature. It is an important method to use so that a single study isn’t overestimated. It also allows for input of clinical experience.

Observational studies are shown to be important for measuring effect sizes, and assessing outcomes that are not the primary focus of clinical trials, such as side effects.

Randomization solves the problem of confounding bias; it addresses systematic error, which is the most important source of error, not chance. It equalizes all potential confounding factors, known and unknown, in all groups so that they equally influence the results, and thus can be ignored. Only then can the results of randomized treatment be interpreted at face value and causal inferences made. Sample size and other factors are relevant, though, and small randomized clinical trials (RCTs) can be misleading. Examples are given.

The basic insight of evidence-based medicine is that randomized studies are more valid in their results than observational studies or case studies or clinical experience, in that order, because of correction for confounding bias. This concept of levels of evidence is the key to understanding EBM.

Clinical trials are conducted to solve the problem of confounding bias by conducting randomized studies. Examples are given with antidepressants for how clinicians see clinical benefit in practice but randomized clinical trials (RCTs) show that most of the benefit has to do with placebo effects, not the drug pharmacology as clinicians would assume. The conduct of clinical trials is discussed, including the risks of false positive and false negative results depending on how data are analyzed or how sample size is planned. Common errors are identified and criticized.

Meta-analysis can be useful, but it also can be misused. It is an observational study of studies, and thus can be biased.

Confidence intervals are described as more relevant than p-values, as measures of precision of effect sizes.

Statistics grew out of the need to better understand clinical experience. The early “numerical method” simply meant counting, which is better than pure clinical experience. But you can’t simply count either, because of the importance of confounding factors that distort your clinical experience; hence, the need to understand statistical methods.

There is a case to be made for evidence-based medicine (EBM), and there is a case to be made against it. Many of the critiques of EBM are ill-founded, but some important criticisms need attention. The issues and concerns around EBM are discussed.

This chapter reviews some basic principles of applying prior concepts in the book to reading a typical research study article, and interpreting the results efficiently and correctly.

The peer review process of publication has limitations, which are discussed. The influence of the pharmaceutical industry can be beneficial and harmful, both of which are examined.

p-values are measures of random error, or chance mistakes. They are meant to be used mainly in RCTs, where confounding bias has been removed. They are misused in observational studies.

The two major sources of error are chance (random error) and confounding bias (systematic error). After correcting for these two kinds of error, one can then assess or assert causation. These are the “Three Cs.”