Module 4: Hypothesis Testing â Notes
(Statistics for Management - QAM601, Semester I; taken from the live AMIGO course, 29 lessons)
Lesson 1: Learning Objectives
Explain the fundamental concepts of hypothesis testing, null and alternate hypothesis creation and Type-I and Type-II errors; learn significance level and critical region, standard error and confidence interval; infer about a single population mean (t and z statistics); compare two populations (mean differences, proportion differences, independent and matched samples); two-population variance ratio inference, ANOVA and hypothesis testing in industry with MS Excel; chi-squared tests for goodness of fit and independence, and an Excel case study.
Lesson 2: 4.1 Fundamentals Concepts of Hypothesis Testing (SLM p.102-103)
Seven numbered-list markers explain the concepts: 1. A hypothesis is a statement about a population's characteristics in the statistical domain; a hypothesis test evaluates how likely the claim is to be valid. 2. A hypothesis test has parts: the null and alternative hypotheses, the test statistic and critical value, which give the P-value and the rejection region (alpha). 3. The null hypothesis (H0) claims the parameter equals the stated value; it is assumed true until shown false. 4. The alternative hypothesis (H1) says the parameter differs from the null in some way (symbols <, > or not equal). 5. The test statistic is the tool used to decide whether to reject H0: the observed sample statistic converted into a standard score assuming H0 is true. 6. The P-value is the probability of obtaining a sample statistic at least as extreme as the observed value (the area to the left or right of the test statistic). 7. The critical value is the standard score that separates the rejection region (alpha) from the rest of the curve.
Lesson 3: 4.1.1 Introduction to Hypothesis Testing (SLM p.103-105)
- A hypothesis test is a decision-making technique based on scientific facts. A result is statistically significant when it is unlikely to have occurred by chance alone, judged against a predefined threshold probability (the significance level). Ronald Fisher coined "test of significance".
- Such tests show whether results contain enough information to cast doubt on conventional wisdom (the null hypothesis), and which study outcomes lead to rejection at a pre-specified significance level. The set of results that lead to rejecting H0 is the critical region. Hypothesis testing is called confirmatory data analysis, versus exploratory data analysis with no pre-set hypotheses; it is a main method of frequentist inference.
- Characteristics of a good hypothesis (accordions): conceptually clear; has empirical referents; specific; related to available research techniques; related to a body of theory.
- Hypothesis testing procedure: the formal process statisticians use to accept or reject statistical hypotheses (assumptions about population parameters). Because studying the whole population is rarely feasible, a random sample is used; if the sample data do not support the hypothesis it is rejected. H0 is the statement about the population parameter; if it is disproved, H1 is accepted. Tests are parametric (assume properties of the population such as sample size and parameters) or non-parametric/distribution-free (assume only nominal or ordinal data).
- Quizzes: Quiz Bank Q1-2.
Lesson 4: 4.1.2 Null and Alternate Hypothesis Creation (SLM p.105)
- Null hypothesis (H0) (4 numbered markers): used to test the researcher's hypothesis; researchers give more weight to negative than positive evidence because negative evidence undermines a hypothesis. It shows the hypothesis's predictions are wrong: it says there is no relationship (a "zero" relationship), e.g. no correlation between efficiency and work commitment. It is non-directional, so testing proceeds in two steps: first whether a relationship exists, then its direction and strength. Experts advise testing H0 to evaluate the research hypothesis indirectly; if the idea has merit the findings should refute H0, and rejecting H0 means accepting the alternative.
- Alternative hypothesis (H1) (4 markers): refutes the null by asserting that the variables are related, e.g. there is a correlation between officers' job devotion and efficiency; it may be directional.
- Quizzes: Quiz Bank Q3-4 (H0 assumes no effect or no difference; a one-tailed alternative suits "a new training programme increases productivity").
Lesson 5: 4.1.3 Type-I and Type-II Errors (SLM p.106)
- A statistically significant result cannot prove a research hypothesis (that would imply 100% certainty); because p-values are probabilities there is always a chance of a wrong conclusion about H0. Every decision has four outcomes: two correct and two errors.
- Type I error (false positive): rejecting a true null hypothesis, i.e. reporting a significant finding that arose by chance. Its probability is the alpha level (the p-value below which H0 is rejected): alpha 0.05 means accepting a 5% chance of being wrong when rejecting H0. A smaller alpha (e.g. 0.01) lowers the Type I risk, but a very low alpha risks missing a true difference (raising the Type II risk).
