Exploring Chi-Square as a Test of Independence Through Contingency Tables

When researchers want to know whether two categorical variables influence each other – say, whether a student’s attitude toward education is connected to their preference for online versus in-person learning – they need a statistical tool that can work with grouped, count-based data rather than numbers on a continuous scale. That tool is the Chi-square (χ²) test of independence. It is one of the most widely used tests in psychology, social science, and behavioral research precisely because so much of the data in these fields comes in categories, not measurements.

What does “independence” mean in statistics?
The contingency table: organizing the data
Observed vs. expected frequencies
Stating the hypotheses
Degrees of freedom and the critical value
A behavioral example: study preferences and academic attitude
What the test can and cannot tell you
Key assumptions to check
Why this test matters in psychological research

What does “independence” mean in statistics?

In everyday language, saying two things are independent means one has no bearing on the other. In statistics, the same logic applies. As the University of Utah’s sociology department explains, two variables are independent when an increase or pattern in one is not associated with any pattern in the other. The Chi-square test of independence is specifically designed to evaluate this: it checks whether the distribution of one categorical variable differs depending on the category of another. If it does, the variables are not independent – they are associated. If it doesn’t, they are statistically independent of each other.

This is different from correlation, which is used for continuous numerical data. With categorical or discrete data, correlation is not a suitable method for assessing relationships between variables. The Chi-square test fills that gap.

The contingency table: organizing the data

Before the Chi-square test can be applied, the data must be arranged in a specific format called a contingency table (also known as a cross-tabulation or two-way table). According to Kent State University’s SPSS tutorial library, a contingency table arranges data so that the categories of one variable appear in the rows and the categories of the other variable appear in the columns. Each cell in the table contains the count of observations that fall into that particular combination of categories.

For example, suppose a researcher surveys 200 students and wants to examine whether attitude toward education (positive or negative) is associated with their preferred learning format (traditional classroom or online). The contingency table might look like this:

	Traditional classroom	Online learning	Row total
Positive attitude	60	40	100
Negative attitude	30	70	100
Column total	90	110	200

These are the observed frequencies – the actual counts from the data collected. The Chi-square test then asks: if attitude and learning preference were truly unrelated, what counts would we expect to see in each cell?

Observed vs. expected frequencies

The heart of the Chi-square test is the comparison between what was observed and what would be expected under the assumption of independence. As Simply Psychology explains, expected frequencies are calculated by multiplying the row total and the column total for each cell, then dividing by the overall total sample size. This formula reflects what the data would look like if the two variables had absolutely no relationship.

Using the table above, the expected frequency for “positive attitude + traditional classroom” would be: (100 × 90) / 200 = 45. But the observed value is 60. That gap between 60 and 45 is exactly what the Chi-square formula measures and accumulates across all cells.

The formula for the Chi-square statistic is:

χ² = Σ [(O − E)² / E]

Where O is the observed frequency and E is the expected frequency. For every cell, you subtract the expected from the observed, square the result, divide by the expected frequency, and then sum all these values. The larger the resulting Chi-square value, the greater the discrepancy between observed and expected frequencies – and the stronger the evidence that the two variables are not independent.

Stating the hypotheses

Like all hypothesis tests, the Chi-square test of independence begins with two competing statements about the population. According to Statistics LibreTexts, the null hypothesis states that the two factors are independent of each other, while the alternative hypothesis states that they are not independent – meaning a real association exists.

For the education attitude and learning format example:

H₀ (Null hypothesis): There is no association between attitude toward education and preferred learning format. The two variables are independent.
H₁ (Alternative hypothesis): There is an association between attitude toward education and preferred learning format. The two variables are not independent.

The test is always one-tailed (right-tailed) because large Chi-square values – those that would result from big differences between observed and expected – always fall in the right tail of the Chi-square distribution. A small Chi-square value would mean the observed and expected counts are close, supporting independence.

Degrees of freedom and the critical value

Once the Chi-square statistic is calculated, researchers need to determine whether it is large enough to reject the null hypothesis. This involves two things: the degrees of freedom (df) and the significance level (usually α = 0.05).

