How to Easily Perform a Chi-Square Test by Hand: A Step-by-Step Guide

Observed Frequencies from 60 Rolls

• 1: 8 times
• 2: 12 times
• 3: 18 times
• 4: 9 times
• 5: 7 times
• 6: 6 times

Step 1: Defining the Null and Alternative Hypotheses

The cornerstone of any inferential statistical test is the precise definition of the Null Hypothesis ($H_0$) and the Alternative Hypothesis ($H_1$). These statements represent the two competing claims about the population from which the data was drawn. The Null Hypothesis always represents the status quo, the assumption of no effect, or in this case, the assumption that the observed distribution is consistent with the hypothesized uniform distribution.

In the context of the die experiment, the hypotheses are formally stated as follows, establishing the framework within which our test statistic will be interpreted. We are essentially testing the proposition that all outcomes are equally probable versus the claim that some bias exists, causing unequal likelihoods among the outcomes. The test aims to find evidence against $H_0$.

• H₀ (Null Hypothesis): The die is fair; the probability of landing on any specific number (1 through 6) is equal ($P(1) = P(2) = dots = P(6) = 1/6$). The distribution of observed frequencies fits the hypothesized uniform distribution.
• H₁ (Alternative Hypothesis): The die is unfair; the probability distribution is not uniform, meaning the die is not equally likely to land on each number. The distribution of observed frequencies does not fit the hypothesized distribution.

The entire purpose of calculating the Chi-Square statistic is to determine if we have enough statistical evidence—meaning a large enough deviation from expectation—to reject $H_0$. If the evidence is weak (the calculated statistic is small), we fail to reject $H_0$, concluding that the observed deviations are likely due to random chance rather than inherent unfairness in the die.

Step 2: Calculating the Observed and Expected Frequencies

Before proceeding with the main calculation, we must organize and explicitly define our Observed Frequencies ($O$) and our Expected Frequencies ($E$). The Observed Frequencies are simply the counts recorded during the experiment, as listed previously. The Expected Frequencies represent the theoretical counts under the assumption that the Null Hypothesis ($H_0$) is perfectly true. Calculating these expected values accurately is critical for setting the baseline of comparison.

Since $H_0$ states that the die is perfectly fair and we rolled it 60 times, the expected count for any single outcome is calculated by multiplying the total number of trials by the theoretical probability for that outcome. For a fair six-sided die, the probability of rolling any single number is 1/6. Therefore, the expected frequency for each category is $mathbf{60 times (1/6) = 10}$.

We compile these figures into a comparative table, which is shown below. This visual organization is critical for the next step, as the Chi-Square formula relies on calculating the difference between the Observed and Expected counts for every category. Note: If we believe the dice is fair, this means we expect it to land on each number an equal amount of times—in this specific case, 10 times each.

Step 3: Calculation of the Chi-Square Test Statistic

The Chi-Square statistic, denoted as $X^2$ or $chi^2$, is the quantitative measure of the total disparity between the observed data and the expected model. It is calculated by summing the standardized, squared differences across all categories. The mathematical definition ensures that larger differences contribute more significantly to the final value, and the squaring ensures that deviations below the expectation do not cancel out deviations above the expectation.

The formal formula for the Chi-Square test statistic ($X^2$) for a Goodness of Fit test is:

• $X^2 = Sigma frac{(O – E)^2}{E}$

We calculate this step-by-step for each category (side 1 through 6). This process involves three internal steps for each row: finding the raw difference $(O – E)$, squaring that difference $(O – E)^2$, and finally dividing the squared difference by the Expected frequency $E$. For instance, for the outcome ‘3’ (Observed=18, Expected=10), the calculation is: $(18 – 10)^2 / 10 = (8)^2 / 10 = 64 / 10 = 6.4$. We repeat this process for all six categories and then sum the resulting standardized values.

The table below details the calculation process, showing the difference, the squared difference, and the final standardized contribution $frac{(O – E)^2}{E}$ for each outcome. The summation of the final column yields our calculated test statistic.

After summing the standardized differences across all categories, we find that the total Chi-Square test statistic, $X^2$, equals $mathbf{9.8}$. This calculated value quantifies the cumulative deviation observed in our 60 rolls relative to what was theoretically expected under the assumption of a fair die. This is the crucial number we must now compare against a theoretical distribution to determine its statistical significance.

Step 4: Determining Degrees of Freedom and the Critical Value

To properly interpret the calculated $X^2$ value, we must establish the appropriate critical threshold. This threshold depends on two parameters: the chosen significance level ($alpha$) and the degrees of freedom ($df$). The significance level, typically set at $alpha = 0.05$ (or 5%), represents the maximum probability of making a Type I error—the error of incorrectly rejecting the true Null Hypothesis—that we are willing to accept.

The degrees of freedom specify which specific Chi-Square distribution curve we must use for comparison. In a Goodness of Fit test, the $df$ is calculated as the number of categories ($k$) minus one ($df = k – 1$). This formula is used because, given the fixed total sample size, once we know the counts of $k-1$ categories, the count of the final category is automatically determined. Since our experiment has 6 categories (the numbers 1 through 6), the calculation is $df = 6 – 1 = mathbf{5}$.

Using the standard Chi-Square distribution table, we look up the value corresponding to $df = 5$ and a significance level of $alpha = 0.05$. The critical value defines the boundary of the rejection region. If our calculated $X^2$ falls into this rejection region (i.e., is larger than the critical value), the result is deemed statistically significant. Consulting the table reveals that for $df=5$ and $alpha=0.05$, the critical value is $mathbf{11.07}$.

Step 5: Decision Making and Interpreting the Results

The final and most crucial step is comparing the calculated Test Statistic ($X^2$) with the Critical Value ($X^2_{text{critical}}$). This comparison dictates the statistical decision regarding the Null Hypothesis ($H_0$).

We found the following key values:

1. Calculated Test Statistic ($X^2$): $mathbf{9.8}$
2. Critical Value ($X^2_{text{critical}}$ for $df=5, alpha=0.05$): $mathbf{11.07}$

The decision rule is straightforward: If the calculated $X^2$ is greater than the critical value, we reject $H_0$. If the calculated $X^2$ is less than or equal to the critical value, we fail to reject $H_0$. In our case, $9.8$ is clearly less than $11.07$. Therefore, we fail to reject the Null Hypothesis.

Failing to reject the Null Hypothesis means that the difference observed between the actual observed frequencies and the expected frequencies is not statistically large enough to conclude, with 95% confidence, that the die is unfair. We conclude that we do not have sufficient evidence to definitively state that the die is biased or that its probability distribution is non-uniform. The minor fluctuations observed are likely attributable to the natural randomness inherent in the sampling process, rather than systematic unfairness.

Summary of the Chi-Square Goodness of Fit Test

The execution of the Chi-Square test by hand reinforces the understanding of how statistical significance is determined by linking observed reality to theoretical probability. The entire process ensures a rigorous quantitative assessment of whether categorical data aligns with a theoretical expectation. The calculated test statistic acts as a measure of accumulated deviation, while the critical value establishes the statistical threshold for making claims about the underlying population distribution.

In conclusion, the Chi-Square goodness of fit test is an invaluable tool for researchers across various disciplines seeking to compare frequency data. By following these methodical steps—from setting clear hypotheses and determining expected values to calculating the Chi-Square statistic and making a formal decision—one can draw reliable conclusions about the distribution of observational data relative to theory.