Hypothesis Testing and Report Writing
UGC NET Commerce: hypotheses, errors, z-test, t-test, ANOVA, chi-square, Mann-Whitney U, Kruskal-Wallis H, rank correlation tests and report writing.
- 25 min
On this page · 19 sections
- Prerequisites
- The Hook: Did the Training Programme Work?
- 1. The Logic of Hypothesis Testing
- 2. The z-Test
- 3. The t-Test
- 4. The F-Test and Analysis of Variance (ANOVA)
- 5. The Chi-Square (χ²) Test
- 6. Mann–Whitney U-Test
- 7. Kruskal–Wallis H-Test
- 8. Rank Correlation Test
- 9. Choosing the Right Test
- 10. Report Writing
- Memory Tricks
- Don’t Confuse
- Common Mistakes
- Practice
- Quick Revision Table
- Key Takeaways
- What’s Next
Prerequisites
- Research Design, Data Collection and Sampling — standard error and the central limit theorem, on which every test statistic rests.
- Probability and Probability Distributions — the normal curve and the critical values 1.96 and 2.58.
- Correlation and Regression — Spearman’s rank correlation, tested for significance here.
- Methods and Steps of Research — Type I and Type II errors at Paper 1 level.
The Hook: Did the Training Programme Work?
1. The Logic of Hypothesis Testing
| Term | Meaning |
|---|---|
| Null hypothesis (H₀) | Statement of no difference / no effect / no relationship; the hypothesis actually tested. Always contains equality (μ = 500) |
| Alternative hypothesis (H₁ or Hₐ) | What is accepted if H₀ is rejected (μ ≠ 500, μ > 500 or μ < 500) |
| Level of significance (α) | Maximum risk of rejecting a true H₀ that we accept; usually 5% or 1% |
| Test statistic | (Statistic − hypothesised parameter) ÷ standard error |
| Critical region (rejection region) | Values of the test statistic for which H₀ is rejected |
| Critical value | Boundary of the critical region |
| p-value | Probability, if H₀ is true, of a result at least as extreme as the one observed. Reject H₀ if p ≤ α |
| Degrees of freedom (df) | Number of values free to vary after restrictions are imposed (n − 1 for one sample) |
| Power | \( 1 - \beta \): probability of rejecting H₀ when it is false |
Errors
| H₀ true | H₀ false | |
|---|---|---|
| Reject H₀ | Type I error (α): “false alarm”, producer’s risk | Correct (power, 1 − β) |
| Accept (do not reject) H₀ | Correct (1 − α) | Type II error (β): “missed effect”, consumer’s risk |
For a fixed sample size, reducing α increases β. The only way to reduce both at once is a larger sample.
One-tailed and two-tailed tests
| H₁ | Test | Critical z at 5% | Critical z at 1% |
|---|---|---|---|
| μ ≠ μ₀ | Two-tailed | ±1.96 | ±2.58 |
| μ > μ₀ | Right-tailed | +1.645 | +2.33 |
| μ < μ₀ | Left-tailed | −1.645 | −2.33 |
Steps
graph TD
A["1. State H₀ and H₁"] --> B["2. Choose level of significance α"]
B --> C["3. Choose the test and compute
the test statistic"]
C --> D["4. Find the critical value
(table, with df) or the p-value"]
D --> E["5. Compare and decide:
reject or do not reject H₀"]
E --> F["6. State the conclusion
in the language of the problem"]Parametric vs non-parametric tests
| Basis | Parametric | Non-parametric (distribution-free) |
|---|---|---|
| Assumption about population | Usually normal, with known form; parameters estimated | No assumption about the shape of the distribution |
| Level of measurement | Interval or ratio | Nominal or ordinal (also interval data that break assumptions) |
| Sample size | Works best with adequate n | Suits small samples |
| Power | More powerful when assumptions hold | Less powerful, more robust |
| Examples | z, t, F (ANOVA), Pearson’s r test | Chi-square, sign test, Wilcoxon, Mann–Whitney U, Kruskal–Wallis H, Spearman’s rank test, Friedman, run test |
Parametric test and its non-parametric counterpart
| Situation | Parametric | Non-parametric |
|---|---|---|
| Two independent samples | Independent t-test | Mann–Whitney U (Wilcoxon rank-sum) |
| Two related (paired) samples | Paired t-test | Wilcoxon signed-rank; sign test |
| Three or more independent samples | One-way ANOVA | Kruskal–Wallis H |
| Three or more related samples | Repeated-measures / two-way ANOVA | Friedman test |
| Association of two variables | Pearson’s r (t-test for r) | Spearman’s rank correlation; chi-square for categories |
2. The z-Test
Used when the sample is large (n > 30 as a working rule) or the population SD is known and the population is normal. The test statistic follows the standard normal distribution.
