How to Choose a Statistical Test: A Simple Guide for Students

Choosing the statistical test is often the most stressful part of planning a quantitative thesis. The good news is that for most student projects the choice comes down to four questions: what you want to find out, what type of variables you have, how many groups or measurements you are comparing, and whether your data meet the assumptions of parametric tests. This guide walks through each question and ends with a summary table. If you prefer an interactive version, try the ScholarCalc test chooser.

Question 1: What is your aim?

Almost every analysis in a thesis falls into one of four families:

Describe your sample and variables: means, standard deviations, frequencies, percentages.
Compare groups or time points: is there a difference between men and women, or between pre-test and post-test?
Relate variables: is there an association between study hours and grades, or can one variable predict another?
Check measurement quality: is the questionnaire reliable, or do two raters agree?

Write each research question or hypothesis down and label it with one of these aims. ‘Is there a significant difference…’ points to a comparison test; ‘Is there a significant relationship…’ points to correlation or regression.

Question 2: What type of variables do you have?

The measurement level of your variables limits which tests make sense:

Categorical (nominal): categories without order, such as gender, school type or yes/no answers.
Ordinal: ordered categories, such as education level or a single Likert item (strongly disagree to strongly agree).
Continuous (interval or ratio): numerical scores such as test marks, age, income or the total or mean score of a multi-item scale.

In a comparison, the grouping variable is the independent variable (usually categorical) and the outcome is the dependent variable. Totals or means of multi-item Likert scales are usually treated as continuous in social science research, while single Likert items are safer treated as ordinal.

Question 3: How many groups, and are they independent?

For comparisons, two further details decide the test:

Number of groups or conditions: two, or three or more.
Independent or related: independent groups contain different people (for example, public and private school students). Related or paired data come from the same people measured more than once (pre-test and post-test) or from matched pairs.

Question 4: Are the parametric assumptions met?

Parametric tests such as t tests, ANOVA and Pearson correlation assume, among other things, that the continuous outcome is approximately normally distributed (within each group, or for the differences in paired designs). With moderate or large samples these tests are fairly robust, but with small samples, strongly skewed data, ordinal outcomes or extreme outliers, a non-parametric alternative based on ranks is safer. Check normality with histograms, Q–Q plots and skewness and kurtosis values; formal tests such as Shapiro–Wilk become over-sensitive in large samples (Ghasemi & Zahediasl, 2012). The descriptive statistics calculator gives skewness and kurtosis instantly.

Summary table

Aim	Data	Parametric test	Non-parametric alternative
Compare 2 independent groups	Continuous outcome	Independent samples t test	Mann-Whitney U
Compare 2 related measurements	Continuous outcome	Paired samples t test	Wilcoxon signed-rank
Compare 3+ independent groups	Continuous outcome	One-way ANOVA + Tukey HSD	Kruskal-Wallis + Dunn
Compare 3+ related measurements	Continuous outcome	Repeated measures ANOVA	Friedman test
Relationship between 2 variables	Both continuous	Pearson correlation	Spearman correlation
Relationship, ordinal data	At least one ordinal	—	Spearman correlation
Predict an outcome from predictors	Continuous outcome	Linear regression	—
Association between 2 categorical variables	Both categorical	—	Chi-square test of independence
Questionnaire reliability	Multi-item scale	Cronbach’s alpha	—
Agreement between 2 raters	Categorical codes	—	Cohen’s kappa

Worked examples

Example 1: Gender and job satisfaction

Research question: ‘Is there a significant difference in job satisfaction between male and female teachers?’ The aim is comparison; the grouping variable (gender) has two independent categories; job satisfaction is the mean of a 10-item scale (continuous). If the scores are roughly normal, use an independent samples t test; otherwise use the Mann-Whitney U test. Report the effect size as Cohen’s d (or r for Mann-Whitney).

Example 2: Training programme with three measurements

Research question: ‘Does anxiety change from baseline to mid-programme to post-programme?’ The same people are measured three times, so the data are related with three conditions: a repeated measures ANOVA, with Mauchly’s test for sphericity, or the Friedman test if the data are clearly non-normal.

Example 3: Study hours and GPA

Research question: ‘Is there a relationship between weekly study hours and GPA?’ Both variables are continuous, so use Pearson correlation if the relationship is linear and there are no extreme outliers, otherwise Spearman. If you also want to predict GPA from several factors (study hours, attendance, sleep), use multiple linear regression. See Pearson vs Spearman for more detail.

