Key takeaways
- The null hypothesis (H0) is the default of no effect; the alternative (Ha) is what you seek evidence for.
- You reject or fail to reject H0—you never prove Ha with absolute certainty.
- One-tailed and two-tailed alternatives must match your theoretical prediction and test choice.
Null and alternative hypotheses are the foundation of every statistical test in quantitative research. Yet they remain confusing for many students writing project reports, theses, and dissertations. What exactly does the null hypothesis claim? How is the alternative different? Why do we 'fail to reject' rather than 'accept'? This guide explains null and alternative hypotheses in plain language with research project examples you can adapt directly for your dissertation proposal and methodology chapter.
What is the null hypothesis (H0)?
H0 is the sceptical default—the claim that there is no effect, no difference, or no relationship in the population. It is what statistical tests are designed to challenge. Examples: no difference between group means (μ1 = μ2), no correlation (ρ = 0), no regression effect (β = 0). H0 represents the status quo your data must overcome.
What is the alternative hypothesis (Ha or H1)?
Ha is your research prediction—the claim that an effect, difference, or relationship exists. Examples: groups differ (μ1 ≠ μ2), variables correlate (ρ ≠ 0), predictor matters (β ≠ 0). Ha is what you believe based on theory or prior evidence and what you hope your data supports.
The testing logic in simple terms
Imagine H0 says a coin is fair. You flip it 100 times and get 70 heads. The test asks: if the coin were fair, how often would you see 70 or more heads by chance? If that probability (p-value) is very small, you reject H0 and suspect the coin is biased. Academic hypothesis testing works the same way with sample means, correlations, and regression coefficients.
Why we 'fail to reject' rather than 'accept'
Statistics deals in probability, not proof. Failing to reject H0 means your data did not provide sufficient evidence against it—not that H0 is definitely true. A non-significant result might reflect a real null effect, insufficient sample size, or measurement error. Precision in language matters in dissertation write-ups.
One-tailed vs two-tailed alternative hypotheses
- Two-tailed Ha: μ1 ≠ μ2 (difference in either direction)—default choice.
- One-tailed Ha: μ1 > μ2 or μ1 < μ2 (specific direction only).
- One-tailed tests allocate all alpha to one tail—slightly more power but cannot detect opposite effects.
- Justify one-tailed tests with strong prior theory before data analysis.
Examples for common research projects
MBA project—customer satisfaction and repurchase: H0: ρ = 0. Ha: ρ > 0. BBA survey—gender and brand preference: H0: no association. Ha: association exists. MCA study—algorithm A vs B performance: H0: μA = μB. Ha: μA > μB. PhD experiment—intervention effect: H0: μintervention = μcontrol. Ha: μintervention ≠ μcontrol.
Writing H0 and Ha in your dissertation
State hypotheses formally in methodology chapter before results. Number them to match results section structure. Use population parameters, not sample statistics. Include direction where applicable. Link each pair to its statistical test explicitly.
Common misconceptions
- 'Accepting H0'—incorrect terminology; use 'fail to reject H0.'
- 'Proving Ha'—statistics provides evidence, not proof.
- Non-significant p-value means H0 is true—false; may indicate low power.
- Significant p-value means effect is large or important—check effect size.
- Ha must always be directional—non-directional is often appropriate.
Null hypothesis significance testing debates
Some researchers criticise over-reliance on p-values and arbitrary alpha thresholds. Modern best practice emphasises effect sizes, confidence intervals, and replication alongside hypothesis testing. For your dissertation, follow your department's expectations while reporting effect sizes and not over-interpreting borderline p-values.
Need help with hypothesis formulation?
Our research methodology support helps students write correct null and alternative hypotheses and connect them to valid statistical tests for dissertations and project reports.