Choosing the wrong statistical test is one of the most technically serious errors in a quantitative dissertation. It is also one of the most avoidable with a systematic approach. The correct test depends on three things: the type of your data, the number of groups you're comparing, and the distribution of your data.
Step 1: Identify Your Data Type
This determines which family of tests applies to you:
- Nominal/Categorical: Named categories with no order (yes/no, gender, industry sector)
- Ordinal: Ordered categories but unequal intervals (Likert scales: 1–5 Strongly Disagree to Strongly Agree)
- Interval/Ratio: Continuous numeric data with equal intervals (age, revenue, temperature, exam score)
Important note on Likert scales: There is genuine academic debate about whether Likert data should be treated as ordinal (use non-parametric tests) or interval (use parametric tests). In MBA research, it is acceptable to treat a 5-point or 7-point Likert scale as interval if the overall scale has been validated — but document your decision and cite supporting methodology literature.
Step 2: What Are You Trying to Test?
Comparing Two Groups
- Independent t-test: Compare means of two separate groups (e.g., UK vs USA employee engagement scores). Requires normal distribution and interval data.
- Paired t-test: Compare same group before and after (e.g., performance pre- and post-training).
- Mann-Whitney U (non-parametric): Use instead of independent t-test when data is ordinal or non-normal.
- Wilcoxon Signed-Rank (non-parametric): Use instead of paired t-test for ordinal or non-normal data.
Comparing Three or More Groups
- One-Way ANOVA: Compare means across 3+ groups. Requires normal distribution and interval data. Post-hoc tests (Tukey, Bonferroni) identify which specific groups differ.
- Kruskal-Wallis (non-parametric): Use instead of ANOVA for ordinal or non-normal data.
Testing Relationships Between Variables
- Pearson Correlation: Measures linear relationship between two continuous variables. Assumes normality.
- Spearman's rho (non-parametric): Use when data is ordinal or non-normal.
- Linear Regression: Predicts a continuous outcome from one or more predictors.
- Logistic Regression: Predicts a binary outcome (yes/no, pass/fail) from predictors.
Step 3: Check Normality
Before choosing parametric vs non-parametric tests, check whether your data is normally distributed. In SPSS: Analyse → Descriptive Statistics → Explore → check the Shapiro-Wilk test (for samples under 50) or Kolmogorov-Smirnov (for larger samples). If Sig. > 0.05, data is normal — use parametric tests. If Sig. < 0.05, data violates normality — use non-parametric alternatives.
Concerned you've chosen the wrong test? Our tutors can review your data type and research question to confirm you're using the correct statistical approach. Book a 1:1 analysis session.