| Test | Statistic |
|---|---|
| Single mean | \( z = \dfrac{\bar{x} - \mu_0}{\sigma/\sqrt{n}} \) (use s for σ if unknown and n is large) |
| Difference of two means | \( z = \dfrac{\bar{x}_1 - \bar{x}_2}{\sqrt{\sigma_1^2/n_1 + \sigma_2^2/n_2}} \) |
| Single proportion | \( z = \dfrac{p - P_0}{\sqrt{P_0Q_0/n}} \) |
| Difference of proportions | \( z = \dfrac{p_1 - p_2}{\sqrt{\hat{p}\hat{q}(1/n_1 + 1/n_2)}} \), \( \hat{p} \) pooled |
3. The t-Test
Developed by William Sealy Gosset, who published under the pen name “Student” in 1908 while working for the Guinness brewery, so the distribution is Student’s t. Use it for small samples (n ≤ 30) from a normal population whose σ is unknown.
Properties of the t-distribution: symmetric and bell-shaped around zero, but with heavier tails than the normal (so critical values are larger); its shape depends on degrees of freedom; as df increases it approaches the standard normal.
| Test | Statistic | df |
|---|---|---|
| Single mean | \( t = \dfrac{\bar{x} - \mu_0}{s/\sqrt{n}} \), with \( s^2 = \frac{\sum(x - \bar{x})^2}{n - 1} \) | n − 1 |
| Two independent means (equal variances) | \( t = \dfrac{\bar{x}_1 - \bar{x}_2}{s_p\sqrt{1/n_1 + 1/n_2}} \), \( s_p^2 \) pooled | \( n_1 + n_2 - 2 \) |
| Paired (dependent) samples | \( t = \dfrac{\bar{d}}{s_d/\sqrt{n}} \), d = after − before | n − 1 (n pairs) |
| Significance of r | \( t = \dfrac{r\sqrt{n - 2}}{\sqrt{1 - r^2}} \) | n − 2 |
The paired t-test suits before-and-after designs on the same units (sales of the same executives before and after training). Using the independent-samples test on paired data is a common error.
4. The F-Test and Analysis of Variance (ANOVA)
ANOVA was developed by R.A. Fisher; the F-distribution is named in his honour. It tests whether three or more population means are equal by comparing the variation between groups with the variation within groups.
Assumptions: independent random samples; normal populations; equal variances (homoscedasticity).
H₀: \( \mu_1 = \mu_2 = \cdots = \mu_k \). H₁: not all means are equal.
One-way ANOVA table
| Source | Sum of squares | df | Mean square | F |
|---|---|---|---|---|
| Between groups (treatments) | SSB | k − 1 | MSB = SSB/(k − 1) | F = MSB/MSW |
| Within groups (error) | SSW | N − k | MSW = SSW/(N − k) | |
| Total | SST = SSB + SSW | N − 1 |
A large F means the group means differ more than chance would allow; F near 1 means the between-group variation is no more than the within-group (chance) variation.
Two-way ANOVA adds a second factor (or blocks, as in the randomised block design): df for rows r − 1, columns c − 1, error (r − 1)(c − 1).
The F-test is also used on its own to compare two population variances: \( F = s_1^2/s_2^2 \) with \( (n_1 - 1, n_2 - 1) \) df, placing the larger sample variance in the numerator so that F ≥ 1.
