Data Analytics with Statistical Software (COM3MN210) — Module 3: Statistical Analysis
Lecture Notes • Complete Study Material
In corporate analytics and empirical business research, managerial decisions frequently hinge upon comparing mean performance metrics across distinct operational environments, consumer segments, treatment conditions, or chronological periods. This module delivers an exhaustive guide to the parametric techniques governing mean comparisons in IBM SPSS. We systematically examine the complete family of Student's t-tests—including the One-Sample t-Test, the Paired-Samples (Dependent) t-Test, and the Independent-Samples (Two-Sample) t-Test—followed by the advanced partitioning of variance via One-Way Analysis of Variance (ANOVA) and Two-Way (Factorial) ANOVA, complete with post-hoc multiple comparison procedures and interaction modeling.
3.1 The Family of Student's t-Tests: Foundations & Assumptions
Origins and Theoretical Concept
The t-test was formulated in 1908 by William Sealy Gosset under the pseudonym "Student" while working as a chemist and quality control researcher for the Guinness Brewery in Dublin. Gosset recognized that when sample sizes are small (n < 30) and the true population standard deviation (σ) is unknown, the standard normal Z-distribution underestimates tail probabilities. The resulting Student's t-distribution features heavier tails that adjust dynamically based on the Degrees of Freedom (df), converging to the standard normal distribution as sample size approaches infinity.
Core Parametric Assumptions
Valid execution of t-tests in SPSS requires that the empirical data satisfy four core mathematical assumptions:
- Continuous Metric Scale: The dependent variable must be measured on an interval or ratio scale (Scale in SPSS).
- Random Sampling & Independence: Observations within each group must be independently sampled from the target population without mutual influence.
- Normality of Distribution: The dependent variable should follow an approximately normal distribution within each comparison group. In SPSS, normality is verified via the Shapiro-Wilk test (p > 0.05 indicates normality), skewness/kurtosis coefficients within [−1.0, +1.0], or visual inspection of Normal Q-Q plots. Note: By the Central Limit Theorem, if sample sizes exceed 30 cases per group, t-tests exhibit substantial robustness against mild departures from normality.
- Homogeneity of Variance: For independent two-sample tests, both groups should possess approximately equal population variances (σ₁² = σ₂²). SPSS automatically tests this using Levene's Test for Equality of Variances.
3.2 One-Sample t-Test in SPSS
Purpose and Business Applications
The One-Sample t-Test evaluates whether the sample mean (x̄) of a continuous metric differs significantly from a predetermined, specified benchmark, historical standard, or hypothesized population mean (μ₀).
Mathematical Formulation
t = (x̄ - μ₀) / (s / √n)
Degrees of Freedom (df): df = n - 1
Where: x̄ = sample mean, μ₀ = hypothesized test value, s = sample standard deviation, n = sample size, and (s / √n) = Standard Error of the Mean (SE).
SPSS Navigation & Output Interpretation
- One-Sample Statistics Table: Displays N, Mean, Std. Deviation, and Std. Error Mean.
- One-Sample Test Table:
- t column: The calculated t-statistic.
- df column: Degrees of freedom (n - 1).
- Sig. (2-tailed): The exact two-sided p-value. If Sig. < 0.05, reject H₀ and conclude that the sample mean differs significantly from the specified benchmark value.
- Mean Difference: (x̄ - μ₀).
- 95% Confidence Interval of the Difference: If the 95% CI does not span zero (both Lower and Upper bounds are either positive or negative), statistical significance at α = 0.05 is confirmed.
3.3 Independent-Samples (Two-Sample) t-Test in SPSS
Purpose and Research Design
The Independent-Samples t-Test evaluates whether the population means of two completely separate, independent groups differ significantly on a continuous dependent variable. Each participant contributes data to only one of the two groups.
- Comparing average monthly customer spend between Prime Members vs. Non-Prime Members.
