Can there be two modes?

Yes. Two values can tie for the largest frequency or probability.

Can nominal categories have a mode?

Yes, because only frequency comparisons are required.

Does a mode identify a typical continuous measurement exactly?

Usually not from raw repetition alone; estimating a density peak involves additional choices and uncertainty.

AI Glossary / Statistics and probability / Intermediate

Mode

Definition

The mode is a value or category with the greatest frequency in a dataset, or greatest probability in a discrete distribution. For a continuous distribution described by a density, a mode is a location where that density is highest. A mode need not be unique, and a sample's most repeated value should not be confused with an estimated peak of an underlying continuous distribution.

What it means

The mode answers a popularity question: which outcome appears most often? This makes it useful for categories such as preferred contact channel, product size, or issue type, where adding category labels to compute a mean would make no sense.
For numerical data, the most frequent value may be far from the mean or median. Calling it a measure of central tendency does not guarantee that it sits in the numerical center or describes a majority of observations.

How it works

For a finite dataset, count how often each distinct value occurs and identify the largest count. Report all ties or apply a clearly stated tie-handling convention. If every observation occurs once, many elementary summaries say there is no mode, while a literal frequency-maximizing definition ties every observed value. State the convention instead of implying a unique answer.
For discrete probability distributions, compare probability masses. For a continuous density, identify its maximum or maxima. Exact repeats are uncommon in unrounded continuous samples, so sample repetition is usually a poor way to estimate the density's peak.
Histograms and density estimates help describe continuous modes. Their apparent peaks depend on bin boundaries, bin widths, smoothing, and sample size. A modal interval is a range, not an exact observed value. Unequal-width histogram bins also require care: count and density describe different comparisons.
Data cleaning affects frequencies. Inconsistent spelling may split one category into several labels, while excessive rounding may create artificial numerical repetition. Missing-value codes should not silently become a substantive mode.
For weighted data, sum the relevant weights within each category rather than simply counting rows, when the objective is a weighted distribution. Explain whether the result describes raw observations or a population estimate.
In machine learning classification, choosing the category with the highest predicted conditional probability is a conditional-mode decision. That choice minimizes expected zero-one classification loss when the probabilities are correct and all mistakes carry equal cost. Different costs can justify another action.

Three practical examples

Customer service: identify the most common complaint category while also reporting its share and the next most frequent categories.
Inventory planning: summarize the most frequently requested size, then inspect the full demand distribution before ordering stock.
AI classification: select the highest-probability label under an equal-cost decision rule, while retaining the probability distribution for uncertainty assessment.

Think of it this way

Think of a shop's best-selling item. It can lead the sales ranking even when most customers collectively buy other items. Being first in the ranking does not make it the average purchase or a choice made by more than half of customers.

Worked example

Consider the original numerical dataset 1, 1, 1, 4, 4, 7, 10. Value 1 appears three times, value 4 twice, and the remaining values once. The unique sample mode is therefore 1.
The sum is 28, so the mean is 28/7 = 4. The fourth sorted observation is also 4, making the median 4. This example shows that the mode can differ from both other summaries.
Now consider 100 fictional contact preferences: chat receives 45 responses, email 35, and telephone 20. Chat is the modal category with 45% of responses, but it is not a majority preference.
If a model assigned those same probabilities to a new person's preference and mistakes had equal cost, always predicting chat would have expected error 1 - 0.45 = 0.55. A modal decision can thus be the best available choice under that loss while still being wrong more often than right.

Illustrative case study

Fictional case: Maple Homeware uses the modal requested size to plan an initial product display. Its analyst notices that the leading size accounts for less than half of orders and that regional patterns differ.
The team reports the full frequency table and separates raw sales from estimated demand affected by stockouts. It uses the mode as one descriptive fact rather than ordering every unit in the most popular size. The resulting plan acknowledges diversity that a single label would hide.

Common misconceptions and mistakes

The mode is not necessarily the middle value or the arithmetic average. It also need not represent more than half the observations.
Do not infer that a continuous population lacks a density mode merely because all sampled measurements are distinct. Conversely, a repeated rounded value can reflect the recording process rather than a genuine population peak.
Symmetry alone does not guarantee that mean, median, and mode coincide: a symmetric distribution can have two separated peaks. A distribution's visible peaks also need not correspond one-to-one with meaningful latent groups.
Avoid presenting a software-selected tie winner as the only mode without checking the program's convention.

Business use

Businesses use modes to summarize leading categories, default choices, common failure types, and popular configurations. Pair the result with counts, percentages, and subgroup information so decision-makers can assess its strength and relevance. Where mistakes have unequal consequences, such as prioritizing service incidents, the most common category may not be the most important one to address.

Check your understanding

Why is chat's 45% share a mode but not a majority?

Answer: It exceeds each alternative separately but not 50% of all responses.

What can change a histogram's apparent mode?

Answer: Bin widths, boundaries, data variation, and other choices involved in representing the distribution.

Related terms

Frequency distribution
Mean
Median
modal class
probability mass
probability density
multimodality
maximum a posteriori classification

Learn next

compare frequency tables with histograms and density estimates, then examine how unequal error costs can change a classification decision.

Sources

Sources: Australian Bureau of Statistics, Measures of central tendency: https://www.abs.gov.au/statistics/understanding-statistics/statistical-terms-and-concepts/measures-central-tendency . NIST/SEMATECH, Measures of Location: https://www.itl.nist.gov/div898/handbook/eda/section3/eda351.htm . Both pages were read. Continuous-density modes and symmetry qualifications are made explicit rather than extending simplified sample descriptions. Examples are original.

Previous: Modality Next: Mode collapse