Sampling

Front 1

Sampling

Back 1

Sampling: means to select subset of units
from a population to collect information to
draw inferences about whole population
– Non-probability sampling
– Probability sampling

Front 2

Non-Probability Sampling

Back 2

• Subjective method to select units from population

• Fast, easy, inexpensive

• Must assume sample represents population in order to make inferences, which is difficult due to biases

Front 3

Non-Probability Sampling

Back 3

• Inclusion probability can’t be calculated due to selection bias so can’t produce reliable estimates

• Used often to generate ideas, as a preliminary
step or as a follow-up step

Front 4

Non-Probability Sampling

6 Types

Back 4

1. Haphazard: “man in the street”

2. Judgment: purposeful selection

3. Volunteer: screened volunteers

4. Quota: sampling up to specific number

Front 5

Non-Probability Sampling

6 Types

Back 5

5. Modified probability: probability sampling
followed by quota sampling

6. Network/snowball: use contacts to find rare
characteristics

Front 6

Probability Sampling

Back 6

• Selection of units based on randomization or chance

• Complex, time-consuming, and costly

• Inclusion probability can be calculated

Front 7

Probability Sampling

Back 7

• Prevalence, incidence, and sampling error can be calculated

• Inferences can be made about population since they are randomly selected and have a non-zero inclusion probability

Front 8

Probability Sampling

9 Examples

Back 8

– Bernoulli

– Simple random (SRS)

– Systematic (SYS)

– Cluster

– Probability-proportion-to-size (PPS)

Front 9

Probability Sampling

9 Examples

Back 9

– The random method of PPS

– The systematic method

– Stratified

– Multi-stage

Front 10

Probability Sampling

Bernolli

Back 10

• Easiest method, like tossing of a coin

• Good for moderate- to large-size samples from
large electronic files

• Quick, cheap, effective but least precise

Front 11

Probability Sampling

Simple Random Sampling (SRS)

Back 11

• One-step selection method ensures every
possible sample of size n has an equal chance

• Similar to drawing names from a hat, each
individual has equal inclusion probability

• Can be done with or without replacement

Front 12

Probability Sampling

Systematic Sampling (SYS)

Back 12

• Individuals are selected from population at
regular intervals using a sampling interval and
random start

Front 13

Probability Sampling

Cluster Sampling

Back 13

• Randomly selecting complete groups

• Less statistically significant than SRS but
cheaper and can produce estimates for
clusters themselves

Front 14

Probability Sampling

Cluster Sampling

Back 14

• Two-step process:
1. Population grouped into clusters
2. Select sample of clusters and interview all

individuals within

Front 15

Probability Sampling

Cluster Sampling

Back 15

• Statistical efficiency depends on how
homogenous units within clusters are, how
many population units are in each cluster, and
how many clusters are sampled

Front 16

Probability Sampling

Probability-Proportional-to-Size (PPS)

Back 16

• Uses auxiliary data and yields unequal
probabilities of inclusion

• Increased precision if size measures are
accurate and variables of interest are
correlated with size of unit

• Improve statistical efficiency/reduce variance

Front 17

Probability Sampling

Probability-Proportional-to-Size (PPS)

Back 17

• Needs to have good quality, up-to-date

sampling frame for use as an MOS or there
may be no statistical gain in efficiency

• Estimation of sampling variance is more
complex than with equal probability models

Front 18

Probability Sampling

Random Method of PPS

Back 18

• Drawing names from hat, but units have as
many pieces of paper in the hat as size
indicates

Front 19

Probability Sampling

The Systematic Method

Back 19

• The MOS are cumulated and running totals
are recorded

• When MOS is less reliable this is not the best
strategy

Front 20

Probability Sampling

Stratified Sampling

Back 20

• Not a selection method, but a way to organize
the population in homogenous, mutually
exclusive groups called strata

• Independent samples are selected from each
stratum and any sample designs can then be
used to sample within strata

Front 21

Probability Sampling

Stratified Sampling

Back 21

• Population can be stratified by any variables
available for all units on the frame before the
survey is conducted

• Good for skewed populations and to ensure
adequate sample sizes for domains of interest

Front 22

Probability Sampling

Multi-Stage Sampling

Back 22

• Select a sample in two or more stages where
the units at each stage are different in
structure and hierarchy
– Primary sampling units
– Second stage units

Front 23

Probability Sampling

Multi-Stage Sampling

Back 23

• Can have any number of stages but complexity of the design increases with the number of stages

• Can be more efficient than one-stage when
clusters are homogenous with respect to
variables of interest

Front 24

Probability Sampling

Multi-Stage Sampling

Back 24

• Can decrease travel time and cost of interviews as
samples are less dispersed