- Comparing sales productivity between employees trained under Method A vs. Method B.
- Evaluating whether website session duration differs between Mobile App users and Desktop Web users.
Levene's Test for Equality of Variances
Before interpreting the t-test, SPSS conducts Levene's Test for Equality of Variances, which tests:
H₀: σ₁² = σ₂² (Variances are equal) vs. H₁: σ₁² ≠ σ₂² (Variances are unequal).
Case 1: Levene's Sig. > 0.05 (Homogeneity Holds)
Case 2: Levene's Sig. ≤ 0.05 (Heteroscedasticity)
SPSS Execution Protocol & Effect Size
Effect Size (Cohen's d): Measures the standardized difference between two means:d = (x̄₁ - x̄₂) / s_pooled.
Interpretation: d = 0.20 (Small effect), d = 0.50 (Medium effect), d = 0.80+ (Large effect).
3.4 Paired-Samples (Dependent / Repeated-Measures) t-Test in SPSS
Purpose and Experimental Design
The Paired-Samples t-Test evaluates mean differences across two related, matched, or repeated observations drawn from the same group of participants across two measurement occasions or experimental conditions.
- Pre-Test vs. Post-Test Studies: Assessing employee sales performance before and after a 30-day digital marketing training bootcamp.
- Matched-Pair Case Control: Comparing expenditure between two stores matched precisely on physical square footage and neighborhood demographics.
- Cross-Over Product Trials: Evaluating customer rating of Product A vs. Product B tested by the same consumer panel.
Mathematical Mechanics: Analyzing Difference Scores
For each subject i, compute the difference score: D_i = X_{post, i} - X_{pre, i}
t = D̄ / (s_D / √n)
Degrees of Freedom (df): df = n - 1 (where n is the number of pairs)
Where: D̄ = mean of the difference scores, and s_D = standard deviation of difference scores.
3.5 One-Way Analysis of Variance (ANOVA) in SPSS
The Rationale: Overcoming Type I Error Inflation
When a business researcher needs to compare the means of three or more independent groups (e.g., comparing mean customer satisfaction across North, South, East, and West branches), running multiple pairwise independent t-tests is statistically invalid.
If an analyst runs t-tests across k = 4 groups, there are 6 distinct pairwise comparisons: C(4, 2) = 6. The overall family-wise error rate (α_FW) inflates dramatically:α_FW = 1 - (1 - α)^c = 1 - (1 - 0.05)^6 ≈ 0.265 (26.5% probability of a false positive!).
One-Way ANOVA resolves this by performing a single, omnibus F-test that tests whether at least one group mean differs from the others, maintaining the overall Type I error rate strictly at α = 0.05.
Partitioning of Variance & The F-Ratio
Total Sum of Squares (SS_Total) = Sum of Squares Between Groups (SS_Between) + Sum of Squares Within Groups (SS_Within)
Mean Square Calculations:
MS_Between = SS_Between / (k - 1) (df_between = k - 1, where k = number of groups)
MS_Within = SS_Within / (N - k) (df_within = N - k, where N = total sample size)
The F-Statistic:
F = MS_Between / MS_Within
Logic: If group means differ significantly due to treatment effects, MS_Between will substantially exceed MS_Within (random individual variation), resulting in a large F-ratio and p < 0.05.
Post-Hoc Multiple Comparisons
An omnibus ANOVA F-test indicates only that at least one group differs, but does not identify which specific pairs differ. Analysts deploy Post-Hoc Tests to examine pairwise contrasts while controlling the family-wise error rate:
| Post-Hoc Test | Variance Assumption | Recommended Usage |
|---|---|---|
| Tukey's HSD (Honestly Significant Difference) | Equal variances assumed (Levene's p > 0.05) | The industry gold standard for pairwise comparisons when sample sizes across groups are approximately equal. |
| Bonferroni | Equal variances assumed | A conservative adjustment that divides α by the number of comparisons. Ideal when testing a pre-planned subset of hypotheses. |
| Games-Howell | Equal variances NOT assumed (Levene's p ≤ 0.05) | Robust non-parametric post-hoc test deployed when groups exhibit unequal sample sizes and heterogeneous variances. |
SPSS Navigation Workflow for One-Way ANOVA
3.6 Two-Way (Factorial) Analysis of Variance in SPSS
Factorial Designs & Interaction Effects
Two-Way ANOVA (General Linear Model → Univariate in SPSS) evaluates the simultaneous effects of two categorical independent factors (Factor A and Factor B) on a single continuous dependent variable.
