TWO-STAGE CLUSTER SAMPLING (WITH QUOTA SAMPLING AT SECOND STAGE)

1
SAMPLING METHODS
2
TWO-STAGE CLUSTER SAMPLING (WITH QUOTA SAMPLING
AT SECOND STAGE)
3
STATISTICAL TABLES Table A Random
Digits
4
SIMPLE RANDOM SAMPLING
5
STRATIFIED RANDOM SAMPLINGGrouped by
characteristic
6
SYSTEMATIC SAMPLING
7
CLUSTER SAMPLING
8
TWO STAGE CLUSTER SAMPLING (WITH RANDOM SAMPLING
AT SECOND STAGE)
9
FLOWCHART
10
TABLE 1
11
TABLE 2
12
POPULATION

• Population units
• e.g. children or adults
• Population observations, characteristics or
  attributes
• e.g. immunization history
• Time and resources are limited so that only
  sample units and sample observations can be
  selected from the population.

13
Total Count versus sampling

• National census is conducted every 10-15 years
• Less accurate over time.
• Less accurate in dynamic (shifting) populations.
• Very expensive.

14
Sample surveys allows obtaining more extensive
information (smaller number of persons)

• Need to train a limited number of interviewer
• More in-depth questions or detailed data
• Can quickly provide useful information
• Relatively low cost

15
"Less is more" Mies Van der Rohe

• A sample should be representative to the
  population of interest.

16
Simple Random Sampling

• Need a list of all eligible persons in the
  population
• Every person has equal chance (equal probability)
  to be selected in the sample
• Basic method, important for comparison with other
  sampling methods
• Provides an unbiased estimate of a variable in a
  population

17
Simple Random Sampling (continued)

• Permits quantitative assessment of sampling error
• Rarely used in actual surveys
• Difficult
• Expensive
• Excessive travel time (different location
• of subjects)
• Excessive local introduction and organization
  time

18
Sampling with replacement

• Individuals from a population of observations may
  appear more than once in a sample of population

19
Sampling without replacement

• Individuals from a population of observations can
  appear only once in a sample of population.
• This is the usual case.
• Number of possible samples N!/n!(N-n)!
  (if order is not important)
• Equal probability selection Method (EPSEM)
• Use of random tables, or computers

20
Systematic Sampling

• Similar Procedure
• List all persons in the population
• Define selection interval
• (Sampled population)/(Sample size)
• N/n
• An integer for ease of field use

21
Systematic Sampling(continued)

• Select a random starting point (first person in
  the sample)
• Next selection the random start
• the random interval
  
• And so on and so forth
• Data should not be ordered in a special way.

22
Stratified random sample

• The population is divided into multiple strata
  based on common characteristics
• e.g.
• Residence (Urban or rural)
• Tribe, ethnicity or race
• Family income (poor, moderate, or wealthy)

23
Stratified random sample(continued)

• A random sample is selected from each stratum
• The samples from each stratum are combined for a
  single estimate of the population mean and
  variance.

24
One-Stage Cluster Sampling

• The population is listed as groups (termed
  clusters), not individuals
• e.g.
• Area of residence (village, town, .. etc.)
• School or classroom within a school
• All clusters are listed and a sample of clusters
  is selected.
• All persons in the selected clusters are
  examined.
• The samples from each of the clusters are
  combined into a single estimate of the population
  mean and variance.

25
Two-Stage Cluster Sampling with Simple Random
Sampling at the Second Stage

• Stage I A random sample of clusters
• Stage II A sample from selected clusters
• The samples from each of the selected clusters
  are combined into a single estimate of the
  population mean and variance.

26
Two-stage Cluster Sampling with Quota Sampling in
the Second Stage

• The population is divided into multiple clusters.
• Stage I A random sample of clusters
• Stage II A random start
• Interviewer continues
  within a cluster until the quota (constant
  number) is filled.
• The samples from each cluster are combined into a
  single estimate of the population mean and
  variance.

27
Two-stage Proportionate to size (PPS) Cluster
Sample with Quota Sampling in the Second Stage

• The population is divided into multiple clusters.
• Stage I A random sample of clusters with
  probability proportionate to their size (PPS)
• "Size" means the number of eligible persons
  residing in the cluster.
• Stage II A random start
• Interviewer continues within a
  cluster until the quota (constant
  number) is filled.

28
Two-stage Proportionate to size (PPS) Cluster
Sample with Quota Sampling in the Second Stage
(continued)

• The samples from each cluster are combined into a
  single estimate of the population mean and
  variance.
• This method is favored by Expanded program on
  Immunization (EPI).
• Note No random selection in the second stage.

29
Probability sample versus Non-probability
sample

• Every person has equal chance (equal probability)
  to be selected in the sample.
• No bias
• Generalization of the results
•  
• On average, the characteristics of people in
  probability samples are similar to those of the
  population from which they were selected,
  particularly if a larger number are chosen.

30
Probability sample versus Non-probability
sample

• Sampling in clinical trials are usually highly
  selected and biased samples of all patients with
  the condition of interest. (Internal validity)
• 1 Use of inclusion/ exclusion criteria
• Restricts the heterogeneity of patients
• Excludes atypical forms of the disease
• Improves chances of patients completing the
  assigned treatment used in the study
• Excludes presence of other diseases
• Excludes an unusually poor prognosis
• Excludes patients with contra-indication for the
  treatment

31
Probability sample versus Non-probability sample
(continued)

• 2 Refusal of patients to participate in the
  study Tend to be systematically different from
  those who agree to enter in the trial
• Socio-economic class
• Severity of disease
• 3 Patients who are thought to be
  unreliable (would not follow the groundrules of
  the trial are usually not enrolled.

32
PRECISION

• Determine the desired level of precision i.e.
  amount of error in parameter estimates that can
  be tolerated by the decision-maker.
• Definitions
• Precision is the size of deviations from the
  average value of some parameters of interest
  obtained by repeated application of sampling
  procedures.
•  
• Accuracy is the size of deviations from the true
  mean of some parameter in a population.
• In surveys, we cannot measure accuracy but can
  measure precision

33
MATCHING

• Is stratified sampling in which numbers selected
  in each stratum are determined by the numbers in
  that stratum in some other sample.
• Main stay in epidemiology.
• 11 is the best.
• Can have up to 51.