Degrees of freedom for a Chi-square test of independence are calculated as:

df = (number of rows − 1) × (number of columns − 1)

For a 2×2 table like the one above, df = (2−1) × (2−1) = 1. Using the degrees of freedom and significance level, researchers look up the critical value in a Chi-square distribution table. According to StatTrek, if the calculated Chi-square statistic exceeds this critical value, the null hypothesis is rejected – meaning there is statistically significant evidence of an association between the two variables.

A behavioral example: study preferences and academic attitude

To make this more concrete, consider a psychology study examining whether students’ attitudes toward academic work (engaged vs. disengaged) are associated with their preferred study behavior (group study vs. studying alone). Researchers survey 150 students and record their responses in a contingency table. After calculating observed and expected frequencies and computing the Chi-square statistic, suppose they obtain χ²(1) = 8.3, p = 0.004.

Since the p-value (0.004) is below the 0.05 significance level, the researchers reject the null hypothesis. They conclude that attitude toward academic work and study behavior preference are not independent – there is a statistically significant association. Engaged students appear more likely to prefer group study, while disengaged students are more likely to prefer studying alone.

This is the kind of finding the Chi-square test is built for. As noted in a peer-reviewed article published in Biochemia Medica, the Chi-square test provides detailed information about how each group performed, giving researchers richer insight than many other non-parametric tools. Unlike correlation or regression, it doesn’t require the data to follow a normal distribution or meet assumptions about variance.

What the test can and cannot tell you

It is important to be clear about what a significant Chi-square result actually means – and what it doesn’t. The test tells you that an association exists between two categorical variables, but it does not indicate which variable influences the other, how strong that association is, or whether a causal relationship is at play. As GraphPad’s statistics guide warns, assuming that a statistically significant result implies causation is one of the biggest mistakes researchers make when interpreting contingency tables.

To measure the strength of the association – not just its statistical significance – researchers use additional measures such as Cramér’s V or the phi coefficient (φ). These effect size measures provide context that the Chi-square statistic alone cannot offer. A large sample, for instance, can produce a statistically significant Chi-square result even when the actual association between variables is trivially small.

Key assumptions to check

The Chi-square test of independence works best when certain conditions are met. According to BachelorPrint’s statistics guide, the test requires that the data consist of raw frequency counts (not percentages), that the variables are categorical, and that observations are independent – meaning each participant appears in only one cell of the table. Additionally, expected frequencies in each cell should generally be at least 5. When this condition is violated, alternatives such as Fisher’s exact test are more appropriate, especially for small sample sizes.

Why this test matters in psychological research

Much of the data collected in psychology and behavioral sciences is categorical by nature. Researchers classify people by their attitudes, preferences, diagnostic categories, group memberships, or behavioral choices. As the SAGE Encyclopedia of Survey Research Methods explains, the Chi-square test represents the most straightforward significance test for examining whether a meaningful relationship exists between two such categorical variables – making it the first step in many forms of bivariate analysis in social and psychological research.

Whether investigating whether therapeutic preferences differ across age groups, whether gender is linked to specific behavioral choices, or whether attitudes toward education predict engagement patterns, the Chi-square test of independence gives researchers a statistically grounded way to move beyond guesswork. It turns frequency counts in a table into a testable conclusion about the real world.

What do you think? If a study finds a significant Chi-square result showing that positive attitudes toward education are associated with online learning preferences, what additional information would you need before drawing any practical conclusions? And how might the size of the sample affect your confidence in that result?

References

PDF 📄

Table of Contents

What does “independence” mean in statistics? 🔗

The contingency table: organizing the data 🔗

Observed vs. expected frequencies 🔗

Stating the hypotheses 🔗

Degrees of freedom and the critical value 🔗

A behavioral example: study preferences and academic attitude 🔗

What the test can and cannot tell you 🔗

Key assumptions to check 🔗

Why this test matters in psychological research 🔗