t-Test, ANOVA, χ²-Test

CSIR-NET / GATE Biotechnology & Life Sciences

L13: t-Test, ANOVA, χ²-Test

Parametric and Non-Parametric Statistical Tests

Complete Theory Notes + Formulas + 10 MCQs

Lecture focus: Student's t-test, ANOVA, Chi-square test, parametric tests, non-parametric tests, assumptions, degrees of freedom, interpretation of P-values, Type I and Type II errors, and selection of appropriate statistical tests.

📚 Index / Table of Contents

1. Introduction to Statistical Hypothesis Testing
2. Parametric Tests
3. Non-Parametric Tests
4. Student's t-Test
5. One-Sample t-Test
6. Independent-Samples t-Test
7. Paired t-Test
8. ANOVA
9. One-Way ANOVA
10. Two-Way ANOVA
11. Post-Hoc Tests
12. Chi-Square Test
13. Chi-Square Goodness-of-Fit Test
14. Chi-Square Test of Independence
15. Type I and Type II Errors
16. Statistical Power
17. How to Select the Correct Test
18. CSIR-NET/GATE Important Points
19. 10 Practice MCQs

1. Introduction to Statistical Hypothesis Testing

Statistical tests are mathematical procedures used to determine whether an observed difference, association, or pattern in experimental data is likely to represent a real biological effect or could reasonably have occurred because of random variation. In biological research, statistical testing is essential because measurements such as enzyme activity, gene expression, cell number, growth rate, protein concentration, metabolite concentration and physiological parameters naturally show variation.

• Population: The complete set of individuals, organisms, cells, observations or measurements about which a researcher wants to make a conclusion.
• Sample: A smaller group selected from the population.
• Parameter: A numerical characteristic of a population, such as population mean μ.
• Statistic: A numerical characteristic calculated from a sample, such as sample mean x̄.
• Null hypothesis (H₀): Usually states that there is no difference, no effect or no association.
• Alternative hypothesis (H₁ or Hₐ): States that a difference, effect or association exists.
• Significance level (α): The maximum probability of rejecting a true null hypothesis that the researcher is willing to tolerate.
• P-value: The probability of obtaining results at least as extreme as the observed results, assuming H₀ is true.
• Critical value: A threshold used to decide whether a test statistic lies in the rejection region.
• Degrees of freedom (df): The number of independent pieces of information available for estimating a parameter or calculating a test statistic.

Exam rule: A P-value smaller than the selected significance level α generally leads to rejection of H₀. A P-value greater than α means that there is insufficient evidence to reject H₀. It is better to say “fail to reject H₀” rather than “prove H₀ true.”

2. Parametric Tests

Parametric statistical tests make assumptions about the probability distribution of the population from which the sample is obtained. Many commonly used biological statistical tests are parametric. The t-test and ANOVA are major examples.

• Parametric tests generally involve population parameters such as mean and variance.
• They are particularly useful for quantitative continuous data.
• Many parametric procedures assume that the observations are approximately normally distributed.
• Independence of observations is usually an important assumption.
• Tests such as t-test and ANOVA often assume equal or approximately equal variances between groups, depending on the exact version of the test.
• Parametric tests can have high statistical power when their assumptions are reasonably satisfied.
• Severe violation of assumptions can make P-values and confidence intervals unreliable.

Common assumptions

• Independence: Measurements should generally be independent unless a paired/repeated-measures design is specifically being analyzed.
• Normality: For many small-sample parametric procedures, the underlying population or relevant residuals should be approximately normal.
• Homogeneity of variance: Groups should have reasonably similar variances for standard versions of several tests.
• Appropriate scale: Data should usually be quantitative and measured on an interval or ratio scale for conventional mean-based tests.

3. Non-Parametric Tests

Non-parametric tests are statistical procedures that generally do not require the same distributional assumptions as classical parametric tests. They are especially useful when data are ordinal, strongly skewed, contain extreme outliers, or do not meet assumptions required by a parametric procedure.

