Histogram
Also known as: frequency distribution chart, distribution plot, bin chart, frequency histogram, data distribution visualization
A histogram is a data visualization showing the distribution of a numerical variable.
Values are grouped into ranges (bins or buckets), and the height of each bar represents the count (or proportion) of data points falling within that range. Unlike a bar chart (which compares categories), a histogram reveals the shape of a distribution: whether it is symmetric or skewed, whether it has one peak (unimodal) or two (bimodal), and whether values cluster tightly or spread widely. Histograms are one of the first visualizations analysts use when exploring a new dataset -- the shape of a distribution often determines which statistical methods are appropriate.
Histogram, explained at three levels
At home
A histogram is a chart that shows you how data is spread out. Imagine you measured 100 people's heights and wanted to see the pattern. You group the heights into ranges (called bins): 150-155cm, 155-160cm, 160-165cm, and so on. For each range, you draw a bar whose height shows how many people fell in that range. The resulting chart (the histogram) shows you the distribution: most people clustered near the middle (say 165-175cm) with fewer people at the extremes. The shape of the histogram tells you a lot about what the data looks like.
At work
Histograms are used in business analytics to understand the distribution of numerical metrics before reporting or modeling: customer age distribution (are we serving mostly young or old customers?), order value distribution (are most orders small with a few large ones, or evenly spread?), wait time distribution (are customers experiencing mostly short waits with rare long ones, or consistently medium waits?). A histogram of any metric you work with regularly tells you whether the mean is a representative summary (symmetric distribution) or misleading (skewed distribution), and whether there are unusual clusters (bimodal distribution) worth investigating.
At home
Reading histogram shapes: symmetric / bell-shaped (values cluster around the center, mean and median close together); right-skewed (most values low, long tail to the right -- typical for income, prices, response times); left-skewed (most values high, long tail to the left -- typical for test scores when a test is easy); bimodal (two peaks -- suggests two distinct subpopulations mixed together); uniform (all values equally common -- typical for random numbers). Each shape suggests different analytical approaches: right-skewed data often benefits from log-transformation before analysis; bimodal distributions suggest the data should be analyzed in separate groups.
At work
Bin width selection affects interpretation: narrow bins make histograms noisy (showing every little local variation); wide bins lose distribution shape detail (combining distinctly different values into the same bar). There is no universally correct bin width, but automated methods (Sturges' rule: log2(n)+1 bins; Freedman-Diaconis rule using IQR and sample size) provide reasonable starting points. For business communication, choose bin widths that are interpretable in the context of the metric: for customer order values, $50 bins might be meaningful; for response times, 30-second bins might be appropriate. Avoid bin widths that cut across natural thresholds in the business ($95-$105 bins that split the $100 price point into two categories).
At home
Kernel density estimation (KDE) as an alternative to histograms: histograms are affected by bin width choice and bin boundary placement (a value at exactly a bin boundary could shift the apparent distribution). Kernel density estimation (KDE) smooths individual data points into a continuous density estimate, producing smoother distribution visualizations not affected by arbitrary bin boundaries. KDE plots (the smooth curve you often see overlaid on histograms in statistical software) provide a better visual impression of the true underlying distribution, particularly for smaller samples where histograms can look jagged. The tradeoff: KDE plots can smooth over genuine discreteness in data (if values are truly discrete, the smoothed density misrepresents the data).
At work
Histograms in quality control (process capability analysis): in manufacturing and service quality, histograms of process output measurements are compared against specification limits to assess capability. A process whose histogram is centered within specification limits with most values far from the limits has high capability (Six Sigma methodology quantifies this with Cp and Cpk indices). A process whose histogram is shifted toward a specification limit or whose spread extends beyond the limits has poor capability and requires process improvement. Histogram analysis of process output is a standard component of Six Sigma DMAIC (Define, Measure, Analyze, Improve, Control) projects.
Real-world examples
At home
Fatima was analyzing her running times from 200 training runs over the last two years. She made a histogram of her finish times for a standard 5km distance. The histogram showed a bimodal distribution: one cluster around 28-30 minutes (her typical times when well-rested) and another cluster around 34-36 minutes (slower times). She investigated and found the slower cluster corresponded to early morning runs in winter -- she was consistently slower in cold, dark mornings. The histogram revealed a pattern she hadn't noticed from individual run records.
At work
Nadia was preparing a customer segmentation analysis. Before running any models, she made histograms of key customer metrics: purchase frequency, average order value, and time since last purchase. The order value histogram was strongly right-skewed (most customers spending under $100, with a long tail of high spenders). She log-transformed order value before including it in the clustering model, which improved the segmentation quality significantly. The histogram had revealed the skewness that informed the appropriate transformation.
Common confusions
- Many people confuse histograms with bar charts — Histograms and bar charts look similar (both use rectangular bars) but represent different types of data. Bar charts compare counts or values across discrete categories (sales by region, products by popularity); the bars can be reordered without changing meaning, and there are typically gaps between them. Histograms show the distribution of a continuous numerical variable grouped into ranges; the bars must be in order and typically touch each other (because the ranges are contiguous). Using a bar chart when you want a distribution (histogram) or vice versa is a common visualization error that misrepresents the data's nature.
- Some assume that a single histogram with one set of bin widths gives a complete picture of the distribution — Histogram appearance is sensitive to bin width and starting point. The same data can look unimodal with wide bins and bimodal with narrow bins, or show different peak locations with different bin starting points. Examining histograms at multiple bin widths (or using a KDE plot alongside the histogram) provides more robust conclusions about distribution shape. For critical analytical decisions based on distribution shape (choosing appropriate statistical tests, deciding whether log-transformation is needed, identifying multimodal structure), plotting multiple bin widths helps distinguish genuine structural features from visualization artifacts.
For analysts
Recommending histograms as a first step in any data exploration is a practical analytical habit to build into client team workflows. Before calculating any summary statistics, fitting any models, or drawing any conclusions, a histogram of each key numerical variable reveals: whether the distribution is symmetric or skewed (determining which summary statistics are appropriate), whether there are unexpected clusters or bimodal structures (suggesting segmentation), whether there are values at impossible extremes (data quality issues), and the general range and spread of values. A five-minute histogram exploration prevents errors that would take hours to diagnose after the fact.
History & origin
Karl Pearson introduced the histogram as a visualization tool in the 1890s. The name combines 'histos' (Greek for mast or upright structure) and 'gram' (written record). Histograms became standard in statistical textbooks and scientific papers through the 20th century. The choice of bin width is a key decision: too few bins obscure the distribution's shape; too many bins produce a noisy, jagged appearance. Optimal bin width selection methods (Sturges' rule, Scott's rule, Freedman-Diaconis estimator) were developed to automate reasonable bin size selection. Modern data visualization tools (ggplot2, Matplotlib, Tableau) compute optimal bins automatically but allow manual adjustment.