5. The Chi-Square (χ²) Test
Introduced by Karl Pearson (1900). A non-parametric test on frequencies (counts), with three main uses:
| Use | H₀ | df |
|---|---|---|
| Goodness of fit | Observed frequencies fit a specified distribution (uniform, binomial, Poisson, normal) | k − 1 − (number of parameters estimated) |
| Independence of attributes (contingency table) | The two attributes are independent | (r − 1)(c − 1) |
| Test of a population variance | \( \sigma^2 = \sigma_0^2 \) | n − 1 ; \( \chi^2 = (n - 1)s^2/\sigma_0^2 \) |
Conditions: observations independent and random; total frequency reasonably large (often N ≥ 50); each expected frequency at least 5 (pool adjacent classes if not); use counts, never percentages. For a 2 × 2 table with small expected frequencies, apply Yates’ correction: \( \sum (\lvert O - E \rvert - 0.5)^2/E \).
Properties of the χ² distribution: values are never negative (0 to ∞); positively skewed, becoming more symmetric as df rises; mean = df, variance = 2 df. χ² = 0 means perfect agreement between observed and expected.
6. Mann–Whitney U-Test
Published by H.B. Mann and D.R. Whitney (1947); it is equivalent to Wilcoxon’s rank-sum test. It tests whether two independent samples come from identical populations (often stated as equal medians), using ranks rather than values. It is the non-parametric alternative to the independent-samples t-test, used for ordinal data or when normality is doubtful.
Procedure:
- Combine both samples and rank all values from smallest (rank 1) to largest; ties get average ranks.
- Find the rank sums \( R_1 \) and \( R_2 \).
- Compute
with the check \( U_1 + U_2 = n_1n_2 \). The test statistic is U = the smaller of the two.
- Small samples: reject H₀ if U ≤ the table critical value. Large samples (both above about 10, by common textbook rule): U is approximately normal with
7. Kruskal–Wallis H-Test
Published by W.H. Kruskal and W.A. Wallis (1952). It extends the U-test to three or more independent samples and is the non-parametric alternative to one-way ANOVA.
- Rank all N observations together (ties get average ranks).
- Find the rank sum \( R_j \) of each of the k samples.
- Compute
- When each sample has more than about five observations, H follows approximately a χ² distribution with k − 1 df. Reject H₀ if H exceeds the critical χ².
8. Rank Correlation Test
Tests whether Spearman’s rank correlation \( R \) (or \( r_s \)) in the sample shows a real association in the population. H₀: \( \rho_s = 0 \) (no association).
- Small n (roughly up to 30): compare R with the critical value from Spearman’s table for n.
- Larger n: \( t = R\sqrt{\dfrac{n - 2}{1 - R^2}} \) with n − 2 df (or, for large n, \( z = R\sqrt{n - 1} \)).
9. Choosing the Right Test
| Question | Data and design | Test |
|---|---|---|
| Is the mean equal to a claimed value? | Large n or σ known | z |
| Is the mean equal to a claimed value? | Small n, σ unknown, normal population | t (df n − 1) |
| Do two group means differ? | Independent, normal, small n | Independent t (df n₁ + n₂ − 2) |
| Did a change work on the same units? | Before–after pairs | Paired t (df n − 1); non-parametric: Wilcoxon signed-rank |
| Do two groups differ? | Ordinal or non-normal, independent | Mann–Whitney U |
| Do three or more means differ? | Normal, equal variances | ANOVA (F) |
| Do three or more groups differ? | Ordinal or non-normal | Kruskal–Wallis H |
| Are two attributes associated? | Counts in a contingency table | χ² test of independence |
| Does the data fit a distribution? | Observed vs expected counts | χ² goodness of fit |
| Are two variances equal? | Two normal samples | F-test |
| Is a rank correlation significant? | Ranks | Rank correlation test (table or t with n − 2 df) |
10. Report Writing
The research report is the last step of research: without it, the findings are lost to the people who must act on them. A report is judged by whether the reader can understand what was done, trust it and use it.
