Tests Based on Sampling Distributions II - Study Notes
Chapter summary
Chapter 3 of 12th Statistics studies inference about population variances and about three or more population means. Both use the F-distribution, named after R.A. Fisher (1890 to 1962), who also developed the analysis of variance (ANOVA).
F-distribution
- If X and Y are independent chi-square variates with m and n degrees of freedom, F = (X/m)/(Y/n) follows the F-distribution with (m, n) d.f.
- The F-statistic is the ratio of two sums of squares of deviations, each divided by its d.f.; it can also be defined as the ratio of two mean square errors.
- Applications: (i) testing the equality of variances of two normal populations, (ii) testing the equality of means of k (more than 2) normal populations, (iii) analysis of variance for two-way classified data.
Test for two normal population variances
- Assumptions: the populations are normal and the two samples are independent. If the populations are not normal, the F-test may not be used.
- Null hypothesis: the two population variances are equal. The alternative may be less than, greater than or not equal.
- Sample variances use divisors m-1 and n-1. Test statistic F0 = sX squared / sY squared, with (m-1, n-1) d.f.
- Upper critical value: f with (m-1, n-1) at level alpha from the table. Lower critical value: f with (m-1, n-1) at 1-alpha = 1 / f with (n-1, m-1) at alpha.
- For a two-sided alternative at 10% the book uses the 0.05 and 0.95 points, because 0.025 and 0.975 tables are not provided.
- When the population means are known, divisors m and n are used and the statistic follows F with (m, n) d.f.
- Because it uses a ratio of variances, the F-test is also called the variance ratio test.
| Example | Working | Decision |
|---|---|---|
| 3.1: sizes 9 and 8, sums of squares 160 and 91 | 20/13 = 1.54; limits 0.286 and 3.73 | H0 not rejected |
| 3.2: smokers 36, non-smokers 10 | F0 = 3.6 against 2.19 | H0 rejected |
| 3.3: schools A and B | 78.5/128 = 0.613 against 0.291 | H0 not rejected |
Analysis of variance (ANOVA)
- ANOVA tests the equality of means of more than two populations, for example the average yields of several crop varieties. Fisher called it the separation of variance ascribable to one group of causes from the variance ascribable to other groups.
- Assumptions: observations are normal, experimental units are assigned to treatments at random, observations are independent, and population variances are equal but unknown.
- Variation due to assignable causes can be detected and controlled; variation due to chance causes is beyond human control.
One-way ANOVA
- One factor at k levels (treatments); treatment i has ni observations and n is the total.
- H0: all k population means are equal. H1: at least one pair of means differs.
- C.F = G squared / n; TSS = sum of all squares - C.F; SST = sum of (treatment total squared / ni) - C.F; SSE = TSS - SST.
- d.f.: total n-1, treatments k-1, error n-k. MST = SST/(k-1), MSE = SSE/(n-k), F0 = MST/MSE.
- Merits: simple layout. Demerits: equal variances needed; normality is hard to verify.
Two-way ANOVA
- The grouping factor is the treatment and the control factor is the block. Blocks reduce the error variance, so the test is more accurate than one-way ANOVA.
- With k treatments and m blocks, n = km. SST = sum of (treatment total squared / m) - C.F; SSB = sum of (block total squared / k) - C.F; SSE = TSS - SST - SSB.
- d.f.: treatments k-1, blocks m-1, error (k-1)(m-1), total n-1.
- F0t = MST/MSE and F0b = MSB/MSE; reject a null hypothesis when its F exceeds the critical value.
- Merit: any number of blocks and treatments. Demerits: many treatments make blocks hard to keep homogeneous; a missing value must be estimated.
| Basis | One-way | Two-way |
|---|---|---|
| Independent variables | One | Two |
| Compares | Three or more levels of one factor | Three or more levels of two factors |
| Observations per group | Need not be equal | Must be equal |
More for this chapter
Book Back Questions20 textbook MCQs · solved
Additional MCQs15 extra MCQs · solved
Practice TestInteractive · instant score
Book Back TestTest yourself on the textbook set
Additional MCQ TestTest yourself on the extra set
Formula SheetAll key formulas