Histograms display the distribution of quantitative data by grouping values into bins and showing frequency or relative frequency as bar height. Unlike bar charts, histograms show continuous data with adjacent bars and reveal the shape of distributions.
Create histograms with different bin widths and observe how it affects the appearance. Compare histograms to actual frequency tables. Practice interpreting histograms from real data sources.
Using histograms for categorical data (requires bar charts). Thinking the bin width doesn't matter (it greatly affects interpretation). Confusing histogram height with total frequency rather than density or relative frequency.
You've already worked with frequency distributions and tables, which organize raw data into counts per category or interval. A histogram is the visual companion to those tables: it translates the numerical frequency information into a spatial picture where you can see the shape of the distribution at a glance. Making this translation correctly — and reading it carefully — is more nuanced than it first appears.
The core construction: divide the range of a quantitative variable into bins (intervals of equal width, typically), count how many observations fall in each bin, and draw a bar for each bin whose height equals the frequency or relative frequency. Unlike a bar chart for categorical data, histogram bars are adjacent with no gaps, because the data is continuous — adjacent bins represent adjacent numerical ranges, not separate categories. The x-axis is a number line; the y-axis shows frequency (raw count), relative frequency (proportion), or frequency density (proportion divided by bin width, so bar areas sum to 1 when bin widths vary). Always check which y-axis convention is being used before comparing histograms.
The most important skill is reading the shape of the distribution from the histogram, not just tallying individual bars. A roughly symmetric bell-shaped histogram suggests approximate normality — equal spread on both sides of the peak. A histogram with a long right tail (peak on the left, tapering rightward) is right-skewed — a few large values pull the distribution out; income and wealth are classic examples. A left-skewed distribution has its tail on the left. Bimodal histograms with two distinct peaks suggest the data may come from two different subpopulations mixed together, which a single summary statistic like the mean would completely obscure. This shape vocabulary feeds directly into your next topic on skewness and kurtosis as formal measures.
Bin width is a genuine analytical choice, not a default setting. Too few bins (wide intervals) and you lose resolution — all the data looks like one undifferentiated lump. Too many bins (narrow intervals) and the histogram becomes noisy — jagged peaks from random sampling variation obscure the underlying shape. A good choice reveals the distribution's structure without overfitting to the specific sample. Rules of thumb like Sturges' rule (k ≈ log₂ n + 1 bins) or the Freedman-Diaconis rule (bin width proportional to IQR / n^{1/3}) give starting points, but visual judgment matters. This tension between resolution and noise is your first encounter with a theme that recurs throughout statistics: fitting the data versus revealing the underlying pattern.
No topics depend on this one yet.