Understanding t‑Tests and ANOVA: A Comprehensive Guide from the Quantitative Data Analysis Training

Introduction

The third session of the quantitative data analysis training brought together researchers, health professionals, and dental public‑health experts to deepen their grasp of inferential statistics. After covering correlations and chi‑square tests in the previous week, the focus shifted to t‑tests and ANOVA, with a strong emphasis on variable types, measurement scales, hypothesis formulation, and practical interpretation of SPSS output.

1. Foundations Revisited

• Variables & Measurement Scales
• Quantitative: continuous (infinite values, e.g., temperature) vs. discrete (whole numbers, e.g., count of visits).
• Qualitative: ordinal/rank (ordered categories) vs. nominal (labels only, e.g., gender, colour).
• Binary/Dichotomous: a special case of nominal with exactly two levels (yes/no, male/female).
• Roles of Variables
• Independent (Predictor/Exposure) X – the factor that may cause change.
• Dependent (Outcome/Response) Y – the result we measure.
• Intervening (Mediating) Variable – sits on the causal pathway between X and Y (e.g., medical care mediates the link between income and longevity).
• Confounding Variable – influences both X and Y but is not caused by X (e.g., education affects both socioeconomic status and health).

2. Inferential Statistics Overview

• Goal: Use sample data to infer population parameters.
• Two Main Paths
• Estimation – point estimates (e.g., sample mean) and interval estimates (confidence intervals).
• Hypothesis Testing – parametric tests (assume normality) vs. non‑parametric tests (no distributional assumptions).
• Data Types
• Categorical outcome → chi‑square.
• Continuous outcome → correlation or t‑test/ANOVA.

3. The t‑Test Family

3.1 Decision Rule & Interpretation

1. Formulate null (H₀) – no difference between means.
2. Formulate alternative (H₁) – a difference exists (directional for one‑tailed, non‑directional for two‑tailed).
3. Compute the t statistic and compare it with the critical value from the t‑distribution (df = n‑1 for one‑sample; df = n₁+n₂‑2 for independent).
4. p‑value < α (0.05) → reject H₀; report the mean difference, confidence interval, and effect size.
5. If the confidence interval does not cross zero, the result is significant.

3.2 Practical SPSS Walk‑through

• One‑sample example: 25 participants, mean = 130, population mean = 100, SD = 15 → t = 10, p < 0.001 → training was effective.
• Paired example: Pre‑test vs. post‑test scores for the same class; compute mean difference, standard error, and 95 % CI.
• Independent example: Compare exam scores of two separate classes; first check Levene’s test for equality of variances.
• Levene p > 0.05 → assume equal variances.
• Levene p ≤ 0.05 → use the “equal variances not assumed” row.

3.3 Non‑Parametric Counterparts

• One‑sample → Sign test or Wilcoxon signed‑rank.
• Paired → Wilcoxon signed‑rank.
• Independent → Mann‑Whitney U. These are invoked when normality is violated or data are ordinal.

4. Analysis of Variance (ANOVA)

• Purpose: Test whether three or more group means differ simultaneously, protecting the overall Type I error rate (α = 0.05).
• Key Concepts
• Between‑group variation – how each group mean deviates from the grand mean.
• Within‑group variation – variability of individual scores around their own group mean.
• F‑ratio = (Mean Square Between) / (Mean Square Within).
• Large F → reject H₀ (means are not all equal).
• Assumptions
• Dependent variable is continuous and approximately normally distributed.
• Homogeneity of variances (Levene’s test).
• Independent observations.
• Interpretation of ANOVA Table
• Sum of Squares (SS), Degrees of Freedom (df), Mean Square (MS), F, p‑value.
• Example: 3 treatment groups for skin‑blister healing (A, B, placebo). Significant F → at least one group differs.
• Post‑hoc Tests
• Conducted only when ANOVA is significant.
• Commonly Tukey HSD, Bonferroni, or Scheffé to pinpoint which pairs differ.

5. Extending Beyond One‑Way ANOVA

• Repeated‑Measures ANOVA – for the same subjects measured > 2 times (e.g., baseline, 1 h, 2 h). Controls Type I error without inflating the number of tests.
• Two‑Way / Factorial ANOVA – handles two or more categorical independent variables simultaneously (e.g., treatment and gender effects).

6. Frequently Asked Questions (selected)

• Can I rearrange post‑test items to reduce carry‑over? Yes – randomising question order helps.
• How do I run a t‑test in SPSS? Load data, choose Analyze → Compare Means → Independent Samples T Test (or Paired‑Samples T Test), specify grouping variable, and review the output.
• What does “degrees of freedom” mean? It reflects the number of independent pieces of information that can vary when estimating a statistic; essentially the “room” for variability.
• When to use repeated‑measures ANOVA? When the same participants are measured at three or more time points or under three+ conditions.
• Do I need ANOVA for categorical dependent variables? No – for a categorical outcome you would use chi‑square or logistic regression; ANOVA requires a continuous dependent variable.

7. What Comes Next?

The next session will cover regression analysis and contingency‑table techniques, rounding off the quantitative data‑analysis series.

Key Takeaways 1. Distinguish variable types (continuous, discrete, nominal, ordinal, binary) and their roles (independent, dependent, mediating, confounding). 2. Choose the correct t‑test based on the number of groups, relationship between observations, and variance‑equality assumptions. 3. Use Levene’s test to decide between “equal variances assumed” and “not assumed” rows for independent‑sample t‑tests. 4. ANOVA extends the t‑test to three or more groups, comparing between‑group and within‑group variance via the F‑ratio. 5. A significant ANOVA requires post‑hoc comparisons (e.g., Tukey HSD) to identify the specific group differences. 6. When normality or variance assumptions fail, switch to the appropriate non‑parametric alternatives (Wilcoxon, Mann‑Whitney, Kruskal‑Wallis).

Mastering the selection, execution, and interpretation of t‑tests and ANOVA equips researchers to draw valid, statistically sound conclusions from their data, while respecting assumptions and controlling error rates.