• Non-parametric tests often use ranks rather than raw numerical values.
• They are useful for ordinal data.
• They are useful when normality assumptions are seriously violated.
• They are often more robust against outliers.
• They may have lower statistical power than an appropriate parametric test when parametric assumptions are actually satisfied.
• They should not automatically be considered “better” than parametric tests.

Parametric test	Common non-parametric counterpart	Typical use
Independent t-test	Mann–Whitney U test	Compare two independent groups
Paired t-test	Wilcoxon signed-rank test	Compare paired measurements
One-way ANOVA	Kruskal–Wallis test	Compare more than two independent groups
Repeated-measures ANOVA	Friedman test	Compare more than two related conditions

4. Student's t-Test

The Student's t-test is a statistical test used to compare means. It is especially useful when the population standard deviation is unknown and the sample size is relatively small. The t-test was developed from work associated with William Sealy Gosset, who published under the name “Student.”

• The t-test is mainly used for comparing means.
• The basic test statistic follows a t-distribution under the relevant assumptions.
• The t-distribution has heavier tails than the standard normal distribution.
• As degrees of freedom increase, the t-distribution approaches the standard normal distribution.
• Different versions include one-sample, independent-samples and paired t-tests.

When is a t-test appropriate?

• When the response variable is quantitative.
• When comparison of means is biologically meaningful.
• When observations satisfy the assumptions of the selected t-test.
• When the study design contains two relevant means or one mean compared with a reference value.

5. One-Sample t-Test

A one-sample t-test determines whether the mean of a sample differs significantly from a specified reference or hypothesized population mean.

• Suppose a researcher measures enzyme activity in 15 samples.
• The researcher wants to determine whether the mean activity differs from a known/reference value of 50 units.
• The null hypothesis is usually H₀: μ = μ₀.
• The alternative hypothesis may be H₁: μ ≠ μ₀ for a two-tailed test.

t = (x̄ − μ₀) / (s / √n)

Where:

• x̄ = sample mean
• μ₀ = hypothesized/reference mean
• s = sample standard deviation
• n = sample size
• df = n − 1

CSIR-NET point: For a one-sample t-test, degrees of freedom = n − 1.

6. Independent-Samples t-Test

An independent-samples t-test compares the means of two independent groups. For example, a researcher may compare the mean growth rate of bacteria cultured under treatment A with the mean growth rate under treatment B, provided the observations in the two groups are independent.

• Group 1 and Group 2 consist of different experimental units.
• Measurements in one group are not naturally paired with measurements in the other group.
• The null hypothesis commonly states that the two population means are equal.
• For the classical pooled-variance version, equal population variances are assumed.
• If variances are unequal, Welch's t-test is commonly used.

Basic pooled-variance concept

t = (x̄₁ − x̄₂) / √[s²_p(1/n₁ + 1/n₂)]

Here, s²_p represents the pooled estimate of variance when the equal-variance assumption is appropriate.

df = n₁ + n₂ − 2

Example: If Group A contains 10 independent observations and Group B contains 12 independent observations, the classical pooled independent t-test has df = 10 + 12 − 2 = 20.

7. Paired t-Test

The paired t-test is used when two measurements are naturally linked. Instead of treating all observations in the two conditions as independent, the analysis focuses on the difference within each pair.

• Before-and-after experiments are common examples.
• The same individual measured before and after treatment produces paired observations.
• The same biological sample measured under two conditions may also generate paired data.
• Matched experimental units can create paired observations.
• The analysis calculates a difference for each pair.
• The one-sample t-test is then applied to these differences.

dᵢ = Xᵢ(after) − Xᵢ(before)

t = d̄ / (s_d/√n)

The degrees of freedom are:

df = n − 1

Key distinction: If the same experimental subjects are measured twice, think “paired t-test,” not independent t-test.

