৪৯তম বিসিএস ⎯ পরিসংখ্যান [৯৮১]

One-way ANOVA is used when we want to test whether the means of more than two groups differ significantly, based on one independent factor (categorical variable).

It partitions total variation into between-group and within-group variance.

Two-way ANOVA involves two independent factors.

It allows us to test not only the main effects of each factor but also their interaction effect.

Equal sample size is not always mandatory (though balanced designs are preferred).

The F-statistic in ANOVA is the ratio of between-group variance (treatment effect) to within-group variance (error effect).

A higher F-value indicates stronger evidence that group means differ significantly.

One-way ANOVA requires homogeneity of variances, independence, and approximate normality.

Equal sample sizes are not required (though they make the test more robust).

Interaction checks if the effect of one factor depends on the level of another factor.

Example: The effect of teaching method (Factor A) on scores may vary depending on gender (Factor B).

The LSD test is a post-hoc multiple comparison test applied after ANOVA finds a significant difference.

It helps identify which specific group means differ.

In ANOVA, the F-test checks whether the variance explained by group differences (between-group variance) is significantly larger than the variance within groups (error variance).

It’s essentially testing the overall model fit.

If multiple t-tests are used, the probability of committing Type I error increases.

ANOVA controls the family-wise error rate by testing all group means simultaneously with a single F-test.

A non-significant F-test indicates we fail to reject the null hypothesis, meaning there is no statistical evidence that group means differ.

It doesn’t prove all means are exactly equal—it just means differences are not significant at the chosen significance level.

ANOVA is designed to test whether the means of 3 or more populations differ significantly.

For two means, a t-test is sufficient.

In one-way ANOVA, H₀: μ₁ = μ₂ = … = μk (all group means are equal).

The alternative: at least one differs.

SSB (or SSA) represents variability between different treatment means.

Two-way ANOVA is used when there are two factors (independent variables).

Here: Fertilizer type + soil type.

Two-way ANOVA can check main effects and whether factors interact (combined influence).

ANOVA only tells if at least one mean differs.

To identify specifically, post-hoc tests (e.g., Tukey, LSD) are required.

Source: Live MCQ lecture slide 

ANOVA uses F-statistic = MSB / MSW (variance ratio).

ANOVA is efficient for multiple groups, preventing α-inflation from many t-tests.

Source: Business statistics, SP Gupta, MP Gupta 

Error variance = within-group variation, due to random factors not explained by treatments.

ANOVA tells that means differ, but post-hoc comparisons are needed to find which ones.

Source: Live Mcq lecture slide 

Without replication → only one observation in each cell (factor combination).

Interaction cannot be estimated in this case.

ANOVA only indicates that not all means are equal. It does not specify which groups differ.

Source : Business Statistics, Md.Abdul Aziz

Larger between-group variability relative to within-group error variance increases the F-ratio → more power.

Two-way ANOVA partitions variation into factor A, factor B, their interaction, and error.

Example: Fertilizer type (Urea, Compost, NPK) → each type is a treatment level.

Extraneous variables may bias results if not controlled (e.g., soil moisture in a fertilizer study).

Spearman’s rho is used for ordinal (rank) data or non-normal distributions.

ANOVA decomposes total variation: 
SSB (explained by treatment/factors)

SSW (unexplained/random error)

Response variable = dependent variable being measured.

Treatment/factor = independent variable manipulated.

Estimation and hypothesis testing in ANOVA rely on normally distributed errors.

Without replication (only one observation per cell), interaction cannot be estimated.

Explanation:
df_between = k–1 = 4–1 = 3,
df_within = N–k = 20–4 = 16.

Explanation: ANOVA doesn’t specify which group(s); only that not all means are equal.

Explanation: If F ≈ 1, between variance ≈ within variance → no significant effect.

Explanation: N = abn = 3×2×4 = 24.

Explanation: Interaction df = (A–1)(B–1).

Explanation: Since F > Fcrit, reject H₀ → at least one branch differs.

