Struggling to choose the right statistical test for your thesis or dissertation? Whether you’re a diploma student at KMTC, working on your master’s at MKU, or completing your PhD at University of Nairobi, understanding when to use parametric versus non-parametric tests is crucial for your research success.

What Are Parametric Tests?

Parametric tests are statistical methods that make specific assumptions about your data, particularly that your data follows a normal distribution (bell curve). These tests use sample data to make inferences about population parameters like the mean and standard deviation.

Think of parametric tests as the “strict teachers” – they’re powerful and efficient, but only if your data follows their rules.

Common Parametric Tests You’ll Use in Your Research:

Independent t-test – Comparing means between two different groups (e.g., male vs female performance)
Paired t-test – Comparing means from the same group at different times (e.g., before and after training)
ANOVA (Analysis of Variance) – Comparing means across three or more groups
Pearson Correlation – Measuring the linear relationship between two continuous variables
Linear Regression – Predicting outcomes based on one or more variables

What Are Non-Parametric Tests?

Non-parametric tests are “distribution-free” methods that don’t assume your data follows a normal distribution. These tests work with ranks, medians, or frequencies rather than means.

Think of non-parametric tests as the “flexible teachers” – they work with data that doesn’t meet strict requirements, though they may be slightly less powerful when parametric assumptions are met.

Common Non-Parametric Tests for Your Thesis:

Mann-Whitney U test – Non-parametric alternative to independent t-test
Wilcoxon Signed-Rank test – Alternative to paired t-test
Kruskal-Wallis test – Alternative to one-way ANOVA
Spearman’s Rank Correlation – Alternative to Pearson correlation
Chi-Square test – Testing relationships between categorical variables
Friedman test – Alternative to repeated measures ANOVA

Key Differences at a Glance

Feature	Parametric Tests	Non-Parametric Tests
Data assumptions	Requires normal distribution	No distribution assumptions
Data type	Interval or ratio scale	Ordinal, nominal, or non-normal data
Sample size	Works well with small or large samples	Better for small samples
Measures used	Mean, standard deviation	Median, ranks, frequencies
Power	More powerful when assumptions met	Less powerful but more robust
Outliers	Sensitive to extreme values	Resistant to outliers

When Should You Use Parametric Tests?

Use parametric tests when ALL of these conditions are met:

1. Normal Distribution

Your data follows a bell-shaped curve. You can check this using:

Shapiro-Wilk test (for samples < 50)
Kolmogorov-Smirnov test (for larger samples)
Q-Q plots (visual inspection)
Histograms

For Kenyan Students: If you’re analyzing exam scores, income levels, or measurement data from experiments, check normality first using SPSS, STATA, or R.

2. Adequate Sample Size

Generally, you need at least 30 observations per group. The Central Limit Theorem suggests that with larger samples (n > 30), the sampling distribution approaches normality even if your data isn’t perfectly normal.

PhD students working with smaller sample sizes (common in qualitative-dominant or mixed-methods research) should be extra cautious.

3. Homogeneity of Variance

Groups you’re comparing should have similar variances. Test this using:

Levene’s test
Bartlett’s test

4. Interval or Ratio Data

Your variables should be continuous measurements (height, weight, test scores, income) rather than categories or ranks.

5. Independence of Observations

Each data point should be independent – one participant’s response shouldn’t influence another’s.

When Should You Use Non-Parametric Tests?

Choose non-parametric tests when:

1. Non-Normal Distribution

Your Shapiro-Wilk or K-S test shows p < 0.05 (indicating non-normality)
Your histogram or Q-Q plot shows severe skewness
You have heavily skewed data (income data, reaction times, customer wait times)

Example for Masters Students: If you’re studying household incomes in Nairobi informal settlements, income data is typically right-skewed – use non-parametric tests.

2. Small Sample Size

You have fewer than 30 participants per group
Your pilot study has limited data
You’re working with rare populations

Common in Kenyan Context: Medical research at KNH, specialized professional groups, or rare disease studies often have small samples.

3. Ordinal Data

Likert scale responses (Strongly Disagree to Strongly Agree)
Rankings (1st place, 2nd place, 3rd place)
Educational levels (Primary, Secondary, Tertiary)
Satisfaction ratings (Poor, Fair, Good, Excellent)

For Diploma & Masters Students: Most questionnaire-based research uses Likert scales – these should typically be analyzed with non-parametric tests, though there’s ongoing debate about treating them as interval data.

4. Presence of Outliers

When your data has extreme values that aren’t errors but genuine observations, non-parametric tests are more robust.

