How to Compare Two Proportions Using a Z-Test in SPSS

A two proportion z-test is specifically utilized to determine if there is a statistically significant disparity between two independent population proportions (p1 and p2). Although SPSS does not offer a distinct module explicitly labeled “Two Proportion Z-Test,” the most accessible and reliable method for generating the necessary output involves the utilization of the Analyze > Descriptive Statistics > Crosstabs function. This method relies on the mathematical equivalence between the squared Z-statistic and the Chi-square statistic with one degree of freedom, particularly effective when dealing with large sample sizes typically encountered in proportion testing.

Understanding the Two Proportion Z-Test

The core objective of the two proportion Z-test is to evaluate the relationship between two categorical variables, where one variable defines the two independent groups being compared (e.g., Class A vs. Class B), and the second variable represents the binary outcome of interest (e.g., Success vs. Failure, or Prefer vs. Do Not Prefer). This test is predicated on certain critical assumptions, primarily that the samples are independent, the data is derived from a binomial distribution, and that the sample sizes are large enough to ensure that the sampling distribution of the difference in proportions approximates a normal distribution.

Statistically, the test works by calculating a pooled estimate of the overall success rate across both samples and then using this estimate to compute the Z-statistic. This calculated Z-statistic measures how many standard errors the observed difference in sample proportions is away from zero (the hypothesized difference). Crucially, the outcome of this Z-test—specifically its associated p-value—is mathematically identical to the outcome derived from the Pearson Chi-square test applied to a 2×2 contingency table, provided that the significance level is interpreted carefully based on whether a one-tailed or two-tailed test is required.

When conducting this test, researchers must ensure their raw data is summarized into counts, which is the input format required for the Crosstabs procedure in SPSS. If the data is provided in a disaggregated format (one row per participant), it must first be aggregated into group counts. Since SPSS expects the data structure to represent the summary statistics—the number of successes and failures within each group—the subsequent steps will involve a mandatory process of weighting cases to ensure the software correctly treats the counts as the actual frequency of observations.

Prerequisites and Data Preparation in SPSS

Effective execution of any hypothesis test in SPSS begins with meticulous data structure setup. For a two proportion Z-test, the data must be organized into a format suitable for frequency analysis. Instead of the typical structure where each row represents a single case, we use a summarized structure with variables representing the group identifier, the outcome preference, and the frequency count. This compact format minimizes the necessary data entry and facilitates the weighting process crucial for accurate results.

For this specific analysis, three variables are essential in the Data View of SPSS: the group variable (e.g., Class), the outcome variable (e.g., Preference), and the count variable (e.g., Count). It is vital that both the group and outcome variables are defined as nominal or categorical variables in the Variable View. The count variable, however, must contain the actual numerical frequencies reflecting how often each combination of group and outcome occurred.

The subsequent step, known as weighting cases, is non-negotiable when using summarized data. If this step is overlooked, SPSS will treat each row in the Data View as a single observation (N=4 in our typical 2×2 setup) rather than recognizing the actual sample size (N=60 in our example). The weighting procedure tells the software to use the numerical value in the ‘Count’ variable as the frequency weight for the row, thereby restoring the true sample sizes required for calculating correct degrees of freedom and accurate test statistics.

Step-by-Step Example Scenario

To illustrate the application of the two proportion Z-test, consider a common educational research scenario. Suppose a university professor manages two large sections of the same introductory statistics course, Class A and Class B. The teacher suspects that the proportion of her students who prefer a specific teaching method is equal across both classes. This forms the basis of our null hypothesis (H₀).

To test this assumption, she surveys the 30 students in each class, asking a simple binary question. The hypothesis test will seek to determine if the observed difference in preference rates between Class A and Class B is substantial enough to achieve statistical significance, allowing us to reject the claim of equality. The results of the survey are summarized below:

• Class A Size: 30 students surveyed.
• Count in Class A who prefer method: 23 students.
• Class B Size: 30 students surveyed.
• Count in Class B who prefer method: 18 students.

From these counts, we determine the sample proportions: Class A preference is 76.7% (23/30), and Class B preference is 60.0% (18/30). The observed difference is 16.7%. The subsequent steps will demonstrate how to input this data and use SPSS to assess whether this difference is likely due to chance.

Phase 1: Entering and Weighting Data

Here is how to enter this summarized data into SPSS. Remember that we must input the counts for both preference categories (Prefer and Do Not Prefer) for each class:

Before we perform the two proportion z-test to determine if the two proportions are equal, we must first weight the cases to correctly reflect the total sample size of 60. To do so, click the Data tab, then click Weight Cases.

In the new window that appears, drag the Count variable into the Frequency variable panel. This instructs SPSS to treat the numerical entries in the ‘Count’ column as the frequency of each row’s category combination:

Then click OK. The analysis will now proceed with the correct total sample size (N=60).

Phase 2: Executing the Crosstabs Procedure

With case weighting complete, we proceed to the statistical analysis. Click the Analyze tab, then click Descriptive Statistics, then click Crosstabs:

In the Crosstabs dialogue box, define the structure of the 2×2 contingency table. Drag the Class variable to the Rows panel and drag the Preference variable to the Columns panel:

To ensure SPSS calculates the statistic necessary for the proportion test, click the Statistics button. In the new window that appears, check the box next to Chi-square:

Then click Continue. Finally, click OK in the main Crosstabs window to generate the output tables.

Interpreting the SPSS Output

The generated output includes the critical table for interpretation: the “Chi-Square Tests” table. This table contains the necessary two-tailed p-value for the comparison of the two proportions.

The two-tailed p-value for the two proportion Z-test is shown in the final table of the output in the Pearson Chi-Square row, under the column Asymp. Sig. (2-sided). This value represents the likelihood of observing the difference in proportions if the null hypothesis were true.

In this particular example derived from the teacher’s survey data, the p-value for the two proportion Z-test is .165. This figure is used to decide whether the observed difference between the 76.7% preference rate in Class A and the 60.0% rate in Class B is statistically meaningful.

Formulating the Hypotheses and Conclusion

The two proportion Z-test relies on the following standard hypotheses:

• H0: The two population proportions are equal (P₁ = P₂).
• HA: The two population proportions are not equal (P₁ ≠ P₂).

To determine whether we reject or fail to reject the null hypothesis (H₀), we compare the calculated p-value (0.165) to the standard significance level (α = 0.05). Since our p-value (.165) is greater than .05, we fail to reject the null hypothesis.

This statistical decision leads to the conclusion that we do not have sufficient evidence to say that the proportion of students in each class who prefer the specific teaching method is significantly different. While there is a numerical difference in the sample proportions, that difference does not reach the threshold for statistical significance, suggesting it is likely due to random sampling fluctuation.

Conclusion and Further Resources

The two proportion Z-test, though implemented indirectly via the Crosstabs and Chi-square procedure in SPSS, remains a fundamental tool for comparing binary outcomes between two groups. Ensuring the data is correctly summarized and weighted is the critical prerequisite for obtaining accurate results from the Pearson Chi-square output.

Bonus: You can also use the arabpsychology Z-test calculator to perform this analysis externally for verification purposes.

The following tutorials explain how to perform other common statistical tasks in SPSS: