Math AI HL: Chi-Squared Test of Independence

When you suspect a relationship between two categorical variables—like whether a person's preferred social media platform is independent of their age group—you need a tool to test that suspicion statistically. The chi-squared test of independence is that tool, allowing you to move from a hunch to a mathematically sound conclusion. Mastering it is crucial for the IB Math AI HL course, as it connects foundational probability with real-world data analysis, forming a bridge between pure math and applied research in fields from sociology to medicine.

Contingency Tables and Observed Frequencies

Every chi-squared test begins with organizing your raw data into a contingency table (also called a two-way table). This table displays the observed frequencies ($O_{ij}$), which are the actual counts of data points falling into each combination of categories for your two variables. For instance, if you are studying the relationship between exercise frequency (Low, Medium, High) and sleep quality (Poor, Good), your contingency table would have 3 rows and 2 columns. Each cell contains the number of individuals in that specific combination. This table is your empirical starting point; the entire test investigates whether the patterns in this table are likely due to chance or indicate a genuine association. Constructing it correctly is the first critical step, as all subsequent calculations depend on its values.

Formulating Hypotheses and Calculating Expected Frequencies

The next step is to state your hypotheses formally. The null hypothesis ($H_0$) always states that the two variables are independent. There is no association between them. The alternative hypothesis ($H_1$) states that the two variables are not independent; an association does exist. You never specify the direction of the association in this test.

To test $H_0$, you must determine what the data would look like if the null hypothesis were true. This involves calculating expected frequencies ($E_{ij}$) for every cell in your contingency table. The expected frequency is the count you would anticipate if the variables were perfectly independent. The formula for the expected frequency of the cell in row $i$ and column $j$ is:

$$E_{ij}=\frac{(\text{Row }i\text{ Total})\times (\text{Column }j\text{ Total})}{\text{Grand Total}}$$

This calculation is intuitive: it proportions the grand total according to the marginal totals. If 40% of all subjects are in Row 1 and 30% are in Column 2, and if the variables are independent, then you'd expect 40% × 30% = 12% of the grand total to be in cell (1,2). Let's apply this. Suppose you have a 2x2 table with row totals of 80 and 120, and column totals of 100 and 100 (Grand Total = 200). The expected frequency for the top-left cell would be $E_{11}=(80\times 100)/200=40$.

The Chi-Squared Test Statistic and Degrees of Freedom

With observed ($O$) and expected ($E$) frequencies in hand, you calculate a single number that quantifies the overall discrepancy between them: the chi-squared test statistic (${\chi }^2$). The formula is:

$${\chi }^2=\sum \frac{(O_{ij}-E_{ij}{)}^2}{E_{ij}}$$

You compute $(O-E{)}^2/E$ for every cell in the table and then sum all these values. This statistic follows a predictable pattern: it is always zero or positive. A value of zero would mean observed counts perfectly match expected counts, providing strong support for independence. Larger values indicate greater divergence from the independence model, thus providing evidence against the null hypothesis.

However, you cannot interpret the ${\chi }^2$ value in isolation. You must consider the degrees of freedom (df), which depends on the table's dimensions and affects the shape of the chi-squared distribution. For a contingency table with $r$ rows and $c$ columns, the formula is:

$$\text{df}=(r-1)\times (c-1)$$

For a 3x4 table, df = (3-1)×(4-1) = 2×3 = 6. The degrees of freedom are crucial because they determine which chi-squared distribution you use to find your p-value or critical value.

Making a Decision: Critical Values and P-Values

To make a formal decision on your hypotheses, you compare your calculated ${\chi }^2$ statistic to a benchmark. There are two equivalent approaches, both requiring the significance level ($\alpha$, often 0.05) and your calculated degrees of freedom.

1. Critical Value Method: Using a chi-squared distribution table, find the critical value for your df and $\alpha$. If your calculated ${\chi }^2$ statistic is greater than the critical value, you reject the null hypothesis. This means the observed discrepancy is too large to be attributed to chance alone at your chosen significance level.
2. P-Value Method: Use your GDC or statistical software to find the p-value associated with your ${\chi }^2$ statistic and df. The p-value is the probability of obtaining a test statistic at least as extreme as the one you calculated, assuming the null hypothesis is true. If the p-value is less than $\alpha$, you reject the null hypothesis.

For IB exam purposes, you must be proficient with both methods. Your GDC can perform the entire test, but you must correctly input the observed contingency table and interpret the output, which includes the ${\chi }^2$ statistic, p-value, and sometimes the calculated expected frequencies.

Interpretation in Context and Conditions for Use

A statistically significant result is not the end of the analysis. You must interpret the result in the context of the original problem. "Reject $H_0$" translates to: "There is sufficient evidence at the 5% significance level to conclude that [Variable A] and [Variable B] are not independent (i.e., there is an association between them)." Conversely, failing to reject $H_0$ means there is insufficient evidence to claim an association; it does not prove the variables are independent.

This test is only valid when specific conditions are met:

• Independence of Observations: Each data point must belong to only one cell and be independent of others (often achieved by random sampling).
• Sample Size: All expected frequencies ($E_{ij}$) must be 5 or greater. This is a non-negotiable rule of thumb for the chi-squared approximation to be valid. If this condition is not met, you may need to combine adjacent categories to increase the expected counts.

Common Pitfalls

1. Misapplying the Test for Independence: This test is exclusively for two categorical variables. A common mistake is trying to use it for numerical data or to test for "goodness-of-fit" (a different chi-squared test) without adjustment. Always confirm your variables are categorical and you are specifically testing for an association between them.

1. Ignoring the Expected Frequency Condition: Calculating a ${\chi }^2$ value when one or more expected frequencies are below 5 invalidates the test. Always calculate and check all $E_{ij}$ before proceeding to interpret the test statistic. On an exam, stating this check is often required for full marks.

1. Confusing Statistical Significance with Practical Importance: A significant p-value indicates an association is unlikely to be due to random chance in your sample. It does not mean the association is strong or causally important. Always pair the statistical conclusion with a comment on the practical meaning, perhaps by looking for which cells have the largest contributions to the ${\chi }^2$ statistic.

1. Incorrect Degrees of Freedom Calculation: Using the formula $\text{df}=r\times c$ instead of $(r-1)(c-1)$ is a critical error that will lead to using the wrong distribution for comparison. The correction accounts for the constraints imposed by the fixed row and column totals.

Summary

• The chi-squared test of independence assesses whether two categorical variables are associated, using observed frequencies from a contingency table.
• The test compares observed frequencies ($O_{ij}$) with expected frequencies ($E_{ij}$) calculated under the assumption of independence: $E_{ij}=(\text{Row Total}\times \text{Column Total})/\text{Grand Total}$.
• The chi-squared test statistic ${\chi }^2=\sum (O_{ij}-E_{ij}{)}^2/E_{ij}$ measures the total discrepancy from independence. Its interpretation depends on the degrees of freedom, calculated as $\text{df}=(r-1)(c-1)$.
• You reject the null hypothesis of independence if the test statistic exceeds the critical value or if the p-value is less than the significance level $\alpha$.
• A valid test requires independent observations and that all expected frequencies are at least 5. Always conclude by stating your decision in the context of the original research question.