- Type II error (false negative): failing to reject a false null hypothesis, i.e. concluding there is no significant effect when there is one. Its probability is beta; power = 1 - beta. The Type II risk falls when the test has enough power (e.g. a larger sample size).
- Quizzes: Quiz Bank Q5-6 (increasing the sample size increases power; a Type I error is rejecting a true null hypothesis).
Lesson 6: 4.1.4 Significance Level and Critical Region (SLM p.106-107)
- Level of significance (alpha): the strength of evidence needed in the sample before the null hypothesis is rejected and the effect is called statistically significant; chosen by the researcher before the experiment. It is the probability of rejecting H0 when it is true (alpha of 0.06 = 6% chance of wrongly concluding a difference exists). Lower significance levels need more convincing proof before H0 is rejected.
- In testing, the p-value is compared with alpha: if p < alpha, reject H0 and call the effect statistically significant (enough evidence in the sample to reject H0 for the population).
- Critical region (rejection region): the set of test-statistic values for which H0 is rejected; if the observed statistic falls in it, reject H0 and accept H1. It fixes how far the sample statistic must be from the null value to count as exceptional. The "best" critical region has the lowest probability of Type I or Type II error (the uniformly most powerful, UMP, region with the largest power function).
- Quizzes: Quiz Bank Q7-8 (alpha = probability of rejecting H0 when true; for a two-tailed test with alpha 0.05 each tail is 0.025).
Lesson 7: 4.1.5 Standard Error (SLM p.107-108)
- The standard error of a statistic is the standard deviation of its sampling distribution (or an estimate of it); for the sample mean it is the standard error of the mean.
- Four points: it increases when the population standard deviation (variance) increases; it decreases as sample size increases (sample means cluster nearer the true population mean); it shows how accurate a sample mean is likely to be compared with the true population mean; a larger standard error means means are more spread out so any one mean is more likely to misrepresent the population mean.
- Quizzes: Quiz Bank Q9-10 (SEM = standard deviation of the sampling distribution of the sample mean; increasing the sample size decreases it).
Lesson 8: 4.1.6 Confidence Interval (SLM p.108)
- A confidence interval is a range of plausible values for an unknown parameter (e.g. the mean), computed from the observed data, with an associated confidence level. The confidence level is the proportion of such intervals, over repeated sampling, that contain the true parameter; e.g. at 90%, 90% of intervals from repeated samples contain the parameter. The level is set before looking at the data; 95% is most common, with 90% and 99% also used.
- Width depends on: sample size (larger sample gives a narrower, better estimate), confidence level (higher level gives a wider interval) and variability in the sample.
- Quizzes: Quiz Bank Q11-12 (a higher confidence level makes the interval wider; use the t-distribution when the population standard deviation is unknown).
Lesson 9: 4.2 Inference: Population (SLM p.109)
- Four markers: a researcher usually cannot study the whole population; a sample is examined to find the true population parameters; statistical inference is applying techniques to sample data to estimate population parameters; its basic presumption is that every member of the population has an equal chance of being selected in a sample.
- Results can still be applied to a larger group when the sample was not random, provided it is representative. Inference assumes sample estimates have a normal (bell-shaped) distribution; the sampling-distribution idea matters even when the population itself is not normal, because the shape of the distribution of sample estimates matters more than that of individual values.
Lesson 10: 4.2.1 t-Statistic for a Single Population Mean (SLM p.109-110)
- When the population standard deviation is unknown, the sample estimate s is used and the sample size n determines the sampling distribution. Use the t-test if the parent population is normal and n < 30 (small-sample case).
- t-alpha denotes the t statistic with cumulative probability (1 - alpha); e.g. cumulative probability 0.95 gives alpha 0.05 and t(0.05). Its value depends on degrees of freedom: t(0.05) = 2.92 for 2 df but 1.725 for 20 df.
- Worked example (ABC light bulbs): claimed mean life 300 days; 15 bulbs give mean 290 and standard deviation 50. t = (x-bar - mu) / (s / sqrt n) = (290 - 300) / (50 / sqrt 15) = -10 / 12.91 = -0.7746, df = 15 - 1 = 14; the cumulative probability is 0.226, so there is a 22.6% chance that 15 random bulbs average 290 days or less if the true life is 300 days.
