Turn raw observations into reproducible bins, then overlay a configurable kernel density estimate.
Bars show observed frequency; the curve is a Gaussian kernel density estimate.
| Count | Minimum | Q1 | Median | Mean | Q3 | Maximum | Std. dev. | Bins | Bandwidth |
|---|
This free histogram maker groups numeric observations into adjacent intervals and optionally overlays a kernel density estimate. Use it to inspect center, spread, skew, tails, gaps, and possible multiple modes without sending your data to a server.
Select Import data to upload a CSV, TSV, TXT, JSON, XLSX, or XLS file, or paste comma-separated data. Choose the numeric column to analyze, review the preview, and exclude rows that should not enter the distribution.
Freedman–Diaconis uses sample size and interquartile range, making it resistant to outliers. Scott uses standard deviation, Sturges grows logarithmically with sample size, and the square-root rule offers a simple baseline. Fixed count gives direct control when a reporting standard requires specific bins.
Count shows observations per bin. Probability and percent show each bin’s share of the sample. Probability density scales bar area to one, which is especially useful when comparing a histogram with a density estimate or with histograms that use different bin widths.
A kernel density estimate places a small smooth function at every observation and adds those functions together. The curve is scaled to the selected histogram normalization, so its vertical position remains comparable with the bars.
The bandwidth controls smoothing. A narrow bandwidth can expose local structure but may turn random sample noise into extra peaks. A wide bandwidth produces a stable curve but can hide shoulders or multiple modes. Silverman and Scott provide reproducible starting rules; the multiplier lets you test sensitivity.
Gaussian, Epanechnikov, and triangular kernels differ in shape and support, but bandwidth and sample quality usually matter more. Boundaries, repeated values, outliers, and small samples can distort the estimate. Compare the KDE with the actual bars and optional rug marks.
Keep the units in the X-axis label and state the normalization when readers might assume counts. PNG and SVG exports include the active bins, KDE settings, colors, titles, grid, and optional transparent background.
Import individual delivery times rather than averages. The bars can reveal a common central range, a long tail of delayed orders, gaps caused by service tiers, or more than one peak. Add rug marks to retain a sense of the observations, then use the density curve as a smooth companion to the counts rather than a replacement for them.
Two histograms are directly comparable only when they use the same bin boundaries, units, and normalization. Automatic rules are useful starting points, but they may choose different boundaries for different samples. Fix the bin specification when comparing periods or groups, and report whether the vertical scale shows counts, proportions, probability, or density.
A small sample can look jagged simply because few observations are available. Rounded measurements create repeated values, and natural limits can distort a symmetric kernel near a boundary. Summarize sample size and collection conditions, investigate data-quality issues, and compare with a box plot when exact quartiles and outliers matter.