Feature	Independent t-test	Paired t-test
Experimental units	Different	Same or matched
Relationship	Independent	Dependent/paired
Main analysis	Difference between group means	Mean of within-pair differences
Typical example	Treatment A vs Treatment B using different samples	Before vs after treatment in same subjects

8. ANOVA: Analysis of Variance

ANOVA stands for Analysis of Variance. Despite its name, the major purpose of ANOVA is to test whether the means of multiple groups are equal. Instead of performing many pairwise t-tests, ANOVA provides an overall test for differences among group means.

• ANOVA is commonly used when comparing three or more groups.
• The null hypothesis generally states that all group means are equal.
• The alternative hypothesis states that at least one population mean differs.
• ANOVA uses an F-statistic.
• The F-statistic is based on a ratio of variance components.
• Between-group variation reflects differences among group means.
• Within-group variation reflects variability among observations within groups.

Why not perform many t-tests?

Suppose a researcher has four treatment groups. Performing separate t-tests for every pair increases the number of statistical comparisons and consequently increases the overall probability of making at least one Type I error. ANOVA first provides an overall test. If the overall test is significant, appropriate post-hoc procedures can then identify which groups differ.

9. One-Way ANOVA

One-way ANOVA is used when one categorical independent factor is being studied across multiple groups. For example, a researcher may compare the effect of four nutrient concentrations on microalgal biomass.

• Independent variable: one categorical factor.
• Dependent variable: quantitative response.
• There may be two or more groups, although ANOVA is especially useful for three or more groups.
• The null hypothesis: μ₁ = μ₂ = μ₃ = ... = μₖ.
• The alternative hypothesis: at least one mean differs.

Variance decomposition

The total variation in the observations can be divided into variation attributable to differences between groups and variation within groups.

>Total Sum of Squares = Between-Group Sum of Squares + Within-Group Sum of Squares

The F-statistic is approximately:

F = Mean Square Between / Mean Square Within

If between-group variation is large compared with within-group variation, the F-statistic becomes large, providing evidence against the null hypothesis.

Degrees of freedom in one-way ANOVA

For k groups and N total observations:

df_between = k − 1

df_within = N − k

df_total = N − 1

Notice that df_total = df_between + df_within.

ANOVA assumptions

• Observations are independent.
• Residuals are approximately normally distributed.
• Variances are approximately homogeneous across groups for the standard one-way ANOVA.
• The response variable is quantitative.
• The experimental design should be appropriate for the intended ANOVA model.

10. Two-Way ANOVA

Two-way ANOVA evaluates the effects of two categorical factors on a quantitative response. It can determine whether each factor has an effect and whether the two factors interact.

For example, suppose a researcher studies the effect of temperature and pH on enzyme activity.

• Factor A: Temperature.
• Factor B: pH.
• Response: Enzyme activity.
• Main effect of temperature: Does temperature influence the response?
• Main effect of pH: Does pH influence the response?
• Interaction: Does the effect of temperature depend on pH?

Interaction is extremely important: A significant interaction means that the effect of one factor changes depending on the level of the other factor.

11. Post-Hoc Tests

A significant ANOVA tells us that not all group means are equal, but it does not by itself identify exactly which pairs differ. Post-hoc tests are used after an appropriate significant omnibus test to investigate specific group differences while controlling the multiple-comparison problem.

• Tukey's HSD: Commonly used for pairwise comparisons after ANOVA.
• Bonferroni adjustment: Adjusts significance thresholds or P-values to control family-wise error.
• Dunnett's test: Particularly useful when several treatment groups are compared with one control.
• Other procedures exist depending on the experimental design and assumptions.

Exam concept: ANOVA answers “Is there evidence that at least one mean differs?” Post-hoc tests help answer “Which groups differ?”

12. Chi-Square (χ²) Test

The chi-square test is primarily used for categorical count data. It compares observed frequencies with frequencies expected under a specified null hypothesis.

