AI Glossary / Statistics and probability / Foundation
Frequency distribution
Definition
A frequency distribution organizes observations by how often their values or categories occur. It can be presented as a table or graph, showing counts for individual values, categories or numerical intervals. Relative frequencies divide those counts by a defined total, while cumulative frequencies add counts in a meaningful order. A frequency distribution describes the recorded data; it is not automatically the true probability distribution of the wider population. Its interpretation depends on the measurement, sampling process, grouping rules and denominator.
What it means
A list of individual observations can be difficult to interpret. Grouping values and recording their frequencies makes common outcomes, sparse regions and patterns of spread easier to see. Categorical data may be displayed with bars, while numerical intervals can be displayed in a histogram. The chosen grouping matters: broad intervals hide detail, and narrow intervals can emphasize noise in a small sample. Different displays can summarize the same data without changing the observations, so the display should make its choices visible rather than suggesting that one appearance is uniquely determined.
How it works
Define the unit of observation and identify valid, missing and excluded records. For categories, use clear labels and specify whether each observation belongs to one category or can contribute to several. For numerical intervals, choose boundaries that are exhaustive for the intended data and do not overlap. Count observations, verify the total and compute relative frequencies if useful. For ordered values, cumulative counts provide totals at or below specified boundaries. In a density-scaled histogram, divide each relative frequency by its bin width, so rectangle area represents the fraction of observations. This is especially important when intervals have unequal widths.
Three practical examples
- Customer operations: tabulate contact reasons to see which recorded categories occur most often.
- Quality analysis: summarize measured dimensions in intervals to inspect spread, skewness and unusual values.
- AI data review: count outcome labels to identify class imbalance, while distinguishing the dataset's composition from the target population's prevalence.
Think of it this way
Think of sorting a pile of objects into labeled trays and counting what lands in each tray. The trays help reveal the composition of the pile. Their labels and sizes matter, and the resulting counts describe that pile rather than every object that might exist elsewhere.
Worked example
Take 12 hypothetical values: [1, 2, 2, 3, 4, 4, 4, 7, 8, 9, 12, 14]. Use intervals [0,5), [5,10) and [10,20), where the left boundary is included and the right boundary excluded. Counts are seven, three and two, summing to 12. Relative frequencies are 7/12, 3/12 and 2/12, approximately 58.33%, 25% and 16.67%. Cumulative counts at the upper boundaries are seven, 10 and 12. The interval widths are five, five and 10. For a density histogram, heights are 7/(12 times 5) = 0.11667, 3/(12 times 5) = 0.05 and 2/(12 times 10) = 0.01667. Multiplying each height by its width recovers its relative frequency, and the total area is one. Simply plotting unequal-width bars with raw counts as heights would not make their areas proportional to frequency.
Illustrative case study
Fictional case: Fern Support Analytics compares the distribution of ticket categories across two months. Analysts notice that a newly introduced category absorbed records previously labeled other. They document the classification change before interpreting it as a shift in customer needs. They also show counts alongside percentages because total ticket volume changed. This invented case illustrates why a clear frequency display must preserve definitions and denominators, not merely produce an attractive chart.
Common misconceptions and mistakes
Do not confuse a count with a percentage or a density height with a probability. Density heights can exceed one when units or narrow intervals make that possible; probability is represented by area. Cumulative frequencies require a meaningful order and are usually inappropriate for arbitrary nominal labels. Multiple-response categories can sum to more than the number of people, so state the counting unit. Missing values should not silently alter the denominator. A histogram's shape depends on binning, and a smooth-looking pattern does not prove a particular population model.
Business use
Businesses can use frequency distributions as an early check on data quality and operational patterns. Retain counts, definitions and the population or sample scope so comparisons remain interpretable. Compare percentages when totals differ, but preserve counts to show scale and identify sparse categories. Use consistent bins for meaningful side-by-side comparisons, or explain why they changed. For weighted survey data, distinguish weighted estimates from raw sample counts. Before using a pattern to guide action, investigate whether it reflects collection practices, coding changes, selection or genuine differences in the underlying activity.
People also ask
Is a frequency distribution always a histogram?
No. It can be a table, bar chart or another suitable display.
Must relative frequencies sum to one?
They do for mutually exclusive, exhaustive categories using the same denominator, apart from rounding.
Does the most frequent interval identify an exact mode?
No. Grouping identifies a modal interval and can hide the precise pattern within it.
Check your understanding
Ten respondents can each choose several reasons, producing 18 selections. Should the category counts necessarily sum to 10?
Answer: No. They sum to selections unless the analysis defines a different counting rule. Percentages must state whether their denominator is people or selections.
Related terms
- Relative frequency
- Histogram
- Bar chart
- Class interval
- Empirical distribution
- Frequency
- cumulative frequency
- frequency density
- class imbalance
- weighted frequency
Learn next
Learn how data type determines an appropriate display, then compare count, relative-frequency and density scales. Study the effects of bin width and denominator choice before interpreting differences between distributions.
More in Statistics and probability
- Fractiles - Fractiles are values that locate specified fractions of a distribution, including quartiles, deciles and percentiles.
- Fractional factorial design - A fractional factorial design is an experimental design that uses a carefully selected subset of the treatment combinations in a full factorial experiment.
- Fractional model - A fractional model is a statistical model for an outcome expressed as a fraction or proportion, usually between zero and one.
- Frequentist statistical inference - Frequentist statistical inference uses data to estimate or test properties of a population or process through procedures evaluated by their behavior under...
- Friedman test - The Friedman test is a rank-based test for comparing three or more related conditions in a complete-block or repeated-measures design.
- Frisch - Waugh - Lovell theorem - The Frisch-Waugh-Lovell theorem is an exact algebraic result for ordinary least squares regression.
Sources
Australian Bureau of Statistics, Frequency distribution: https://www.abs.gov.au/statistics/understanding-statistics/statistical-terms-and-concepts/frequency-distribution NIST/SEMATECH, Histogram and normalization: https://www.itl.nist.gov/div898/handbook/eda/section3/histogra.htm Australian Bureau of Statistics, Statistical terms and concepts glossary: https://www.abs.gov.au/statistics/understanding-statistics/statistical-terms-and-concepts/statistical-terms-and-concepts-glossary
Previous: Free word order Next: Frequency principle/spectral bias