Frequency Distribution

A frequency distribution is a summary of how often each value or category (or range of values) occurs in a dataset. It can be presented as a table, histogram, bar chart, or frequency polygon.^[1][2] Frequency distributions reveal the shape, central tendency, spread, and outliers of data, providing the foundation for selecting appropriate statistical tests and understanding the basic properties of a dataset.

Before conducting any statistical analysis, researchers need to understand the distributional characteristics of their data. A frequency distribution is the most fundamental tool for this purpose. It answers the basic question "what do the data look like?" by showing whether values are clustered around a central point, spread evenly, or skewed. The shape of the distribution informs which statistical tests are appropriate (many parametric tests assume that the data, or a model's residuals, are approximately normal) and whether data transformations are needed.^[1] Frequency distributions also reveal outliers, data entry errors, floor and ceiling effects, and other features that can affect the validity of subsequent analyses. In public health and epidemiology, frequency distributions of disease counts, exposure levels, or demographic characteristics are the starting point for nearly every investigation.

For categorical data, a frequency distribution lists each category and the number (or proportion) of observations in that category.^[2] For continuous data, the researcher first divides the data range into intervals (bins or classes) and then counts the number of observations falling within each interval. The choice of interval width affects the appearance of the distribution; too few intervals obscure patterns, while too many intervals fragment the data.^[1][2]

Common visual representations include histograms (bars whose areas represent frequencies, so their heights do too when the intervals are of equal width, with adjacent bars touching to indicate continuous data), bar charts (similar but with spaces between bars, used for categorical data), frequency polygons (lines connecting points plotted at the midpoint of each interval, at the height of its frequency), and cumulative frequency curves (ogives). In addition to raw frequencies, researchers often compute relative frequencies (proportions or percentages) and cumulative frequencies (the running total of frequencies up to each value or interval, which is meaningful only when the values or categories have a natural order).

Key features revealed by frequency distributions include modality (unimodal, bimodal, or multimodal), symmetry versus skewness (positive skew if the tail extends to the right, negative skew if to the left), kurtosis (how heavy the tails are relative to a normal distribution, that is, how prone the data are to extreme values; despite a common description, it does not measure how peaked the distribution is), and the presence of gaps or outliers.^[1][3] These features inform decisions about data transformation, the choice between parametric and non parametric tests, and the interpretation of summary statistics (the mean is sensitive to skewness and outliers, while the median is more resistant to these effects).^[1]

Frequency distributions should be constructed as one of the first steps in any data analysis, whether for a survey, experiment, observational study, or secondary data analysis. They are essential for data cleaning (identifying implausible values), exploratory data analysis (understanding distributional shape), assumption checking (checking for approximate normality before parametric analysis), and communicating basic findings to non technical audiences.^[2] They are appropriate for any type of variable (nominal, ordinal, interval, or ratio) with suitable adjustments in format.

A survey of 1,200 university students asks how many hours per week they spend studying. The raw data range from 0 to 52 hours. The frequency distribution with 5 hour intervals shows: 0 to 4 hours (8%), 5 to 9 hours (15%), 10 to 14 hours (28%), 15 to 19 hours (22%), 20 to 24 hours (14%), 25 to 29 hours (7%), 30 to 34 hours (4%), and 35 or more hours (2%). The distribution is unimodal with a peak at 10 to 14 hours and a positive skew. Two students report more than 48 hours, which the research team flags as possible data entry errors for verification.