- Quizzes: Quiz Bank Q13-14.
Lesson 11: 4.2.2 z-Statistic for a Single Population Mean (SLM p.110-112)
- A z-test tests whether population means differ when variances are known and the sample is large; test statistics are assumed normal. Five markers: it checks whether two population means differ when variances are known and n is large; it applies when the test statistic follows a normal distribution; a z-statistic (z-score) is the number that results; z-tests are closely related to t-tests, but t-tests suit small samples; t-tests assume the standard deviation is unknown, z-tests assume it is known.
- Standardising values to z-scores lets researchers (1) find the probability of a score in a standard normal distribution and (2) compare scores from different samples with different means and standard deviations.
- Conditions: the population is normal and/or the sample is large (n > 30). Z = (x-bar - mu) / (sigma / sqrt n).
- Example (lumber): supposed mean length 8.5 ft; 61 pieces average 8.3 ft with standard deviation 1.2: Z = (8.3 - 8.5) / (1.2 / sqrt 61) = -1.3. If the z value is far from zero the null is unlikely to be valid and is rejected; otherwise it cannot be discarded.
- Quizzes: Quiz Bank Q15-16.
Lesson 12: 4.2.3 Population Proportion Hypothesis Testing (SLM p.112-114)
- Independent vs dependent samples: samples from two populations are independent when there is no correlation between them (one measurement per individual); they are dependent (paired) when each measurement in one sample is matched with a specific measurement in the other (two measurements per individual, e.g. twins or spouses). The difference of interest for independent means is mu1 - mu2. Examples: 15 men vs 15 women watching TV = independent; 15 husband-wife couples = paired.
- Proportion test: H0: pi = pi0 against Ha: pi not equal pi0 (two-tailed) or pi > or < pi0 (one-tailed), with the standard test statistic for a proportion.
- Examples: (a) an apple wholesaler claims only 4% defective; a sample of 600 has 36 defective (6%), so test the claim; (b) 470 tails in 1,000 throws of an unbiased coin: is the difference from 0.5 due to sampling fluctuation? (worked solutions are images).
- Quizzes: Quiz Bank Q17-18 (businesses test proportions to see whether a sample proportion differs significantly from a claimed one; a p-value greater than alpha indicates weak evidence against the null).
Lesson 13: 4.3 Inference: Comparing Two Populations (SLM p.114-115)
- Differences in the estimates for two groups can be computed and reported as a confidence interval for the difference (upper and lower bounds). If the interval contains 0 there is no discernible difference between the groups; if it excludes 0, one population is at higher (or lower) risk than the other. The difference may reflect an excess or reduced risk due to exposure.
- Risk disparity example: the high-pollution city has 5.51 extra deaths per 1,000 person-years; lowering its pollution to the low-pollution city's level could cut deaths by 5.51 per 1,000 person-years.
Lesson 14: 4.3.1 Inference on Two Mean Differences (SLM p.115-116)
- (a) Population mean known: for a random sample from a normal population, H0: mu = mu0 against Ha: mu not equal mu0 (two-tailed); reject H0 at the 5% level if |z-calculated| > 1.96. Example: a company claims bikes give 40 km/l; 20 bikes average 42 km/l with sigma 2 km/l, so z-calculated exceeds 1.96 and H0 is rejected.
- (b) Population standard deviation unknown: use the sample estimate s. Two options (markers): use the t-test if the parent population is normal and n < 30 (one- or two-tailed); use the standard normal test if n >= 30 (large sample; the population may or may not be normal). Example: 40 shops with mean daily sales Rs 528 and standard deviation Rs 600, testing H0: mu = 400 at 5%; since n > 30 a normal test applies and the computed value is below 1.96, so the claim is not contradicted.
- Quizzes: Quiz Bank Q19-20 (paired t-test for before/after training measurements; Welch's t-test for independent samples with unequal variances).
Lesson 15: 4.3.2 Inference on Two Proportion Differences (SLM p.116-118)
- A test of two proportions is like a test of two means, with the parameter p instead of mu. p-hat = x/n; with two samples we expect (p1-hat - p2-hat) to be near (p1 - p2). The null hypothesis is p1 = p2; under H0 both equal a common proportion p, which is estimated by a pooled proportion from the two samples.