Explanation:

MSA = SSA / df_between = 120/2 = 60
MSE = SSE / df_within = 60/12 = 5
F = MSA/MSE = 60/5 = 12.0

Explanation:

df_A = a–1 = 3–1 = 2
df_B = b–1 = 4–1 = 3
df_AB = (a–1)(b–1) = 2×3 = 6

Explanation:

k = 4, N = 40 → df_between = 3, df_within = 36
SSB = SST – SSE = 360 – 240 = 120
MSB = SSB/df_between = 120/3 = 40

Explanation: Since 2.8 > 2.76, reject H₀ → at least one diet mean differs significantly.

Explanation: The F statistic = MSB / MSW. MSW is a pooled estimate of variance from all groups. If group variances are not homogeneous, this pooled estimate becomes biased, making the F-test unreliable.

Explanation: If interaction is significant, the effect of one factor depends on the level of the other. Hence, interpreting main effects alone can be misleading.

Explanation: Repeated-measures ANOVA is used when the same subjects are observed under multiple conditions → e.g., repeated blood pressure measurements.

Explanation: A non-significant F means we lack evidence to claim differences; it does not prove the means are identical. It only means observed differences could be due to random variation.

Explanation: Replication provides a basis for error estimation.

Explanation:
One-way ANOVA is generally taught as a method to compare 3 or more means.
But mathematically, if there are only two groups, the F-test and t-test give identical results.
 -Why?
For two groups, the between-groups degrees of freedom = 1.
In this case, the F statistic = (t statistic)².
So, testing with ANOVA or with a two-sample t-test leads to exactly the same decision (same p-value, same significance conclusion).

Explanation: 

ANOVA rests on three major assumptions:

Normality – data within each group are approximately normal.
Homogeneity of variance – all groups have equal variances.
Independence of observations – each observation is independent of others.

Why independence is critical:
In ANOVA, the within-group mean square (MSW) is used as an estimate of error variance (σ²).
This estimate assumes that errors (residuals) are independent random deviations from the mean.
If observations are correlated (not independent), the error variance estimate becomes biased.
 Example:

Suppose we are comparing exam scores of 3 teaching methods.
If students copy from each other or belong to the same study group, their scores are not independent.
The variation within groups would then appear artificially small, making F larger than it should be.
This increases the risk of Type I error (false significance).
 
Contrast with other assumptions:
If normality is slightly violated, ANOVA is fairly robust (especially with large n).
If equal variance is violated, ANOVA may still work if group sizes are equal.
But if independence is violated, the whole logic of variance partitioning (between vs within) collapses.
 
Key takeaway:
Independence ensures that the error term is truly random. Without it, the denominator of the F-test (MSW) is wrong, and the ANOVA results are invalid.

Explanation
Total number of observations = N = k×n.
In ANOVA, total degrees of freedom = N–1.
Why N–1?

Because when calculating total variability (SST), all deviations are measured from the grand mean.
Once the grand mean is fixed, only N–1 deviations are free (one is dependent).
? Example:
Suppose k = 4 groups, each with n = 5 observations → N = 20.

df_total = N – 1 = 20 – 1 = 19.
So the general formula:

df(total) = kn–1 

Explanation
ANOVA assumptions are:

-Independence of observations
-Normality of residuals
-Homogeneity of variances (equal population variances across groups)
Among these, the equal variance assumption is critical.
Why? Because the within-group mean square (MSW) is a pooled estimate of group variances.
If one group has a much larger variance, the F statistic can become misleading.
Example:

Suppose we compare 3 teaching methods with class sizes: 10, 10, 10.
If variances are equal, MSW = good error estimate.
If one group has variance 100 and others ~10, MSW will be inflated, making real differences harder to detect (Type II error).
In unequal sample sizes (unbalanced design), this problem becomes even worse → higher risk of false conclusions.

ANOVA → Analysis of Variance

A statistical method to compare means of 3 or more groups by partitioning total variation into between-group and within-group variation, then testing with an F-test.
ANCOVA → Analysis of Covariance

An extension of ANOVA that adjusts group means by including one or more covariates (continuous variables that may influence the dependent variable).
It combines ANOVA (group comparisons) with regression (covariate adjustment).
 Example:

ANOVA: Comparing average exam scores across 3 teaching methods.
ANCOVA: Comparing the same exam scores but adjusting for students’ prior GPA (covariate).