You have cleaned the data and studied each variable alone. Now the real question arrives: how do two variables move together, and which pairing of plot answers it fastest?
Match the plot to the variable pair. Univariate plots describe one column. Bivariate plots describe how two columns relate. The single decision that drives everything is the type of each variable.
Categorical versus categorical. Cross two factors with a bar chart of counts. A stacked bar chart puts one factor on the axis and stacks the second factor's segments inside each bar, good for showing total size plus composition. A split bar chart (dodged, side-by-side) places each second-factor level as its own bar, better for comparing levels directly. Switch stacked bars to proportions when group totals differ, so composition is comparable across unequal bars.
KEY: Stacked shows the whole and its parts. Split (dodged) shows level-to-level comparison. Proportional stacking removes the distortion of unequal group sizes.
Categorical versus numeric. You want to see how a numeric variable's distribution changes across the levels of a factor.
Common mistakes
- Calling collinearity an interaction. A tight predictor-versus-predictor scatter shows redundancy, not joint effect. Graders explicitly denied credit for the interaction answer at .
- Reading the target as a predictor. On a heat map that includes the target (like TOTAL_KWH), its correlations are the signal you want, not a collinearity flaw among predictors.
- Ignoring lines of points. A vertical stack at zero square feet or horizontal stack at zero price marks missing or non-market values, not a trend; interpret both lines.
Bottom line
- Bivariate exploration compares two variables at once to reveal association, not just distribution.
- The variable-type pair dictates the plot: pick by whether each side is categorical or numeric.
- Categorical vs categorical: stacked or split (dodged) bar charts of counts or proportions.
- Categorical vs numeric: split boxplots or overlaid/faceted histograms of the numeric variable by group.
Exam shortcut
First classify each variable as categorical or numeric, then read the plot straight off the pair: two categoricals get bar charts, a categorical and a numeric get boxplots or histograms, two numerics get a scatterplot. When a heat map or tight scatter shows near-±1 correlation between two predictors, write "collinearity" and name a fix (significance, stepwise, or regularization), never "interaction." Any straight vertical or horizontal line of points means constant...
The full lesson (about 2,213 words, 15 min read) adds 2 worked examples, all 6 common mistakes, a self-check, free in the app.
Learning objectives
- 2f
Browse all free Exam PA lessons or jump into free Exam PA practice questions.