Statistics for outcomes research

Statistics is the language of data and numbers. The appropriate use of statistical methods ensures accurate interpretation of research results. This article is intended to orient the reader to major concepts underlying the selection and application of univariate and multivariate statistical techniques. Data can be categorized as being qualitative or quantitative. Both types of data should first be analyzed with descriptive statistics in order to understand the underlying distribution of values. Qualitative data is analyzed with tests examining the frequency of observations (i.e. Chi-square test). Quantitative data should be further classified as being parametric (normally distributed) or non-parametric (non-normally distributed). This classification is important when selecting a test to compare two or more groups of observations. The t-test is usually used when data is normally distributed, whereas other tests (i.e. Mann-Whitney, Wilcoxon) are more appropriate with non-normal distributions. Multivariate statistics such as linear regression or logistic regression are able to separate the effects of multiple variables on a single outcome of interest. Linear regression is used for continuous outcomes such as height, weight, or age. Logistic regression deals with binary (yes/no) outcome variables such as whether or not a complication occurred. This brief review is meant to advance the reader's knowledge of basic statistical methods and warn of potenial pitfalls in data analyses.