Types of report
| Basis | Types |
|---|---|
| Audience | Technical report (for specialists: full methodology, data, statistical detail) · Popular report (for general readers and managers: simple language, findings and implications first, charts) |
| Form | Written · Oral (presentation; needs visual aids and a short written summary) |
| Length | Short report (letter or memo form for routine matters) · Long report (formal study) |
| Other | Interim / progress report · Final report · Research monograph · Journal article |
Layout of a research report
| Part | Contents |
|---|---|
| Preliminary pages | Title page; letter of transmittal (business reports); certificate and declaration (dissertations); preface or acknowledgements; table of contents; list of tables and figures; abbreviations; executive summary or abstract |
| Main text | Introduction (background, statement of the problem, objectives, hypotheses, scope, significance) · Review of literature · Research methodology (design, population and sample, tools, data sources, statistical techniques, limitations) · Results and analysis (tables, figures, test results) · Discussion and interpretation · Findings, conclusions and recommendations · Summary |
| End matter | Appendices (questionnaire, raw data, detailed tables) · Bibliography or references · Index (in books and monographs) |
An executive summary is a short, self-contained version for decision-makers: problem, method in a sentence, key findings and recommendations. It is written last but placed first.
Steps in writing the report
graph LR
A["Logical analysis
of the subject matter"] --> B["Final outline"]
B --> C["Rough draft"]
C --> D["Rewriting and
polishing"]
D --> E["Final bibliography"]
E --> F["Final draft"]- Logical analysis of the subject matter — organise the material logically or chronologically.
- Preparation of the final outline — the framework of chapters and headings.
- Preparation of the rough draft — what was done, how, and what was found.
- Rewriting and polishing — check logic, consistency, grammar, and that conclusions follow from data.
- Preparation of the final bibliography.
- Writing the final draft — clear, objective language; no jargon; illustrations and tables in place.
Qualities of a good report
- Clarity and simplicity: short sentences; technical terms defined; one idea per paragraph.
- Objectivity: the researcher’s opinions are separated from the findings; written in the impersonal style.
- Accuracy: data, calculations, quotations and references checked.
- Logical sequence and coherence: each chapter leads to the next; conclusions follow from the analysis.
- Brevity with completeness: “long enough to cover the subject but short enough to maintain interest” is the common textbook rule.
- Visual support: tables and charts numbered, titled and sourced; each discussed in the text.
- Acknowledgement of limitations: sample, method and scope limits stated honestly.
- Proper documentation: every source cited consistently in one style (APA, MLA, Chicago); footnotes for explanations and references.
Mechanics and documentation
- Footnotes serve two purposes: to identify the source of quoted or borrowed material, and to add explanatory notes that would interrupt the text.
- Abbreviations in footnotes: ibid. (same source as the previous note), op. cit. (work already cited, different page), loc. cit. (same work and same page cited earlier), et al. (and others), cf. (compare). See Thesis Writing, Referencing and ICT for full citation styles.
- A bibliography lists all works consulted; a reference list lists only works cited.
- Tables and figures are numbered in sequence and given titles; follow one style consistently for where titles and source notes go.
Memory Tricks
Don’t Confuse
| This | vs | That |
|---|---|---|
| Type I error (α): reject true H₀; producer’s risk | vs | Type II error (β): accept false H₀; consumer’s risk |
| Level of significance α | vs | Power 1 − β; confidence level 1 − α |
| One-tailed 5%: 1.645 | vs | Two-tailed 5%: 1.96 |
| Independent t | vs | Paired t (same units before and after) |
| Mann–Whitney U: two independent groups | vs | Wilcoxon signed-rank: two related groups |
| ANOVA: three or more means, parametric | vs | Kruskal–Wallis: three or more groups, ranks |
| χ² of independence: df (r − 1)(c − 1) | vs | χ² goodness of fit: df k − 1 − parameters estimated |
| Technical report: for specialists, method-heavy | vs | Popular report: for managers, findings first |
| Bibliography: all works consulted | vs | References: only works cited |
| Executive summary: placed first, written last | vs | Summary/conclusions: at the end of the main text |
Common Mistakes
Practice
All questions below are original practice questions written in the UGC NET format. They are not previous-year questions.
