Sampling

• A research study can collect data from:

  1. Every unit of the universe (census study)

  2. A small number taken from the whole (sample study)

• Census study is more reliable as it surveys every unit of the universe.

• Exhaustive survey of the universe is sometimes possible but often limited by time, place, and expense.

• Conditions like time, place, and psychology of groups are always changing, affecting survey outcomes.

• In some problems, such as business surveys, late submission of results can render them useless.

• Measuring the whole series is sometimes possible but often unnecessary and undesirable.

• Examples:

  • Not all rice or grapes need to be examined before purchasing.

  • The whole sambar curry need not be tested.

  • Whole blood need not be tested; testing a sample is enough.

• Even with sufficient time and resources, examining every part may be impracticable (e.g., contents of a mine cannot be fully examined unless completely used up).

• Researchers are usually forced to use a sample, a cross-section or small subgroup representative of the larger universe.

• This cross-section is called a representative sample.

• Sampling involves choosing at random an adequate number of items from a large population to study.

• Following the Law of Statistical Regularity, sampling helps find out the characteristics of the whole population.

Objective of Sampling

• The main objective of sampling is to get an accurate picture of the whole universe by examining a portion that possesses the same characteristics as the entire universe.

• Sampling aims to economize on money and time.

• Another objective is to determine the reliability of estimates obtained from the sample.

• Inference from the sample to the universe can only be expressed in terms of probabilities, not with absolute certainty or mathematical proof.

• The entire theory of sampling is closely linked to the theory of probability.

Essentials of Sampling

• Sample must be representative of the entire universe, containing characteristics in proportion to their presence in the whole universe.

• The value of a sample depends on how well it represents the entire universe (e.g., a teaspoon of sambar tested by a housewife to check the entire pot).

• Sample size should be adequate for making accurate generalizations.

• A proper estimate of sampling error is necessary to determine the adequacy of sample size for the desired degree of accuracy.

• The number of samples is directly proportional to the square of the sampling error and inversely proportional to the square within which the result should be correct.

• A sample that is not representative is called a biased sample.

• Causes of bias include:

  • Instruments used

  • Personal qualities of the observer

  • Defective techniques or other causes

• Bias is difficult to eliminate but can be reduced by proper care.

• Human beings are generally poor instruments for random selection due to personal choice or judgment creeping in.

• The Literary Digest 1936 presidential poll is a dramatic example of consequences of a biased sample despite a very large sample size.

• The Fortune Poll, with only 4,500 cases, predicted Roosevelt’s popular vote with just 1% error, highlighting that sample quality is more important than size.

• Reasons for the Literary Digest poll failure in 1936:

  1. Mailing lists largely from telephone directories and automobile registration lists, overweighting upper socio-economic classes.

  2. Ballots sent by mail; higher income groups more likely to return ballots than lower income groups, combined with selective factors like protest voting.

  3. Political alignments shifted sharply in 1932 and 1936, with wage earners, relief recipients, farmers, and minorities voting solidly for Roosevelt, causing the Literary Digest’s biased sample to fail in capturing this shift.

Sampling Techniques

• Techniques of sampling can be grouped into two main categories:

• A. Probability sampling techniques (also called Random sampling)

• B. Non-probability sampling techniques (also called Non-random or Purposive sampling methods)

Probability Sampling

• The universe is divided into homogeneous groups or regions, and units representative of each group are purposely selected by the researcher’s discretion.
• Exercise of personal discretion may lead to personal bias, affecting the entire result.

Types of Probability Sampling:

1. Random sampling:

• Does not mean haphazard selection; starts with an up-to-date sampling frame.

• Every member has an equal chance of selection proportional to the population.

• Risk of picking a biased group (e.g., all young males or all smokers), making the sample unrepresentative.

• Method is cheap and quick.

• Methods include:

  • Drawing lots

  • Using a pack of cards

  • Selecting digits at random from Tables of numbers (e.g., Trippetts Table)

  • Arranging units alphabetically, numerically or geographically and selecting every nth unit (systematic sampling).

2. Stratified sampling:

• Population divided into strata based on characteristics like age, sex, class.

• Each stratum is sampled randomly to improve representativeness.

• Ensures right balance of groups (e.g., men and women, age groups, social classes).

• Also called proportional stratified sampling when groups appear in proportion.

• Example: Golthorpe and Lockwood’s study of workers in Luton factories by skill level.

3. Cluster sampling:

• Population divided into many clusters; a sample of clusters is taken.

• Used to reduce costs such as listing and travel expenses in social surveys.

4. Multi-stage sampling:

• Samples selected in multiple stages.

• Example: Random sample of census tracts, then random sample of streets within tracts, then every fourth house, then every third adult within the house.

5. Multi-phase sampling:

• Certain detailed questions asked only to a fraction of the sample.

• General information collected from the entire sample.