- Application: comparing proportions in one category across two populations. Example: of 131 women, 120 would return a misdelivered $100 bill; of 69 men, 52 would. The 95% confidence interval for the difference (men minus women) is -0.2746 to -0.0502; both ends are negative, so a greater percentage of women than men would return it.
- Quizzes: Quiz Bank Q21-22 (failing to reject H0 means not enough evidence of a difference; p > alpha means weak evidence against H0).
Lesson 16: 4.3.3 Independent and Matched Samples (SLM p.118-119)
- Matched samples (matched pairs, paired or dependent samples) are paired so that participants share every characteristic except the one under review; a participant can be a person, object or thing. Matched pairs often assign one person to a treatment group and another to a control group (matching). The "pairs" may be the same individuals: the same participants measured before and after an intervention, or measured twice under two different interventions.
- An independent sample is the opposite: unrelated groups, typically chosen at random (simple random sampling).
- Quizzes: Quiz Bank Q23-24 (before/after training is a matched design; the paired t-test analyses matched means).
Lesson 17: 4.3.4 Two Population Variances Ratio Inference (SLM p.119-121)
- Checking the equal-variances assumption is a key step before using pooled variances. The F-test tests whether two population variances are equal, in a two-tailed form (variances differ) or one-tailed form (one variance is larger or smaller, but not both); the choice depends on the problem (e.g. whether a new process is less variable).
- F statistic: f = [s1^2 / sigma1^2] / [s2^2 / sigma2^2], with numerator df v1 = n1 - 1 and denominator df v2 = n2 - 1. Worked example with 12 men and 7 women, s1 = 35, sigma1 = 30, s2 = 45, sigma2 = 50: f = (35^2/30^2) / (45^2/50^2) = 1.361 / 0.81 = 1.68 (v1 = 6, v2 = 11 when the women's data is in the numerator); reversing gives 0.595 (v1 = 11, v2 = 6).
- Assumptions: populations roughly normal (bell-shaped) and independent samples. Tips (markers): to make a right-tailed test put the larger variance in the numerator; for two-tailed tests divide alpha by two before finding the critical value; square standard deviations to obtain variances; if the F table lacks your degrees of freedom use the larger critical value (this lowers Type I error).
- Quizzes: Quiz Bank Q25-26 (the F-distribution tests the ratio of two variances; an F-test checks whether the spread of two populations differs).
Lesson 18: 4.4 ANOVA (SLM p.121)
- ANOVA (analysis of variance), created by Ronald Fisher: one-way ANOVA tells you whether there are statistical differences between the means of three or more independent groups. A marketer can use it to test a hypothesis: the null hypothesis is that the group means are identical; a statistically significant result shows a difference between populations.
Lesson 19: 4.4.1 Analysis of Variance (ANOVA) (SLM p.121-122)
- Variance is the average squared deviation of data points from the mean. In a sample, sum the squared deviations from the mean and divide by n - 1 (sample variance); for a population of N observations divide by N.
- F statistic in ANOVA = between-group variance / within-group variance. Key assumption: variances across groups are equal (plus normality and independent samples).
- Quizzes: Quiz Bank Q27-28.
Lesson 20: 4.4.2 Hypothesis Testing in Industry (MS Excel) (SLM p.122-123)
- Scenario: a manufacturer of metal car-engine components investigates whether the mean dimensions have changed significantly. Steps: (1) collect data (30 components in an Excel sheet); (2) set H0: mu = mu0 (mean dimension unchanged) and H1: mu not equal mu0; (3) compute the sample mean and standard deviation with AVERAGE() and STDEV.S(); (4) set the significance level alpha (usually 0.05); (5) find the critical t-value with T.INV.2T() or T.INV() using n - 1 degrees of freedom; (6) reject H0 if |t calculated| exceeds the critical t, otherwise do not reject; (7) interpret: rejecting H0 means enough evidence of a significant change in mean dimensions, not rejecting means insufficient evidence.
- Excel's built-in functions make these calculations easy and support defensible process decisions.
- Quizzes: Quiz Bank Q29-30 (Excel "t-Test: Paired Two Sample for Means" does a paired t-test; two-tailed z p-value = 2*NORM.S.DIST(-ABS(z), TRUE)).
Lesson 21: 4.5 Chi-Squared Tests
- Introductory lesson for the chi-squared family of tests (goodness of fit and independence), detailed in L22.