Example 4: School type and internet access

Research question: ‘Is internet access at home associated with school type (public/private)?’ Both variables are categorical, so use the chi-square test of independence and report Cramér’s V as the effect size. Check that expected counts are large enough (a common guideline is at least 5 in most cells).

Tips for choosing well

Match the test to the hypothesis, not the other way round. Write your hypotheses first and choose one test for each.
Avoid multiple t tests for three or more groups. Use ANOVA with post hoc tests to keep the Type I error rate under control.
Plan before you collect data. The test you choose determines the type of data you need and the sample size. See how to calculate sample size.
Report effect sizes with every test, as APA 7 recommends (APA, 2020). See effect size interpretation.
Check with your supervisor when your design is more complex, for example two or more independent variables (factorial ANOVA), mixed designs, mediation or moderation, or structural equation modelling. These go beyond the tests in this guide.

What about more advanced designs?

Two-way ANOVA, ANCOVA, MANOVA, logistic regression, factor analysis and structural equation modelling are all common in postgraduate research. The same four questions still apply, but the number of variables increases. Textbooks such as Field (2018), Pallant (2020) and Tabachnick and Fidell (2019) include detailed decision charts for these methods.

Paired or independent: a common trap

Students often treat related data as independent. Examples of related data include pre-test and post-test scores from the same students, ratings of two products by the same customers, and matched pairs such as siblings or students matched on prior achievement. Analysing them with an independent samples test ignores the pairing and usually loses power. On the other hand, two separate classes measured once each are independent, even if they were taught by the same teacher. Ask yourself: is each score in group A linked to one specific score in group B? If yes, the data are paired.

One test per hypothesis

Write down every hypothesis and the test you will use for it before you collect the data. This analysis plan belongs in your methodology chapter and helps you avoid running many tests and reporting only the significant ones, which inflates the false-positive rate (see the p-value explained).

Roman Urdu mein khulasa

Statistical test chunne ke liye chaar sawal poochein: (1) aap ka maqsad kya hai, describe karna, compare karna, relationship dekhna ya reliability check karna? (2) variables kis qisam ke hain, categorical, ordinal ya continuous? (3) kitne groups hain aur kya woh alag log hain ya wahi log bar bar measure hue? (4) kya data normal hai? Masalan do alag groups ke liye independent t test (ya Mann-Whitney), wahi log do martaba ho to paired t test (ya Wilcoxon), teen ya zyada groups ke liye ANOVA (ya Kruskal-Wallis), do continuous variables ke rishte ke liye Pearson (ya Spearman), aur do categorical variables ke liye chi-square.

Frequently asked questions

What test should I use to compare two groups?

Use an independent samples t test for a continuous outcome with roughly normal data, or the Mann-Whitney U test for ordinal or clearly non-normal data.

Which test compares more than two groups?

One-way ANOVA with Tukey HSD post hoc tests, or the Kruskal-Wallis test with Dunn’s post hoc tests for non-normal or ordinal data.

Can I use parametric tests on Likert scale data?

For totals or means of multi-item scales, parametric tests are widely used. For single Likert items, non-parametric tests are safer.

How do I know if my data are normal?

Look at histograms and Q–Q plots and check skewness and kurtosis. Formal tests like Shapiro–Wilk are very sensitive in large samples, so interpret them together with the plots.

Which test shows a relationship between two categorical variables?

The chi-square test of independence, with Cramér’s V as the effect size.

Related calculators

References

American Psychological Association. (2020). Publication manual of the American Psychological Association (7th ed.). https://doi.org/10.1037/0000165-000
Field, A. (2018). Discovering statistics using IBM SPSS Statistics (5th ed.). SAGE.
Ghasemi, A., & Zahediasl, S. (2012). Normality tests for statistical analysis: A guide for non-statisticians. International Journal of Endocrinology and Metabolism, 10(2), 486–489. https://doi.org/10.5812/ijem.3505
Pallant, J. (2020). SPSS survival manual: A step by step guide to data analysis using IBM SPSS (7th ed.). Routledge.
Tabachnick, B. G., & Fidell, L. S. (2019). Using multivariate statistics (7th ed.). Pearson.

Written and maintained by the ScholarCalc team. Last reviewed: November 2026.