• Chi-square tests are based on frequencies or counts.
• The data are usually categorical.
• The test compares observed and expected counts.
• The chi-square statistic is non-negative.
• Large discrepancies between observed and expected frequencies produce larger χ² values.
• Two major applications are goodness-of-fit and test of independence.

χ² = Σ [(O − E)² / E]

Where:

• O = observed frequency
• E = expected frequency

13. Chi-Square Goodness-of-Fit Test

The goodness-of-fit test evaluates whether observed categorical frequencies are consistent with a specified theoretical or expected distribution.

A classical genetics example involves testing whether observed offspring numbers are consistent with an expected Mendelian ratio.

• Observed counts are obtained experimentally.
• Expected counts are calculated from the proposed ratio or theoretical model.
• The χ² statistic quantifies the discrepancy between observed and expected counts.
• A small discrepancy produces a small χ² value.
• A large discrepancy produces a large χ² value.

Expected frequency

If the total number of observations is N and the expected proportion of a category is p:

E = N × p

Degrees of freedom

For a simple goodness-of-fit test with k categories and no additional parameters estimated from the data:

df = k − 1

When parameters are estimated from the data, the degrees of freedom may need additional adjustment. Therefore, the exact model and assumptions should always be considered.

14. Chi-Square Test of Independence

The chi-square test of independence examines whether two categorical variables are associated. For example, a researcher may investigate whether disease status is associated with exposure category.

• The observations are organized into a contingency table.
• Rows represent categories of one variable.
• Columns represent categories of another variable.
• The null hypothesis states that the two categorical variables are independent.
• The alternative hypothesis states that the variables are associated.

Expected frequency in a contingency table

E = (Row Total × Column Total) / Grand Total

Degrees of freedom

df = (r − 1)(c − 1)

where r is the number of rows and c is the number of columns.

Test	Data type	Main purpose
One-sample t-test	Continuous	Compare one sample mean with a reference mean
Independent t-test	Continuous	Compare two independent means
Paired t-test	Continuous	Compare two related measurements
One-way ANOVA	Continuous	Compare multiple independent means
Two-way ANOVA	Continuous	Study two factors and interaction
Chi-square GOF	Categorical counts	Compare observed and expected frequencies
Chi-square independence	Categorical counts	Test association between categorical variables

15. Type I and Type II Errors

Statistical decisions are not completely free from error. Two fundamental errors are Type I and Type II errors.

Type I Error

• Occurs when a true null hypothesis is rejected.
• It is often described as a false positive.
• The probability of Type I error is represented by α.
• A common significance level is α = 0.05.
• If α = 0.05, the researcher accepts a 5% long-run probability of rejecting H₀ when H₀ is actually true, under the conditions defining α.

Type II Error

• Occurs when a false null hypothesis is not rejected.
• It is often described as a false negative.
• The probability of Type II error is represented by β.
• Statistical power is 1 − β.

Reality	Decision: Reject H₀	Decision: Fail to reject H₀
H₀ is true	Type I error	Correct decision
H₀ is false	Correct decision / Power	Type II error

16. Statistical Power

Statistical power is the probability that a statistical test correctly rejects a false null hypothesis under a specified alternative. It is commonly represented as:

Power = 1 − β

Power generally increases when:

• Sample size increases.
• The true effect size becomes larger.
• Measurement variability decreases.
• The significance level α is increased, although doing so also increases the risk of Type I error.
• The study design becomes more efficient.

17. How to Select the Correct Statistical Test

Choosing the correct statistical test is one of the most important skills for competitive examinations and biological research. The decision should be based on the experimental design, type of variable, number of groups, relationship between observations and distributional assumptions.

