Does analyzing one outcome guarantee there is only one statistical test?

No; multiple contrasts and subgroups can still generate many tests.

AI Glossary / Statistics and probability / Foundation

Univariate design

Definition

A univariate design is an experimental design with one dependent variable or outcome. It can still include several independent variables, multiple treatment conditions and many observations. The term describes the number of outcomes being analyzed, not the number of predictors or experimental factors. For example, a study of how training and software version affect one completion-time outcome is univariate even though it has two factors.

What it means

The outcome is the quantity the study aims to explain, compare or predict. Calling a design univariate says that one such response is the focus. It does not specify whether the response is continuous, binary or a count, nor whether observations are independent or repeated. Those features determine suitable methods. The term also differs from univariable regression, which often refers to a model with one predictor. Precise language prevents a study with many predictors and one outcome from being mistaken for a multivariate-outcome analysis.

How it works

Define the outcome and its measurement rule, then specify factors, assignment, experimental units and observation timing. Choose an analysis appropriate to the response and dependence structure: a continuous outcome may use a linear model, while a binary outcome may require a different model. Multiple factors can contribute main effects and interactions without creating additional dependent variables. If several outcomes are analyzed jointly, the analysis becomes multivariate in the response sense. Repeated measurements of the same outcome require particular care because different analytical formulations can treat them as correlated repeated responses or as a vector of outcomes.

Three practical examples

Example 1: A two-group AI workflow experiment measures only task completion time. Example 2: A factorial study varies training and interface version while measuring one accuracy score. Example 3: A production experiment changes temperature and machine settings while recording one defect-count outcome. Each has one dependent variable, but their predictors, measurement scales and suitable statistical models differ. Calling them all univariate does not make their designs or assumptions interchangeable.

Think of it this way

Think of a dashboard with one main output gauge and several control knobs. Changing several knobs does not add more output gauges. A univariate design can therefore study a complex set of influences on one response. It still needs a clear explanation of which knob combinations are compared and how reliably the gauge measures the intended result.

Worked example

Consider a two-by-two experiment with training level basic or enhanced and interface version A or B. The sole outcome is tasks completed correctly in a fixed session. Suppose cell means are eight for basic/A, ten for basic/B, eleven for enhanced/A and fifteen for enhanced/B. The enhanced-minus-basic difference is three tasks under A and five under B. Their difference, two tasks, is an interaction contrast: the training difference depends on the interface. With twelve independently assigned participants in each cell, the study contains 48 participants and one outcome per participant. It is univariate despite having two factors and four conditions. The cell means alone do not provide the within-cell variability needed to calculate uncertainty or declare the interaction statistically significant.

Illustrative case study

In this fictional case study, Cypress Learning evaluates an AI training tool. The team initially describes its two-factor experiment as multivariate because it changes both tutorial style and interface layout. A review clarifies that the experiment measures only one dependent variable, a prespecified task score. The design is therefore univariate in its outcome structure. Analysts keep the two factors and their interaction in the analysis plan while documenting the score's limits and the independence assumptions behind their comparisons.

Common misconceptions and mistakes

Common mistakes include equating univariate with one predictor, assuming one outcome eliminates all multiple-testing issues, and ignoring repeated observations from the same participant. A study can make many contrasts or subgroup comparisons on one outcome, creating multiplicity concerns. Another mistake is believing the label guarantees a simple or causal analysis. Causal interpretation depends on assignment and other design assumptions; complexity can arise from clustering, missingness, nonlinear effects or interactions even with a single response.

Business use

Businesses can use the distinction when commissioning experiments and reviewing analysis plans. Identify the primary outcome, clarify whether other measures are secondary or jointly analyzed, and preserve the link between the outcome and the decision. One well-defined response can make interpretation clearer, but it may omit important tradeoffs such as quality versus speed. If several outcomes matter, acknowledge them and plan their analysis rather than selecting the most favorable result afterward. The number of outcomes is one design feature among many.

Check your understanding

How many dependent variables appear in the worked example?

Answer: One, the task score.

How many experimental factors are there?

Answer: Two.

What is the interaction contrast?

Answer: Five minus three, or two tasks.

Can significance be determined from the four means alone?

Answer: No; information about variation and the sampling or assignment design is also needed.

Related terms

Independent variable
Factorial design
Related terms include dependent variable
interaction
multivariate analysis and repeated-measures design

Learn next

Learn next how a factorial linear model represents main effects and interactions for one continuous outcome. Then compare that setup with joint analysis of several outcomes and with repeated measurements. Practice separating the number of factors, response variables, experimental conditions and observational units in a study description.

Linear model

Sources

Source: University of Vermont, Statistical Methods for Psychology glossary, Univariate design, https://www.uvm.edu/~statdhtx/methods8/Glossary/Glossary.html. This primary teaching source defines the term by the number of dependent variables. The factorial task-score example and interaction arithmetic are original and independently checked. The training-tool case is fictional, and no significance or causal-effect claim is inferred from the illustrative means alone.

Previous: Univariate analysis Next: Univariate feature selection