Tests Based on Sampling Distributions - I - Study Notes
Chapter Summary
When the sample size is small (n < 30), the sampling distributions of test statistics are far from normal and the large sample tests of Chapter 1 cannot be used. This chapter uses two sampling distributions instead: Student's t-distribution for tests about means, and the chi-square distribution for tests about a variance, independence of attributes and goodness of fit.
Student's t-distribution
- Found by W.S. Gosset (1876-1937), who worked in a brewery in Ireland and published under the pseudonym "Student".
- If \( X \sim N(0,1) \) and \( Y \sim \chi^2_n \) are independent, \( T = X/\sqrt{Y/n} \) has the t-distribution with n degrees of freedom.
- The degrees of freedom of t are those of the corresponding chi-square variable.
Properties
- Symmetrical with mean zero; t ranges from \( -\infty \) to \( \infty \).
- The normal curve is higher in the middle; t has a greater spread and more area in the tails.
- The curve is asymptotic to the X-axis.
- Its shape depends on the degrees of freedom; as they grow it approaches N(0,1).
- It does not depend on population parameters, only on the degrees of freedom.
Applications
| Test | Statistic | d.f. |
|---|---|---|
| Single mean, variance unknown | \( t = \dfrac{\bar{x}-\mu_0}{s/\sqrt{n}} \) | n-1 |
| Two independent means, equal unknown variances | \( t = \dfrac{\bar{x}-\bar{y}}{s_p\sqrt{1/m+1/n}} \) | m+n-2 |
| Paired t-test (correlated samples) | \( t = \dfrac{\bar{d}}{s/\sqrt{n}} \) | n-1 |
The sample standard deviation uses divisor n-1: \( s = \sqrt{\tfrac{1}{n-1}\sum (x_i-\bar{x})^2} \). The pooled standard deviation is \( s_p = \sqrt{\dfrac{(m-1)s_x^2+(n-1)s_y^2}{m+n-2}} \).
Critical values and rejection rules
| \( H_1 \) | Critical value | Reject \( H_0 \) if |
|---|---|---|
| \( \mu \ne \mu_0 \) | \( t_{n-1,\alpha/2} \) | \( |t_0| \ge t_{n-1,\alpha/2} \) |
| \( \mu > \mu_0 \) | \( t_{n-1,\alpha} \) | \( t_0 > t_{n-1,\alpha} \) |
| \( \mu < \mu_0 \) | \( -t_{n-1,\alpha} \) | \( t_0 < -t_{n-1,\alpha} \) |
Worked examples in the book
- Tooth paste sales: n = 26, \( \bar{x} = 216 \), s = 8, \( \mu_0 = 200 \): \( t_0 = 10.20 > t_{25,0.05} = 1.708 \), campaign helped.
- Class scores: n = 10, \( \bar{x} = 85 \), s = 8.756: \( |t_0| = 1.806 < 2.262 \), average 90 accepted.
- Two batches: \( s_p = 3.021 \), \( t_0 = 1.3957 < t_{16,0.005} = 2.921 \), performance equal.
- Batteries: \( s_p = 18.03 \), \( t_0 = 1.15 < 2.060 \), no difference.
- Salesmen before and after training: \( \bar{d} = 2 \), s = 3.94, \( t_0 = 1.6052 < t_{9,0.05} = 1.833 \), no increase.
Chi-square distribution
- Karl Pearson (1857-1936) introduced the chi-square goodness of fit test in a paper of 1900 and founded the first university statistics department at University College, London, in 1911.
- If \( Z \sim N(0,1) \), then \( Z^2 \) is chi-square with 1 degree of freedom.
- The sum of squares of n independent standard normal variables is chi-square with n d.f. (additive property).
- Properties: continuous; a single parameter n; shape depends on n; mean n and variance 2n; if U and V are independent chi-square with \( n_1 \) and \( n_2 \) d.f., U + V is chi-square with \( n_1+n_2 \) d.f.
Test for a population variance
- \( H_0 : \sigma^2 = \sigma_0^2 \); statistic \( \chi_0^2 = \dfrac{(n-1)s^2}{\sigma_0^2} \) with n-1 d.f. (mean unknown).
- If the mean \( \mu \) is known, \( \sum (x_i-\mu)^2/\sigma_0^2 \) with n d.f. is used.
- Book examples: 8 weights, \( \chi_0^2 = 3.375 \) lies between 1.69 and 16.01; n = 9, \( s^2 = 5.4 \), \( \chi_0^2 = 4.8 < 15.507 \); n = 20, \( s^2 = 0.024 \), \( \chi_0^2 = 25.3 > 10.117 \).
Independence of attributes
- Attributes are qualitative characteristics such as literacy or employment status. Cross-classified data form a contingency table.
- Expected frequency \( E_{ij} = \dfrac{R_i \times C_j}{N} \).
- \( \chi_0^2 = \sum\sum \dfrac{(O_{ij}-E_{ij})^2}{E_{ij}} \) with (m-1)(n-1) d.f.
- For a 2 x 2 table: \( \chi_0^2 = \dfrac{N(ad-bc)^2}{(a+b)(c+d)(a+c)(b+d)} \) with 1 d.f.
- N should be greater than 50 and no expected frequency below 5.
- Book examples: age and performance, \( \chi_0^2 = 22.152 > 5.991 \); fathers' education and intelligence, \( \chi_0^2 = 43.269 > 3.841 \).
Goodness of fit
- \( \chi_0^2 = \sum \dfrac{(O_i-E_i)^2}{E_i} \) with \( \sum O_i = \sum E_i \) and k-1-s d.f.
- Classes with \( E_i < 5 \) are pooled with neighbouring classes before counting k.
- Binomial recurrence: \( E(x+1) = \dfrac{n-x}{x+1}\cdot\dfrac{p}{q}\cdot E(x) \). Poisson recurrence: \( E(x+1) = \dfrac{m}{x+1}\cdot E(x) \).
- Book examples: five coins 640 times, \( \chi_0^2 = 0.575 < 9.488 \), binomial fits; defective pens, \( \chi_0^2 = 39.803 > 5.991 \), Poisson does not fit; grades in ratio 4:3:2:1, \( \chi_0^2 = 54.375 > 7.81 \); telephone digits, \( \chi_0^2 = 58.542 > 16.919 \).
More for this chapter
Book Back Questions14 textbook MCQs · solved
Additional MCQs18 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