5. Categorical/Nominal Data

Gender (Male/Female)
Yes/No responses
County of origin
Type of institution (Public/Private)

Step-by-Step: Choosing the Right Test for Your Research

Decision Tree for Your Thesis:

Step 1: What type of data do you have?

Categorical (nominal)? → Chi-Square test or Fisher’s Exact test
Ordinal or non-normal continuous? → Continue to Step 2
Normal continuous? → Continue to Step 3

Step 2: How many groups are you comparing? (Non-parametric route)

Two independent groups → Mann-Whitney U test
Two related/paired groups → Wilcoxon Signed-Rank test
Three or more independent groups → Kruskal-Wallis test
Three or more related groups → Friedman test
Relationship between two variables → Spearman’s correlation

Step 3: How many groups are you comparing? (Parametric route)

Two independent groups → Independent t-test
Two related/paired groups → Paired t-test
Three or more independent groups → One-way ANOVA
Three or more related groups → Repeated measures ANOVA
Relationship between two variables → Pearson correlation
Predicting an outcome → Linear regression

Practical Examples for Kenyan Students

Example 1: Comparing Teaching Methods (Education Research)

Research Question: Is there a significant difference in mathematics performance between students taught using the CBC curriculum versus the 8-4-4 system?

Scenario A – Use Independent t-test (Parametric):

You have 50 students in each group (n = 100 total)
Test scores are normally distributed (Shapiro-Wilk p > 0.05)
Variances are equal (Levene’s test p > 0.05)
Scores are continuous (0-100)

Scenario B – Use Mann-Whitney U test (Non-parametric):

You only have 20 students in each group
Scores are highly skewed
You have significant outliers

Example 2: Customer Satisfaction Study (Business Research)

Research Question: Do customers rate service quality differently across three Nairobi hospitals?

Data: 5-point Likert scale (1 = Very Dissatisfied to 5 = Very Satisfied)

Recommended Test: Kruskal-Wallis test (non-parametric)

Why? Likert scale data is ordinal, not truly continuous. Even though some researchers treat it as interval data, the conservative approach is non-parametric, especially for diploma and master’s level work.

Example 3: Income and Education Relationship (Economics/Social Sciences)

Research Question: Is there a relationship between years of education and monthly income among informal sector workers in Nairobi?

Use Spearman’s Correlation (Non-parametric) because:

Income data is typically right-skewed
You may have extreme outliers (very high earners)
The relationship may not be perfectly linear

Common Mistakes to Avoid in Your Thesis

1. Assuming Likert Scales Are Interval Data

Many students automatically use parametric tests on Likert scales. While some statisticians accept this for 5+ point scales with large samples, your thesis examiner may expect non-parametric tests. When in doubt, use non-parametric tests or justify your choice clearly in your methodology chapter.

2. Not Testing Assumptions

Don’t just assume your data is normal. Your examiner will look for:

Normality test results
Homogeneity of variance tests
Sample size justification

Include these in Chapter 3 (Methodology) or Chapter 4 (Results).

3. Ignoring Sample Size

With n < 30, be very cautious about using parametric tests, even if your data looks normal.

4. Misinterpreting “Non-Significant” Normality Tests

If your Shapiro-Wilk test gives p > 0.05, it means you fail to reject the null hypothesis of normality – your data could be normal. This is different from proving normality. Look at multiple indicators (histograms, Q-Q plots, skewness/kurtosis values).

5. Using the Wrong Test for Paired Data

If you’re comparing before-and-after measurements on the same participants (e.g., pre-test and post-test), you MUST use paired tests:

Paired t-test (parametric)
Wilcoxon Signed-Rank test (non-parametric)

Don’t use independent samples tests!

How to Report These Tests in Your Thesis

Reporting Parametric Tests:

Independent t-test example: “An independent samples t-test revealed that male students (M = 65.4, SD = 8.2) scored significantly higher than female students (M = 58.7, SD = 9.1) on the mathematics examination, t(98) = 3.76, p < 0.001, d = 0.76.”

One-way ANOVA example: “A one-way ANOVA indicated significant differences in customer satisfaction across the three hospitals, F(2, 147) = 8.43, p < 0.001, η² = 0.10. Post-hoc Tukey tests revealed…”

Reporting Non-Parametric Tests:

Mann-Whitney U test example: “A Mann-Whitney U test showed that customer wait times were significantly longer in Hospital A (Mdn = 45 minutes) compared to Hospital B (Mdn = 30 minutes), U = 1247, z = -3.21, p = 0.001, r = 0.35.”