Quick decision guide

• One quantitative sample vs known value: One-sample t-test.
• Two independent quantitative groups: Independent t-test.
• Two related measurements: Paired t-test.
• Three or more independent quantitative groups: One-way ANOVA.
• Two experimental factors: Two-way ANOVA.
• Categorical observed vs expected counts: Chi-square goodness-of-fit.
• Two categorical variables: Chi-square test of independence.
• Two independent groups with ordinal/non-normal data: Mann–Whitney U test may be appropriate.
• Two paired groups with ordinal/non-normal data: Wilcoxon signed-rank test may be appropriate.
• More than two independent non-normal groups: Kruskal–Wallis test may be appropriate.
• More than two related non-normal groups: Friedman test may be appropriate.

18. CSIR-NET / GATE Important Points

🔥 High-Yield Revision Points

• t-test is mainly used to compare means.
• One-sample t-test compares a sample mean with a reference mean.
• Independent t-test compares two independent means.
• Paired t-test analyzes within-pair differences.
• For a one-sample t-test, df = n − 1.
• For a classical pooled independent t-test, df = n₁ + n₂ − 2.
• ANOVA is used to compare multiple means.
• ANOVA produces an F-statistic.
• F = MS between / MS within.
• Large F suggests between-group variation is large relative to within-group variation.
• One-way ANOVA has one factor.
• Two-way ANOVA has two factors.
• Two-way ANOVA can test interaction between factors.
• ANOVA does not automatically tell which specific groups differ.
• Post-hoc tests are used for pairwise comparisons following an appropriate significant omnibus test.
• Chi-square tests generally analyze categorical frequency data.
• χ² = Σ(O − E)²/E.
• For a simple goodness-of-fit test, df is often k − 1.
• For an r × c independence table, df = (r − 1)(c − 1).
• Expected count in a contingency table = row total × column total / grand total.
• Type I error = rejecting a true H₀.
• Type II error = failing to reject a false H₀.
• α is associated with Type I error.
• β is associated with Type II error.
• Power = 1 − β.
• A P-value is interpreted conditional on the null model; it is not the probability that H₀ is true.
• Statistical significance does not necessarily imply biological significance.
• Sample size, effect size and variability strongly influence statistical power.

19. Parametric vs Non-Parametric Tests: Quick Comparison

Feature	Parametric	Non-parametric
Distribution assumptions	Usually stronger assumptions	Generally fewer distributional assumptions
Typical data	Quantitative	Ordinal, ranks or non-normal quantitative data
Examples	t-test, ANOVA	Mann–Whitney, Wilcoxon, Kruskal–Wallis, Friedman
Summary focus	Often means and variances	Often ranks or distributions
Outlier sensitivity	Can be sensitive	Often more robust
Power	Can be high when assumptions hold	Can be lower when parametric assumptions are actually satisfied

20. Common Mistakes in Statistical Questions

• Mistake 1: Using an independent t-test for paired observations.
• Mistake 2: Performing many unadjusted t-tests instead of an appropriate ANOVA framework for multiple groups.
• Mistake 3: Thinking that a significant ANOVA automatically identifies every pair of groups that differs.
• Mistake 4: Using a chi-square test for continuous measurements such as enzyme concentration without first defining appropriate categorical counts.
• Mistake 5: Confusing Type I and Type II errors.
• Mistake 6: Interpreting P < 0.05 as proof that the experimental hypothesis is true.
• Mistake 7: Thinking that P > 0.05 proves there is absolutely no biological difference.
• Mistake 8: Ignoring experimental design when selecting a statistical test.
• Mistake 9: Ignoring the assumptions behind a statistical method.
• Mistake 10: Confusing statistical significance with biological importance.

21. Practice MCQs – 10 Questions

Q1. Which statistical test is most appropriate for comparing the means of two independent groups when the assumptions of the classical test are reasonably satisfied?

A. Paired t-test B. Independent-samples t-test C. Chi-square goodness-of-fit test D. Friedman test

Correct Answer: B. Independent-samples t-test
The independent t-test compares the means of two independent groups.

