Frequency Table
A frequency table is a table that organises a dataset by listing each distinct value or class of values alongside the number of times it occurs, turning a long, unordered list of observations into a compact display that can be read at a glance. Its basic layout is two columns, category and frequency; richer versions add columns for relative frequency, cumulative frequency, and percentages. It is the working document of descriptive statistics: nearly every chart of counts is drawn from a frequency table, and nearly every analysis of categorical data begins by building one.
The anatomy of a good frequency table is simple, and each part has a job. Rows are the categories or classes, listed in a meaningful order — natural order for ordinal or numeric classes, descending frequency for nominal categories, so the most common stand out. The frequency column holds the counts and ends with a total row that must equal the sample size. A relative frequency column converts counts to shares, and for ordered classes a cumulative column adds the running total. Classes for numeric data must be mutually exclusive and exhaustive, so every observation lands in exactly one row; a good table states its boundary convention, such as lower edge included, to avoid ambiguity about whether a value of exactly 10 belongs to the class 0 to 10 or 10 to 20. Missing values deserve a row of their own rather than silent deletion, since the number of non-responses is itself information.
The table extends naturally to two variables, and this is where it becomes a working tool of inference: a two-way table, also called a cross-tabulation or contingency table, counts how often each combination of categories occurs. Karl Pearson introduced the term contingency table in the early 1900s in connection with measuring association between categorical variables, and his 1900 chi-square statistic, used to decide whether the observed counts in such a table differ from what independence would predict, remains the standard test. Row and column totals, called marginal frequencies, recover the one-way tables for each variable separately, while the cells show the joint pattern. Converting cells to row or column percentages answers different questions, such as the share of treated patients who recovered versus the share of the recovered who were treated, and choosing the wrong direction is a classic source of misreading.
In practice, frequency tables are everywhere, usually under software names. Spreadsheets build them with pivot tables, R with the table function, Python's pandas with value_counts and crosstab, SQL with GROUP BY and COUNT. In data science they are the first diagnostic of categorical features: a table reveals a category that appears once (probably a typo), inconsistent labels, a dominant class that makes a classifier trivial, or a suspiciously empty cell. In research reporting, the so-called Table 1 of nearly every clinical paper is a set of frequency tables describing the study groups, and in quality control the Pareto table, ranking defect types by frequency with cumulative percentages, directs effort to the few causes that explain most of the problems, following Joseph Juran's popularisation of the principle in the mid twentieth century.
The common pitfalls are about presentation rather than arithmetic. Percentages shown without the underlying counts make small samples look precise, so tables should carry both. Too many classes defeat the purpose of summarising, while a single catch-all "other" row that is the largest category hides what matters. Tables with sparse cells break the assumptions of the chi-square test, for which a common rule of thumb asks for expected counts of about five or more per cell, and Fisher's exact test is the usual alternative for small two-by-two tables. Collapsing a two-way table over a third variable can even reverse an association, the situation known as Simpson's paradox, which is why stratified tables are worth building.
A frequency table is the tabular presentation of a frequency distribution, not a separate concept from it. The distribution is the idea of how counts are spread; the table is its numeric display, the histogram or bar chart is its graphical display, and the pie chart is a graphical display of its relative frequencies for categories. Individual columns are the frequency, the relative frequency, and the cumulative frequency, each with its own entry; the table is simply where they sit together.
Read a frequency table as data compressed into counts without losing the story: every observation assigned to exactly one row, every column a different question answered about the same counts, and the totals that must reconcile as the quiet guarantee that nothing has been dropped or double-counted.