Bivariate Analysis

Bivariate analysis is a statistical approach that examines the relationship between exactly two variables simultaneously. It encompasses a range of techniques, including cross tabulation, correlation, simple linear regression, t tests, and chi square tests, each suited to different combinations of variable types (continuous, categorical, or ordinal).

Bivariate analysis is the starting point for understanding associations in research data. Before building complex multivariable models, researchers typically explore pairwise relationships to identify which variables are associated with the outcome of interest, to check for collinearity among predictors, and to verify that the data meet the assumptions required for more advanced analyses. Bivariate results can suggest candidate variables for regression models (although selecting predictors by their bivariate p values is discouraged, because it can discard a variable that matters once others are adjusted for), provide preliminary evidence for hypothesized relationships, and serve as the basis for descriptive epidemiological reports. For readers of research, understanding bivariate analysis is essential for interpreting the tables and figures that appear in virtually every quantitative study.

The choice of bivariate technique depends on the types of variables being analyzed. When both variables are continuous, Pearson's correlation coefficient measures the strength and direction of the linear relationship, while Spearman's rank correlation is used when the relationship is monotonic but not necessarily linear or when the data are ordinal. Simple linear regression models the dependent variable as a linear function of the independent variable and provides an estimate of the slope (the change in the outcome per unit change in the predictor) and its statistical significance.

When one variable is categorical (with two groups) and the other is continuous, the independent samples t test compares the group means. When the categorical variable has three or more groups, one way ANOVA is used. When both variables are categorical, the chi square test of independence evaluates whether the distribution of one variable differs across categories of the other. For ordinal variables, the Mann Whitney U test (two groups) or Kruskal Wallis test (three or more groups) provides nonparametric alternatives.

Each bivariate method produces a test statistic, a p value, and often an effect size measure that quantifies the strength of the association. The p value indicates whether the observed association is statistically significant at the chosen alpha level, while the effect size indicates whether the association is practically meaningful.

It is important to recognize that bivariate analysis does not control for the influence of other variables. An association observed between two variables may be confounded by a third variable that is related to both. For this reason, bivariate analysis is typically a preliminary step that is followed by multivariable analysis (for example, multiple regression), which can adjust for confounders and test more complex models.

Use bivariate analysis as an initial exploratory step when examining the relationships between pairs of variables in your dataset. It is appropriate for hypothesis testing when the research question involves only two variables and no confounders need to be controlled. It is the correct approach for simple descriptive analyses, for checking assumptions before multivariable modeling, and for exploring candidate predictors for regression models, although predictors should not be chosen by their bivariate p values, because a variable that looks unimportant alone can matter once others are adjusted for. It is not appropriate when the research question requires controlling for confounding variables, when the outcome depends on multiple interacting predictors, or when the goal is causal inference in the presence of confounders.

A common mistake is interpreting a statistically significant bivariate association as evidence of a causal relationship. Correlation does not establish causation, and unadjusted bivariate associations may be driven entirely by confounding variables. Another error is reporting only p values without effect sizes, which leaves readers unable to assess the practical significance of the findings.

Researchers sometimes apply parametric bivariate tests (such as Pearson's correlation or the t test) to data that violate the underlying assumptions (normality, linearity, homogeneity of variance) without checking those assumptions or using appropriate alternatives. Using the wrong test can produce misleading p values and conclusions. Finally, conducting many bivariate tests without adjusting for multiple comparisons inflates the probability of false positive findings, as discussed under the Alpha Level entry.

A researcher studying risk factors for workplace burnout in 450 hospital nurses conducts bivariate analyses to examine associations between burnout (measured as a continuous score on the Maslach Burnout Inventory) and several potential predictors. Pearson's correlation shows a significant positive association between burnout score and weekly overtime hours (r = 0.38, p < 0.001). An independent samples t test reveals that nurses in the intensive care unit have higher burnout scores than those in general wards (mean difference = 4.7, 95% CI: 2.1 to 7.3, p < 0.001). These bivariate findings inform the construction of a multiple regression model that adjusts for age, experience, and social support simultaneously.