Histogram

When to use a histogram in Licklider, how bin count is determined, and what the current limitations are.

A histogram shows the distribution of a single continuous variable by dividing values into bins and displaying the count of observations in each bin. It is useful for understanding the shape of a distribution — whether it is symmetric, skewed, unimodal, or has multiple peaks.

This page is about a descriptive figure. A histogram does not produce a p-value, effect size, or formal test result by itself. Its role is to help you see how one variable is distributed before deciding how strongly the data supports a modeling or comparison claim.

When to use a histogram

Histograms are most useful when:

  • You want to understand the distribution of one continuous variable
  • You are checking whether data meets normality assumptions
  • Multimodality or skewness is suspected and you want to visualize it directly

When Licklider detects a multimodal or irregular distribution, it may suggest a histogram as an alternative to a violin plot, which would produce a misleading smooth shape for that data.

That suggestion is meant to reduce a common display error: a smooth density shape can look more certain than the raw distribution really is when the data has multiple peaks, gaps, or visibly irregular structure.

Single variable only

The current histogram in Licklider works with one continuous variable at a time. Displaying multiple groups side-by-side, overlaid, or stacked in the same histogram is not currently supported. If you need to compare distributions across groups, a violin plot or box plot with individual points may be more appropriate.

How bin count is determined

Licklider calculates the number of bins using the square root of the number of observations, rounded up, and clamps the result between 5 and 30. For example:

  • 25 observations → 5 bins
  • 100 observations → 10 bins
  • 900 observations → 30 bins (maximum)

This default is a display heuristic, not a statistical proof that one particular binning is correct. The goal is to avoid a histogram that is so coarse that important structure disappears, or so fine that small sample noise dominates the shape.

The 5-to-30 clamp serves the same purpose. At very low bin counts, the display becomes too blunt to show meaningful shape. At very high bin counts, the figure becomes visually unstable and can exaggerate random variation rather than real distributional structure.

The bin count can be adjusted by specifying a number in the Chat:

  • "Use 15 bins"
  • "Show a histogram with 20 bins"

Different bin counts can make the same data look more or less skewed, more or less peaked, or more or less multimodal. Licklider does not use the automatic bin choice to prove normality, non-normality, or the presence of a true mixture distribution.

What the figure shows

Each bar represents a bin interval. The height of the bar indicates the number of observations whose values fall within that interval.

Use the figure as a visual summary of shape, spread, and rough irregularities. It is helpful for spotting features worth checking further, but it should not be treated as a standalone confirmation that a variable is normal, non-normal, or generated by multiple subpopulations.

A normal distribution curve overlay is not currently available. To check normality visually, Licklider also generates Q-Q plots as part of the normality check results in the Inspector.

That boundary is intentional. A smooth reference curve can be useful, but it can also make readers over-interpret a descriptive figure as if it were a formal fit check. In Licklider, histogram shape is meant to be read together with the Q-Q plot and the normality check results when assumption checking matters.

Minimum data requirement

A histogram requires at least 5 observations.

This minimum exists so the plot has enough values to form visible bins at all. It does not mean that 5 observations are enough to judge distribution shape reliably. With very small samples, a histogram can be shown, but the apparent shape is often unstable and should be interpreted cautiously.

What this page does not cover