Q2. In a one-sample t-test with n = 15 observations, what is the number of degrees of freedom?

A. 14 B. 15 C. 13 D. 16

Correct Answer: A. 14
For a one-sample t-test, df = n − 1 = 15 − 1 = 14.

Q3. Which statistical test is primarily used to compare the means of three or more independent groups?

A. One-way ANOVA B. Paired t-test C. Chi-square test D. Wilcoxon signed-rank test

Correct Answer: A. One-way ANOVA
One-way ANOVA is used to test for differences among multiple group means when one factor is being considered.

Q4. In one-way ANOVA, the F-statistic is calculated as:

A. MS within / MS between B. MS between / MS within C. SS total / SS between D. SS within / df total

Correct Answer: B. MS between / MS within
The ANOVA F-statistic compares variation among group means with variation within groups.

Q5. Which test is particularly appropriate for determining whether observed categorical frequencies differ from expected frequencies?

A. Chi-square goodness-of-fit test B. Paired t-test C. One-way ANOVA D. Pearson correlation

Correct Answer: A. Chi-square goodness-of-fit test
The goodness-of-fit test compares observed categorical counts with theoretically expected counts.

Q6. A researcher measures blood glucose in the same subjects before and after treatment. Which test is generally appropriate for comparing the two measurements if parametric assumptions are satisfied?

A. Independent t-test B. Paired t-test C. Chi-square test D. One-way ANOVA

Correct Answer: B. Paired t-test
The same subjects are measured twice, so the observations are naturally paired.

Q7. Which of the following represents a Type I error?

A. Failing to reject a false null hypothesis B. Rejecting a true null hypothesis C. Accepting a true null hypothesis D. Increasing sample size

Correct Answer: B. Rejecting a true null hypothesis
Type I error is a false-positive decision and is associated with α.

Q8. For an r × c contingency table, the degrees of freedom for a chi-square test of independence are:

A. r + c B. rc C. (r − 1)(c − 1) D. r − c

Correct Answer: C. (r − 1)(c − 1)
For the standard chi-square test of independence, df = (r − 1)(c − 1).

Q9. Which statement about a significant one-way ANOVA is correct?

A. Every group differs significantly from every other group B. At least one population mean differs, under the model and assumptions C. All groups have identical means D. The null hypothesis is automatically proven false in every possible sense

Correct Answer: B. At least one population mean differs, under the model and assumptions
A significant omnibus ANOVA does not specify which groups differ. Appropriate post-hoc analysis may be required.

Q10. Which of the following is the correct relationship between statistical power and Type II error probability β?

A. Power = β B. Power = α + β C. Power = 1 − β D. Power = 1 − α

Correct Answer: C. Power = 1 − β
Statistical power is the probability of correctly rejecting a false null hypothesis under a specified alternative.

22. Final Quick Revision

• t-test → comparison of means.
• One-sample t-test → sample mean vs reference mean.
• Independent t-test → two independent groups.
• Paired t-test → two related measurements.
• ANOVA → comparison of multiple means.
• F-statistic → MS between / MS within.
• One-way ANOVA → one factor.
• Two-way ANOVA → two factors + interaction.
• χ² test → categorical frequency data.
• χ² = Σ(O − E)²/E.
• GOF df → commonly k − 1.
• Independence df → (r − 1)(c − 1).
• Type I error → reject true H₀.
• Type II error → fail to reject false H₀.
• Power → 1 − β.
• Small P-value → evidence against H₀, under the specified model.
• Significant ANOVA → at least one mean differs; post-hoc testing may identify where.

📌 Study Strategy for CSIR-NET/GATE: Memorize the purpose of each test, identify the data type, determine whether observations are independent or paired, remember the important degrees-of-freedom formulas, and understand Type I/Type II errors. Many examination questions can be solved quickly by first identifying the experimental design rather than trying to remember a formula.