Level 1: Foundation
- What are the degrees of freedom for a χ² test on a 4 × 3 contingency table?
- Which non-parametric test replaces one-way ANOVA?
- Under what pen name did W.S. Gosset publish the t-distribution?
Answers
- \( (4 - 1)(3 - 1) = \) 6.
- Kruskal–Wallis H-test.
- “Student” (1908).
Level 2: Exam-style MCQs
Q1. (Practice) A sample of 36 has mean 105; the population SD is 18. To test H₀: μ = 100 against H₁: μ > 100 at 5%, the test statistic and decision are:
- (a) z = 1.67; reject H₀
- (b) z = 1.67; do not reject H₀
- (c) z = 0.28; do not reject H₀
- (d) t = 1.67; reject H₀
Answer and explanation
(a). \( z = (105 - 100)/(18/6) = 5/3 \approx 1.67 \). For a right-tailed test at 5% the critical value is 1.645, and 1.67 > 1.645, so H₀ is rejected. Using the two-tailed 1.96 would wrongly give (b).
Q2. (Practice) Customer satisfaction ranks (ordinal) from three independent branches are to be compared. The appropriate test is:
- (a) One-way ANOVA
- (b) Mann–Whitney U-test
- (c) Kruskal–Wallis H-test
- (d) Paired t-test
Answer and explanation
(c). Three or more independent groups with ordinal data. Mann–Whitney handles only two groups; ANOVA needs interval data and normality.
Q3. (Practice) In a one-way ANOVA with 4 groups and 20 observations in all, the degrees of freedom for the F-ratio are:
- (a) (3, 16)
- (b) (4, 20)
- (c) (3, 19)
- (d) (4, 16)
Answer and explanation
(a) (3, 16). Between groups k − 1 = 3; within groups N − k = 16.
Q4. (Practice) Consider the following statements about the χ² distribution:
- Its mean equals its degrees of freedom.
- It can take negative values.
- It is positively skewed and becomes more symmetric as df increases.
Which of the statements given above is/are correct?
- (a) 1 and 2 only
- (b) 1 and 3 only
- (c) 2 and 3 only
- (d) 1, 2 and 3
Answer and explanation
(b) 1 and 3 only. χ² is a sum of squares and can never be negative.
Q5. (Practice) In a Mann–Whitney test with \( n_1 = 8 \) and \( n_2 = 10 \), \( U_1 = 22 \). Then \( U_2 \) is:
- (a) 58
- (b) 18
- (c) 80
- (d) 102
Answer and explanation
(a) 58. \( U_1 + U_2 = n_1n_2 = 80 \), so \( U_2 = 80 - 22 = 58 \). The test statistic is the smaller value, 22.
Q6. (Practice) In a research report, the questionnaire used for data collection is normally placed in:
- (a) The introduction
- (b) The research methodology chapter
- (c) The appendix
- (d) The executive summary
Answer and explanation
(c) The appendix. The methodology chapter describes the tool; the full instrument goes in the end matter.
Level 3: Assertion–reason, match and sequence
For assertion–reason questions, use these options:
- (a) Both A and R are true, and R is the correct explanation of A
- (b) Both A and R are true, but R is not the correct explanation of A
- (c) A is true, but R is false
- (d) A is false, but R is true
Q1. (Practice) Assertion (A): For a small sample from a normal population with unknown σ, the t-test is preferred to the z-test. Reason (R): The t-distribution has heavier tails than the normal distribution, allowing for the extra uncertainty of estimating σ from a small sample.
Answer and explanation
(a). The heavier tails are exactly the allowance for estimating σ by s; that is why t is used.
Q2. (Practice) Assertion (A): Reducing the level of significance from 5% to 1%, with the sample size unchanged, increases the probability of a Type II error. Reason (R): A smaller α shrinks the rejection region, making it harder to reject a false null hypothesis.
Answer and explanation
(a). With n fixed, α and β move in opposite directions; R explains the mechanism.
Q3. (Practice) Assertion (A): The Mann–Whitney U-test can be applied to ordinal data. Reason (R): The Mann–Whitney U-test requires the populations to be normally distributed.
