AI Glossary / Statistics and probability / Beginner
Cell
Definition
In statistics, a cell is a particular combination of categories or factor levels, commonly represented by the intersection of a row and a column. A cell can contain a count, a summary or a set of observations collected under the same treatment conditions. Its meaning comes from the table or study design, not merely its position on a screen.
What it means
In a contingency table, a cell identifies observations sharing a combination of categorical values, such as “mobile device” and “purchase completed.” In a factorial experiment, it identifies a treatment combination, such as one interface version under one task difficulty. A cell mean summarizes the outcomes observed in that combination, while a cell count gives the number of observations. These are different quantities. The statistical meaning is related to, but more specific than, an arbitrary spreadsheet cell containing text or a formula.
How it works
Define the categories or factor levels forming each dimension. Assign observations according to explicit rules, then calculate the required count or summary for each combination. In a two-factor design with a levels of one factor and b levels of another, the complete crossed layout has a × b cells. More factors create higher-dimensional combinations even when displayed across several tables. Keep cell sizes with reported means so readers can assess how much evidence supports each summary. Distinguish an observed zero from missing data, an unmeasured combination and a combination that is impossible by definition.
Three practical examples
- Conversion reporting: count completed and incomplete purchases separately for each device category and calculate rates using the relevant denominators.
- Experimental design: allocate independent participants across combinations of interface version and task difficulty, recording the sample size in each cell.
- AI evaluation: summarize performance by language and document type, checking whether small or empty cells conceal gaps in the evaluation data.
Think of it this way
Think of labeled compartments in an organizer. The labels describe which observations belong together, while the contents tell you how many observations or which measurements are present. An empty compartment may mean nothing was observed, nothing was collected or the combination does not exist; those meanings must be distinguished.
Worked example
Consider a two-by-two experiment with interfaces A and B and tasks classified as easy or hard. The A-easy cell has 20 independent participants with mean score 90; A-hard has five with mean 50. The B-easy cell has five participants with mean 94; B-hard has 20 with mean 60. There are four cells and 50 participants. Within easy tasks, B exceeds A by four points; within hard tasks, B exceeds A by ten. But pooling participants gives A an overall mean of (20 × 90 + 5 × 50)/25 = 82 and B a mean of (5 × 94 + 20 × 60)/25 = 66.8. The pooled comparison reverses direction because task difficulty is distributed differently across interfaces. Giving easy and hard cells equal weights yields A = (90 + 50)/2 = 70 and B = (94 + 60)/2 = 77. Neither weighting choice should be hidden: each answers a different descriptive question. These hypothetical summaries do not provide sampling uncertainty or establish a causal treatment effect.
Illustrative case study
Fictional case: Willow Analytics compares two document-classification tools across language and document type. The overall accuracy favors one tool, but a cell-level report reveals that the tools were tested on different mixes of documents. The team creates a matched evaluation set, retains cell counts and identifies combinations with too little evidence for a reliable comparison. It avoids reporting a blank cell as zero accuracy. This fictional case illustrates evaluation design and does not claim that either tool is superior.
Common misconceptions and mistakes
A cell is not necessarily one observation. It may summarize many independent units or many repeated measurements from a few units, which have different inferential implications. A zero count is not automatically a structural zero; the combination may be possible but absent from the sample. Row and column totals are marginal summaries, not additional independent cells. Averaging cell means without considering weights can answer a different question from averaging all observations. Small cell sizes can make estimated rates unstable, and a large overall sample does not guarantee adequate evidence in every subgroup.
Business use
Businesses should report cell definitions and denominators whenever tables support comparisons. In dashboards, make clear whether values are counts, percentages, means or model estimates, and state whether percentages use row, column or grand totals. Inspect sparse combinations before drawing subgroup conclusions or training models that rely on them. Where data are sensitive, apply appropriate disclosure controls without confusing suppressed values with zeros. A table should help readers understand the structure of the evidence rather than conceal it behind a single overall number.
People also ask
Does a two-by-three table have five cells?
No. It has six interior category combinations: two multiplied by three, excluding marginal totals.
Is a cell mean the same as a row mean?
No. A row can combine several cells, and its mean depends on how their observations or summaries are weighted.
Can a study have cells without observations?
Yes. Their interpretation depends on whether the combination was unobserved, omitted or impossible.
Check your understanding
Q: Why does interface A have the higher pooled mean in the worked example despite lower means within both task categories?
Answer: A: Most A participants received easy tasks, while most B participants received hard tasks. The unequal cell sizes change the mixture being compared.
Related terms
- Contingency table
- Factorial design
- Marginal totals
- Sample size
- cell frequency
- cell mean
- marginal total
- structural zero
- interaction and Simpson's paradox
Learn next
rebuild the worked example with equal cell sizes and compare pooled and equally weighted means. Then construct a two-way count table and calculate one cell's row percentage, column percentage and overall percentage, labeling the denominator each time.
More in Statistics and probability
- Case-control study - A case-control study selects people or other units according to whether they have an outcome of interest, then compares their prior exposures or...
- Categorical Naive Bayes - Categorical Naive Bayes is a classification method that models each input feature as a categorical variable and combines its class-conditional probabilities...
- Cause-effect diagram - A cause-effect diagram is a visual tool for organizing possible contributors to a clearly defined problem or outcome.
- Censoring - Censoring occurs when an observation provides only partial information about a value, such as knowing that an event time exceeds a recorded follow-up time.
- Census - A census is a statistical collection that seeks information from every unit in a defined population, rather than selecting only a sample of those units.
- Centering - Centering transforms a variable by subtracting a reference value from each observation, most commonly the variable's mean.
Sources
Sources: Howell, Statistical Methods for Psychology glossary, cell definition: https://www.uvm.edu/~statdhtx/methods8/Glossary/Glossary.html ; Penn State STAT 503, factorial designs and cell means: https://online.stat.psu.edu/stat503/Lesson05 ; Penn State STAT 504, two-way tables and cell counts: https://online.stat.psu.edu/stat504/Lesson03 .
Previous: CDR coding Next: Cell decomposition