Lesson 22: 4.5.1 Chi-squared Tests for Goodness of Fit and Independence (SLM p.123-127)
- The chi-square test fits a theoretical frequency distribution to observed data. Uses: (1) comparing sample proportions, (2) testing independence in a contingency table, (3) goodness of fit. Degrees of freedom v = n - k.
- Assumptions (accordions): simple random sample; adequate sample size; expected count of at least 5 in each cell (otherwise use Yates's continuity adjustment); independence of observations (use McNemar's test for matched pairs).
- Conditions (markers): two observed data sets, or one observed and one expected set; same sample size for both sets; every cell count at least 5; categorical variables.
- Example 1 (tires by shift): Chi-Sq = 8.647, DF = 4, p = 0.071; 3 cells have expected count below 5, so the result is not significant.
- Example 2 (gender vs preferred condiment): chi-square 2.95, df 2, p = 0.2288, so the two are not related at the 10% level.
- Quizzes: Quiz Bank Q31-32.
Lesson 23: 4.5.2 Case Study: Using MS Excel (SLM p.127-133)
- Sports hypothesis testing (Arsenal football case): a one-sample t-test checks whether a team's recent average goals per game (about 2.3) differs from a historical benchmark (about 1.5-2.1). Data were pulled via the football-data.org API with Python.
- A null distribution was simulated by drawing 500 random samples of 38 goals without replacement and taking each mean; the 95% interval was about 1.87 to 2.42 goals per game.
- p-value = share of simulated means above the observed value = 0.32, above alpha 0.05, so H0 is not rejected: the higher average is coincidence.
- Excel: standard error of the mean = STDEV.S(range)/SQRT(COUNT(range)); T.TEST compares two batches' means.
- Quizzes: Quiz Bank Q33-34.
Lesson 24: Summary
- A hypothesis test is a decision-making technique; a result is statistically significant when it is unlikely under chance alone given the significance level.
- A Type I error (false positive) is rejecting a true null hypothesis.
- Standard error shows how accurate a sample mean is likely to be; larger SE means the means are more spread out.
- When sigma is unknown the sample estimate s is used. A z-test is used when the sample is large and variances are known; the test statistic is assumed normal.
Lesson 25: Glossary
- Confidence Interval: a range of values where the true value lies, computed from observed data.
- Z-Test: tests whether two population means differ when variances are known and the sample is large.
- p-Value: probability of outcomes at least as extreme as observed, assuming H0 is true.
- Simple random sample: each member of the population has an equal chance of selection.
- Degrees of Freedom (d.f.): the number of independent values free to vary in a set of observed frequencies.
Lesson 26: Check Your Understanding
- 20-question formative quiz (passing mark 0%); full set with answers in Quiz Bank Q35-54.
Lesson 27: Exercise (notebook questions)
- Characteristics of a hypothesis in scientific research. 2. Level of significance: meaning, determination, implications. 3. Factors affecting confidence-interval width (sample size, confidence level, variability; precision and standard error). 4. Population-proportion hypothesis testing with independent samples. 5. Comparing two populations using confidence intervals for the difference in estimates.
Lesson 28: Learning Activities
- Group discussion: null and alternative hypotheses for a new teaching method vs the traditional method in maths (H0: no significant difference in performance). 2. Group discussion: standard error (definition, significance) in a sleep duration vs academic performance study.
Lesson 29: Further Readings and Bibliography
- Levin, Rubin et al., Statistics for Management (Pearson, 7th ed.); Mann, Introductory Statistics (Wiley); James et al., Intro to Statistical Learning (Springer); Srivastava et al., Quantitative Techniques for Managerial Decision Making; Levin and Kirkpatrick, Quantitative Approaches to Management; Budnick et al., Principles of Operations Research; Sharma, Operations Research.
⨠Quick Revision Pack (Module 4)
60-second summary
- Hypothesis test steps: state H0 and H1 â choose Îą â pick the test and compute the statistic â compare with the critical value (or p with Îą) â decide â state in plain words.
- H0 = no effect / no difference (status quo). H1 = what the researcher suspects. p < Îą â reject H0.
- Errors: Type I = reject a true H0 (probability Îą, false positive). Type II = fail to reject a false H0 (probability β, false negative). Power = 1 - β; bigger sample â more power.
- Standard error falls as n rises. Confidence interval widens with a higher confidence level, narrows with larger n.