Answer and explanation
(c). A is true; it works on ranks. R is false: it is distribution-free.
Q4. (Practice) Match List I with List II.
| List I (Test) | List II (Associated with) |
|---|---|
| A. t-test | I. Karl Pearson |
| B. χ² test | II. R.A. Fisher |
| C. Analysis of variance | III. W.S. Gosset (“Student”) |
| D. H-test | IV. Kruskal and Wallis |
- (1) A-III, B-I, C-II, D-IV
- (2) A-II, B-I, C-III, D-IV
- (3) A-III, B-II, C-I, D-IV
- (4) A-I, B-III, C-II, D-IV
Answer and explanation
(1) A-III, B-I, C-II, D-IV.
Q5. (Practice) Match the situation with the test.
| List I (Situation) | List II (Test) |
|---|---|
| A. Sales of 12 salespersons before and after training | I. χ² test of independence |
| B. Association between gender and preferred payment mode | II. Paired t-test |
| C. Mean yields of four fertilisers | III. Mann–Whitney U |
| D. Ranks of two independent groups of consumers | IV. ANOVA |
- (1) A-II, B-I, C-IV, D-III
- (2) A-II, B-IV, C-I, D-III
- (3) A-III, B-I, C-IV, D-II
- (4) A-IV, B-I, C-II, D-III
Answer and explanation
(1) A-II, B-I, C-IV, D-III.
Q6. (Practice) Arrange the steps of hypothesis testing in order:
A. Compute the test statistic
B. Formulate the null and alternative hypotheses
C. Compare with the critical value and decide
D. Fix the level of significance
(1) B, D, A, C
(2) D, B, A, C
(3) B, A, D, C
(4) A, B, D, C
Answer and explanation
(1) B, D, A, C. α must be fixed before the statistic is computed, so the decision rule is not influenced by the result.
Q7. (Practice) Arrange the following parts of a research report in their usual order:
A. Bibliography
B. Research methodology
C. Table of contents
D. Findings and recommendations
E. Review of literature
(1) C, E, B, D, A
(2) C, B, E, D, A
(3) E, C, B, D, A
(4) C, E, D, B, A
Answer and explanation
(1) C, E, B, D, A. Preliminary pages, literature, methodology, findings, end matter.
Quick Revision Table
| Item | Key point |
|---|---|
| H₀ | No difference; the hypothesis tested |
| Type I / II | Reject true H₀ (α, producer’s risk) / accept false H₀ (β, consumer’s risk) |
| Power | 1 − β |
| z critical | 1.645 / 1.96 (5%: one / two tail); 2.33 / 2.58 (1%) |
| z-test | Large n or known σ |
| t-test | Student (Gosset), 1908; small n, unknown σ; df n − 1, n₁ + n₂ − 2, n − 1 pairs |
| ANOVA | Fisher; F = MSB/MSW; df k − 1, N − k |
| χ² | Pearson, 1900; \( \sum(O - E)^2/E \); df (r − 1)(c − 1); E ≥ 5; Yates for 2 × 2; mean = df |
| U-test | Mann–Whitney, 1947; two independent groups; \( U_1 + U_2 = n_1n_2 \) |
| H-test | Kruskal–Wallis, 1952; k groups; χ² with k − 1 df |
| Rank correlation test | Table for small n; \( t = R\sqrt{(n - 2)/(1 - R^2)} \), df n − 2 |
| Report layout | Preliminary pages → introduction → literature → methodology → results → findings and recommendations → appendices, bibliography |
| Report types | Technical vs popular; oral vs written |
| Writing steps | Analysis → outline → rough draft → polish → bibliography → final draft |
Key Takeaways
What’s Next
- Quick Revision: Business Statistics and Research Methods — the whole unit on one page
- Thesis Writing, Referencing and ICT — Paper 1’s treatment of report format and citation styles
- Research Ethics and Plagiarism — honest reporting and the UGC 2018 regulations
- Central Tendency, Dispersion and Skewness — back to the start of the unit for revision