Kruskal-Wallis test example: “The Kruskal-Wallis test revealed statistically significant differences in job satisfaction ratings across the three departments, χ²(2) = 12.87, p = 0.002. Follow-up Mann-Whitney tests with Bonferroni correction showed…”

Software Tools for Running These Tests in Kenya

SPSS (Most Common in Kenyan Universities)

Where: MKU, UoN, Strathmore, JKUAT computer labs
For Parametric: Analyze → Compare Means → Independent Samples T-Test
For Non-Parametric: Analyze → Nonparametric Tests → Legacy Dialogs

STATA (Popular for Economics/Health Research)

Available at UoN Economics department, KEMRI
Command examples: ttest, ranksum, kwallis

R (Free & Powerful – Growing in Kenyan Academia)

Download free from r-project.org
Parametric: t.test(), aov()
Non-parametric: wilcox.test(), kruskal.test()

Excel (Limited but Accessible)

Can perform basic t-tests
Not recommended for PhD-level analysis
Acceptable for simple diploma projects with supervisor approval

Need Help with Your Statistical Analysis?

Choosing between parametric and non-parametric tests is just the beginning. At Tobit Research Consulting, we help diploma, masters, and PhD students across Kenya with:

✓ Statistical test selection – We analyze your data and recommend appropriate tests
✓ Assumptions testing – Checking normality, homogeneity, and other requirements
✓ Data analysis in SPSS, STATA, R – Complete analysis with interpretation
✓ Results chapter writing – APA-formatted reporting of your statistical findings
✓ Statistical consultations – One-on-one sessions to understand your results

Trusted by students from: MKU, University of Nairobi, Kenyatta University, Strathmore, Daystar, JKUAT, Moi University, Egerton, and more.

📧 Contact us today: [Your contact information]
📍 Located in Nairobi – Available for in-person and online consultations

Frequently Asked Questions

Q: Can I use parametric tests if my sample size is large (n > 100) even if data isn’t perfectly normal?
A: Generally yes, due to the Central Limit Theorem. However, if you have severe skewness or extreme outliers, non-parametric tests are still safer. Document your decision in your methodology.

Q: My supervisor says to use t-test on Likert scales, but I’ve read they’re ordinal. What should I do?
A: This is a contentious issue in statistics. Many researchers treat summed Likert scales (e.g., total score from multiple items) as interval data, especially with 5+ points. Follow your supervisor’s guidance but acknowledge the limitation in your discussion chapter.

Q: What if my normality test is significant (p < 0.05) but my histogram looks normal?
A: Normality tests can be overly sensitive with large samples. Look at multiple indicators: histogram, Q-Q plot, skewness (-1 to +1 is acceptable), and kurtosis (-2 to +2). If most suggest normality, you can justify parametric tests.

Q: Do I need to transform my data if it’s not normal?
A: Transformation (log, square root) is one option, but using non-parametric tests is often simpler and more appropriate, especially at diploma and master’s level.

Q: How do I cite this guide in my thesis?
A: While this guide provides practical advice, always cite primary statistical textbooks and peer-reviewed articles in your actual thesis. We recommend: Field (2018), Pallant (2020), or Creswell & Creswell (2018).

Quick Reference: Test Substitutions

Parametric Test	Non-Parametric Alternative	Use When
Independent t-test	Mann-Whitney U test	Comparing 2 independent groups, non-normal data
Paired t-test	Wilcoxon Signed-Rank test	Comparing 2 related groups, non-normal data
One-way ANOVA	Kruskal-Wallis test	Comparing 3+ independent groups, non-normal data
Repeated measures ANOVA	Friedman test	Comparing 3+ related groups, non-normal data
Pearson correlation	Spearman correlation	Measuring relationships, ordinal or non-normal data
Linear regression	Various options	Consult with statistician for alternatives

Final Tips for Your Thesis Success

Test assumptions BEFORE choosing your test – Don’t assume parametric tests will work
Report both descriptive and inferential statistics – Means/medians AND test results
Use appropriate post-hoc tests – Tukey, Bonferroni for parametric; pairwise comparisons with corrections for non-parametric
Interpret effect sizes – Not just p-values (Cohen’s d, r, η²)
Create clear tables and figures – Follow APA 7th edition formatting
Seek statistical consultation early – Don’t wait until your data is collected

Still confused about which test to use for your specific research? Contact Tobit Research Consulting for personalized statistical guidance tailored to your thesis requirements and Kenyan university standards.

Keywords: parametric tests Kenya, non-parametric tests thesis, statistical analysis MKU, SPSS help Nairobi, research statistics UoN, Mann-Whitney test Kenya, t-test vs Mann-Whitney, thesis data analysis Kenya, PhD statistics help, masters thesis statistics

Parametric vs Non-Parametric Tests: Complete Guide for Research Students in Kenya