Crucially, Factorial ANOVA evaluates three distinct hypotheses in a single analysis:
- Main Effect of Factor A: Does Factor A significantly influence the dependent variable, averaged across all levels of Factor B? (e.g., Does Store Location: Urban vs Rural affect revenue?)
- Main Effect of Factor B: Does Factor B significantly influence the dependent variable, averaged across all levels of Factor A? (e.g., Does Promotional Strategy: Discount vs BOGO affect revenue?)
- Interaction Effect (A × B): Does the effect of Factor A depend upon the specific level of Factor B? An interaction occurs when the impact of promotion depends critically upon whether the store is located in an urban or rural market.
- Locate the rows labeled: Factor A, Factor B, and Factor A * Factor B (the interaction).
- Examine the F and Sig. columns for each row.
- Priority Rule: If the Interaction Effect (Factor A * Factor B) is statistically significant (Sig. < 0.05), you must exercise caution when interpreting the Main Effects, because the effect of one factor is contingent upon the level of the other.
- Partial Eta Squared (η_p²): Quantifies the proportion of variance explained by each factor after controlling for other model terms. (η_p² = 0.01 small, 0.06 medium, 0.14+ large).
3.7 Comprehensive Review & Self-Assessment Exercises
- Under what operational circumstances would a researcher deploy a Paired-Samples t-Test rather than an Independent-Samples t-Test?
- Why is it statistically invalid to conduct multiple two-sample t-tests to compare five group means? Explain the concept of family-wise error rate inflation.
- What is the function of Levene's Test for Equality of Variances in an independent t-test output, and how does the analyst determine whether to read the top row or bottom row?
- Distinguish between a Main Effect and an Interaction Effect in a Two-Way Factorial ANOVA model.
- What does an omnibus ANOVA F-ratio evaluate, and why are Post-Hoc tests (such as Tukey's HSD) necessary following a significant F-statistic?
- A software firm tests a new UI on 50 users, recording average task completion time as 42.5 seconds (s = 8.2 seconds). The existing software benchmark is 45.0 seconds. State the null and alternative hypotheses, and calculate the degrees of freedom for a One-Sample t-test.
- Explain the criteria for choosing between Tukey's HSD and Games-Howell post-hoc procedures in SPSS One-Way ANOVA.
- Describe what visual pattern on an SPSS Profile Plot indicates the presence of a strong interaction effect between two factors.
- Detail the step-by-step navigation path in SPSS to perform a Paired-Samples t-test evaluating employee productivity before and after training.
Scenario Problem: A hospitality chain evaluates guest spending (in INR per day) across three room tiers (Standard, Deluxe, Executive). A One-Way ANOVA is conducted in SPSS across 150 guests (50 per tier). The SPSS ANOVA output reports:
- Did the dataset satisfy the assumption of homogeneity of variance? Justify using Levene's test output.
- Interpret the ANOVA F-statistic and p-value. What is your statistical decision regarding the omnibus hypothesis?
- Based on Tukey's HSD post-hoc test results, specify which pairs of room tiers differ significantly in daily guest spending.
- Formulate a concise executive summary paragraph reporting these findings in standard academic style for corporate management.
Download Module 3 Notes (PDF)
Calicut University • FYUGP 2024 Syllabus
Finished this module?
Continue reading the next module or return to the subject overview.