AP Statistics: Comparing Distributions

Comparing distributions is the statistical engine for making sense of differences between groups. Whether you’re analyzing the effectiveness of two medical treatments, comparing test scores across different teaching methods, or evaluating engineering tolerances for two manufacturing processes, this skill moves you from simply describing a single dataset to making meaningful, evidence-based comparisons. This process formalizes observation into statistical argumentation.

The Core Purpose: From Description to Comparison

In earlier studies, you learned to describe a single distribution using the Four Pillars of Descriptive Analysis: shape, center, spread, and outliers. Comparing distributions means applying these same pillars to two or more groups simultaneously, but with a crucial shift in language and focus. Instead of stating "the distribution is roughly symmetric," you now state "Distribution A is more symmetric than Distribution B." You trade absolute description for comparative analysis. The goal is to identify and quantify differences, supported by visual evidence and numerical summaries, while always considering the real-world context of the data. This is a foundational step toward formal inferential statistics, where you test if observed differences are likely due to chance.

Tools for Visual Comparison

Effective comparison starts with choosing the right visual tool. Each graphical method highlights different aspects of the distributions and is suited to particular data types and sample sizes.

Back-to-Back Stemplots (or stem-and-leaf plots) are ideal for comparing two small to moderately sized datasets. They preserve the raw data values while aligning two distributions along a common scale. To create one, you place a shared stem in the center column. The leaves for one group extend to the right, and the leaves for the other group extend to the left. This side-by-side display makes comparing shapes, clusters, gaps, and general ranges immediate. For instance, a back-to-back stemplot of reaction times for a treatment vs. control group can quickly reveal if one group has a tighter cluster of values (less spread) or a concentration of leaves at lower stems (a lower center).

Parallel Boxplots (side-by-side boxplots) are the most efficient tool for comparing multiple distributions, especially when focusing on the Five-Number Summary (minimum, $Q_1$, median, $Q_3$, maximum). By drawing boxplots for each group on the same axis, you can instantly compare their medians (centers), interquartile ranges, or IQRs (spreads), and any outliers. The alignment of the boxes tells a clear story: if the boxes do not overlap, it suggests a substantial difference in the middle 50% of the data. Parallel boxplots are excellent for larger datasets where stemplots become cluttered and for making clear, high-level comparisons of center and spread.

Overlaid Histograms are powerful for comparing the shapes of two distributions, particularly when using density scaling (where the area represents proportion). By plotting two histograms on the same axes, often with semi-transparent bars or different shading, you can see how the densities overlap and diverge. This is critical for assessing whether one group is skewed while another is symmetric, or if one distribution is bimodal. A key technical consideration is ensuring both histograms use the same bin widths and scale on the vertical axis; otherwise, the comparison is invalid. Overlaid histograms provide an intuitive sense of the overall distributional differences at a glance.

The Comparative Description Framework

With your visual in hand, you must now articulate the comparison systematically. This requires using comparative or superlative language and citing specific visual and numerical evidence.

1. Compare Shape: Describe the overall form. Are both symmetric? Is Group A strongly right-skewed while Group B is approximately normal? Is one unimodal and the other bimodal? Use the visual (histogram or stemplot) as your primary evidence.
2. Compare Center: Identify which group has the larger typical value. The median (from a boxplot or stemplot) is the preferred measure for comparison as it is resistant to outliers. Do not just state the medians; say, "The median reaction time for Group X (450 ms) is substantially higher than for Group Y (320 ms), indicating a slower typical response."
3. Compare Spread: Describe the variability within each group. Use the IQR (from a boxplot) or the general range visible in a stemplot/histogram. For example, "Group A has a much wider IQR (${\text{IQR}}_A=25$) than Group B (${\text{IQR}}_B=10$), showing its middle 50% of data is more dispersed."
4. Note Outliers: Identify any outliers present in either group, referencing the 1.5 x IQR rule in the context of boxplots. Mentioning outliers is essential because they can influence measures like the mean and can be of substantive interest themselves.

Your final description should synthesize these points. A strong comparison might read: "While both distributions of part diameters are roughly symmetric, the distribution from Machine 2 is clearly centered at a higher value (median = 10.2 mm vs. 9.8 mm) and shows less variability (IQR = 0.3 mm vs. 0.5 mm). Machine 1 produced one part that was a clear outlier below the lower fence."

Contextualization and Statistical Evidence

A comparison is incomplete without context. The numbers and shapes are meaningless unless tied back to the problem's setting. After stating, "The median salary for engineers is higher than for technicians," you must add the context: "...which aligns with the greater years of specialized education typically required for engineering roles." Furthermore, you must provide statistical evidence for every claim. Point to the graphic: "As seen in the parallel boxplots, the median line for engineers is visibly above the entire box for technicians." Quote the numbers: "This difference is quantified by a median of $85,000comparedto$55,000." This combination of visual reference, numerical summary, and contextual interpretation transforms your analysis from a mechanical exercise into a coherent statistical narrative.

Common Pitfalls

1. Making Claims Without Visual/Numerical Evidence: Stating "Group A has a higher center" without pointing to the higher median on the boxplot or providing its value is an unsupported assertion. Correction: Always pair comparative statements with a direct reference to the graph or a specific statistic.

1. Using Absolute Language: Saying "Distribution A is symmetric" and "Distribution B is symmetric" when comparing them misses the point of the task. Correction: Use comparative language: "Both distributions are roughly symmetric, though Distribution B is more perfectly bell-shaped than A."

1. Ignoring Scale and Alignment When Comparing Graphs: Creating two separate graphs with different axes scales will visually distort the comparison. A boxplot for Group A with a y-axis from 0-100 placed next to one for Group B with a y-axis from 50-150 is misleading. Correction: Always use parallel boxplots on a single, common scale, or overlaid histograms with identical axes.

1. Over-Interpreting Small Differences: Not every visible difference is meaningful, especially with small sample sizes. A slightly higher median in a back-to-back stemplot based on 10 data points may not be significant. Correction: Note the difference, but qualify it by mentioning the sample size. Your role here is to describe the data's story, not to declare definitive truth, which is the job of formal inference.

Summary

• Comparing distributions elevates descriptive statistics by using the Four Pillars—shape, center, spread, and outliers—to analyze differences between two or more groups.
• The primary visual tools are back-to-back stemplots (for small, detailed comparisons), parallel boxplots (for efficient comparison of centers and spreads), and overlaid histograms (for clear shape and density comparison).
• Every comparative claim about center (typically using the median) or spread (using the IQR or range) must be supported by direct statistical evidence from the graph or numerical summaries.
• Use consistent comparative language (e.g., "higher," "more spread out," "less skewed") rather than simply describing each group in isolation.
• Always conclude your analysis by connecting the statistical differences back to the real-world context of the data, completing the narrative from number to meaning.