Sampling Techniques and Statistical Inference - Study Notes
1. Basic concepts of sampling
- Sampling: the process of selecting a sample from a population. Examples: tasting a spoonful of food, testing a few drops of blood, checking a handful of grain.
- Population: the group of individuals or items under study. It is finite if its units are countable (weights of Class XII students in a school) and infinite if uncountable (grains in a sack, germs in a patient's body).
- Sample: a group selected from the population so that it represents it; the number of units in it is the sample size.
- Parameter: a statistical constant of the population, such as \( \mu \) or \( \sigma^2 \).
- Statistic: any measure computed from a sample, such as \( \bar{x} \) or \( s^2 \). Parameters are usually unknown and are estimated from statistics.
2. Types of sampling
Sampling is either non-random (non-probability) or random (probability). The unit studies random sampling. Dr. Yates: every member of a parent population has had an equal chance of being included. Harper: a random sample is one in which every item in the population has an equal chance of being included.
(i) Simple random sampling
- Every unit has an equal and independent chance of selection. Sampling is done without replacement.
- Probability of drawing a unit at the first draw is \( \frac{1}{N} \), at the second draw \( \frac{1}{N-1} \), and so on.
- Lottery method: number the units on identical slips, shuffle and draw. Also called unrestricted random sampling; not usable for an infinite population.
- Random number tables: L.H.C. Tippett, Fisher and Yates, Kendall and Babington Smith, Rand Corporation. Tippett's four-digit table is the most popular.
- Merits: no personal bias; saves time, money and labour; needs little advance knowledge of the population. Demerits: needs a complete up-to-date list; a small sample may not be representative.
(ii) Stratified random sampling
- A heterogeneous population is divided into homogeneous, non-overlapping sub-groups called strata; a random sample is drawn from each.
- Proportional allocation: sample from a stratum = stratum size \( \times \frac{n}{N} \). Example 8.5: 195, 190 and 115 people out of 500 give 26, 26 and 16 of a sample of 68.
- Merits: more representative; can be kept small without losing accuracy; easy to administer. Demerits: stratifying needs money, time and experience; faulty strata cause bias and more variability.
(iii) Systematic sampling
- Choose the first unit at random from the first \( k \) units, then every \( k \)-th unit.
- Sampling interval \( k = \frac{N}{n} \). For 20 items out of 6,000, \( k = 300 \); a start at 50 gives 50, 350, 650, ..., 5750.
- Merits: simple, spreads the sample evenly, saves time. Demerits: not truly random; if \( N \) is not a multiple of \( n \), \( k \) is not an integer.
3. Sampling and non-sampling errors
- Sampling errors arise by chance in the course of enumeration and are inherent in sampling: faulty selection by a defective technique, substitution of a convenient sample, faulty demarcation of sampling units.
- Non-sampling errors arise from human factors: negligence of investigators or respondents, untrained investigators, a wrong questionnaire, a wrong statistical measure, incomplete investigation.
4. Sampling distribution and standard error
- From a population of size \( N \) there are \( {}^{N}C_n = \frac{N!}{n!(N-n)!} \) possible samples of size \( n \). The values of a statistic over all these samples form its sampling distribution.
- The standard deviation of the sampling distribution is the standard error (S.E.).
| Statistic | Standard error |
|---|---|
| Sample mean | \( \frac{\sigma}{\sqrt{n}} \) |
| Sample proportion | \( \sqrt{\frac{PQ}{n}} \) |
| Sample standard deviation | \( \frac{\sigma}{\sqrt{2n}} \) |
| Sample variance | \( \sigma^2 \sqrt{\frac{2}{n}} \) |
Example 8.8: a die thrown 9,000 times, \( P = \frac{1}{3} \): S.E. \( = \sqrt{\frac{PQ}{n}} = \sqrt{\frac{(1/3)(2/3)}{9000}} \approx 0.005 \).
5. Estimation
- Estimation: obtaining the most likely value of a parameter using a statistic. Estimator: the statistic used. Estimate: its observed value.
- A good estimator is unbiased (\( E(T_n) = \theta \)), consistent (\( T_n \to \theta \) in probability as \( n \to \infty \)), efficient (efficiency \( E = \frac{V_1}{V_2} \), a relative property) and sufficient (uses all the information in the sample).
- Point estimation gives a single value; interval estimation gives limits within which the parameter is expected to lie.
- Confidence interval for the mean: \( \bar{x} - Z_{\alpha/2} \frac{\sigma}{\sqrt{n}} < \mu < \bar{x} + Z_{\alpha/2} \frac{\sigma}{\sqrt{n}} \); \( 1 - \alpha \) is the confidence coefficient. If \( \sigma \) is unknown, use \( s \).
6. Testing of hypothesis
- Initiated by J. Neyman and E.S. Pearson; also called statistical decision making.
- Null hypothesis \( H_0 \) (R.A. Fisher): tested for possible rejection under the assumption that it is true. Alternative \( H_1 \): complementary to \( H_0 \); \( \mu \neq \mu_0 \) is two-tailed, \( \mu > \mu_0 \) right-tailed, \( \mu < \mu_0 \) left-tailed.
- Type I error: rejecting \( H_0 \) when true. Type II error: accepting \( H_0 \) when false.
- Level of significance \( \alpha \): probability of Type I error, usually 5% or 1%, fixed in advance. Critical region: region leading to rejection of \( H_0 \). Critical value: separates rejection and acceptance regions.
- Steps: set \( H_0 \) and \( H_1 \), choose \( \alpha \), compute \( Z = \frac{\bar{x} - \mu}{\sigma/\sqrt{n}} \), compare: if \( |Z| < Z_\alpha \) accept \( H_0 \); if \( |Z| > Z_\alpha \) reject it. A sample is large if \( n > 30 \).
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