- Which test? Ī known or n âĨ 30 â z; Ī unknown and n < 30 (normal population) â t; two variances â F; three or more means â ANOVA (F = between / within variance); categorical data â chi-square; before/after on the same units â paired t.
Diagrams
Choosing the right test
flowchart TB
S["What are you testing?"] --> M{"Means, proportions,
variances or categories?"}
M -->|"One mean"| O{"Ī known or n >= 30?"}
O -->|"Yes"| Z["z-test"]
O -->|"No (n < 30)"| T["t-test (df = n - 1)"]
M -->|"Two means"| TW{"Matched pairs?"}
TW -->|"Yes"| PT["Paired t-test"]
TW -->|"No"| IT["Independent t / z test"]
M -->|"Proportion(s)"| PR["z-test for proportion(s)"]
M -->|"Two variances"| F["F-test"]
M -->|"3+ means"| A["ANOVA"]
M -->|"Categories / independence"| C["Chi-square"]
The four outcomes of a decision
| H0 is true | H0 is false | |
|---|---|---|
| Reject H0 | Type I error (ι) | Correct (power = 1 - β) |
| Do not reject H0 | Correct | Type II error (β) |
flowchart LR
A["Compute test statistic"] --> B{"In rejection region
or p < Îą ?"}
B -->|"Yes"| C["Reject H0
(significant)"]
B -->|"No"| D["Fail to reject H0
(not enough evidence)"]
Memory tricks
Type I vs Type II: Type I = Innocent convicted (rejecting a true H0, false alarm). Type II = guilty goes free (missing a real effect).
"If p is low, H0 must go." p < Îą â reject H0. "If p is high, H0 stays nearby."
Never say "accept H0" â say "fail to reject H0" (the data are not strong enough).
z vs t: "Big sample or known Ī = z (zee, large). Small sample + unknown Ī = t (tiny)."
Degrees of freedom: one-sample t: n - 1; two-sample F: (n1 - 1, n2 - 1); chi-square contingency: (rows - 1)(columns - 1).
Chi-square needs expected counts âĨ 5 in every cell; matched pairs need a different test (McNemar).
Formula sheet
| Test | Statistic |
|---|---|
| z for a mean | z = (xĖ - Îŧ) / (Ī / ân) |
| t for a mean | t = (xĖ - Îŧ) / (s / ân), df = n - 1 |
| z for a proportion | z = (pĖ - p0) / â(p0 (1 - p0) / n) |
| Standard error (mean) | SE = Ī / ân |
| Confidence interval | xĖ Âą z (or t) x SE |
| F (two variances) | F = s1² / s2² (larger over smaller for right-tail) |
| ANOVA | F = MS between / MS within |
| Chi-square | β = ÎŖ (O - E)² / E |
Practice problems (with answers)
- z-test: claim Îŧ = 500 g, Ī = 20, n = 64, xĖ = 495. Îą = 0.05 two-tailed. â z = (495 - 500)/(20/8) = -2.0; |z| > 1.96 â reject H0.
- t-test: claim Îŧ = 300 days, n = 15, xĖ = 290, s = 50. â t = -10/(50/â15) = -0.775, df = 14; far inside the acceptance region â do not reject H0.
- Proportion: wholesaler claims 4% defective; sample 600 has 36 defective. â pĖ = 0.06; SE = â(0.04 x 0.96/600) = 0.0080; z = 0.02/0.0080 = 2.5 > 1.96 â reject the claim.
- Confidence interval: xĖ = 50, s = 10, n = 100, 95% (z = 1.96). â SE = 1; CI = 50 Âą 1.96 = (48.04, 51.96).
- F-test: s1² = 36 (n1 = 13), s2² = 20 (n2 = 11). â F = 36/20 = 1.8, df = (12, 10); compare with the table value.
- Chi-square: observed 30, 20, 50; expected 33.3 each. â β = (30-33.3)²/33.3 + (20-33.3)²/33.3 + (50-33.3)²/33.3 = 0.33 + 5.31 + 8.37 = 14.0, df = 2 â significant at 5% (critical 5.99).
- Power: you want to lower the chance of a Type II error. â Increase sample size (or raise Îą, or reduce variability).
- Short answer (5 marks): Difference between independent and paired samples? â Independent = unrelated groups; paired/matched = each observation linked to a specific partner (before/after on the same person), analysed with a paired t-test.