Sampling

Damodar Suar, Ph.D.

IIT Kharagpur
 Sampling and its importance
• Researchers conduct studies on a subset of the relevant population (or
on the sample) rather than on all elements of the population (or on the
census).
• A sample becomes representative when it mirrors the
characteristics/attributes of the population.
• Benefits of Sample: Gathering data on a subset of the population saves
time, cost, labour, and offers greater scope for a thorough and intensive
enquiry.
• Sampling is based on two premises:

– characteristics of sample elements should represent the population.

 – while some elements in the population overestimate population characteristics,
other elements underestimate. Consequently, they counterbalance each other,
and accurate characteristics of the population surface in a sample survey
• The number of elements in a sample is called the sample size.
• sample size depends on:
– (a) the size of the relevant population and
– (b) the dispersion of variables of research interest.
• The power of a sample to produce a close approximation to the population
depends on
– (a) the sample size,
– (b) the methods by which we draw the sample, and
– (c) the measurement of non-respondent bias.
• Sample Unit
• The smallest component of a sample is the sample unit– an
individual, an employee, a CEO, a family, an ad, a school, a district,
a prose passage/speech, a disaster survivor, a suicide case, etc. It
depends on the nature of the study and the investigative
phenomena. All elements in a sample is the sample size.
Adequate sample size can be rationally determined :
– (a) in advance before conducting the study by using effect-
size, power, alpha level
– (b) Rule of Thumb: Factorial design: 20 cases per cell,
correlation an path diagram 1:10, reliability:300-400, SEM
> 200.

– (c) Collecting pilot data: n = SD2 X (Z2 / E2)

 Methods to Estimate Sample Size from
Sample Mean
1. (a) Specify the amount of error (E) that can be allowed : It is the
maximum allowable difference between sample mean and
population mean. In a 5 point rating scale, it is 1/10th of the scale = .5
(b) In case of proportions, if allowable error between population and
sample proportion on a variable can be set to 5 per cent, then E =
0.05.
2. Specify the confidence level. The confidence level is set to 95 per
cent. Determine the number of standard error (Z) associated with
confidence level.
• 3. (a) Estimate the standard deviation of the population.
(b) Estimate the population proportion.
• 4. (a) On scaling items, calculate the sample size (n) using
the formula for standard error of mean (δχ), δχ = E/Z =
SD√n.
The formula can be simply n = SD2 × (Z2/E2).
(b) On proportions, calculate the sample size (n) using
the formula for standard error of the proportion.
The formula to estimate the sample size (n)
= p × q × (Z2/E2)
 Sampling Methods
• Where probability of selecting a population element is
unknown, the sample is decided arbitrarily on the basis of
judgement, convenience, defined criteria, and cost
estimation. This is non-probability sampling.
• Mixed sampling where the sampling is made in different
stages, partly according to the laws of chance and partly
according to the fixed sampling rule.
 Cond…
• (a) Simple random: each unit has equal and independent chance of
being selected, mechanical and manual randomization
• (b) Stratified random: Divide the population into strata, homogeneity
within, and draw sample from each stratum, proportional and
optimum allocation,
• (c ) Systematic: random start (i) and then the sample with
predetermined spacing (k)
• (d) Cluster: Divide population into groups where considerable
variation.
 Methods: Probability Sampling Advantages and Disadvantages
• Simple Random: Each element in the • Easy to implement with the
population has equal and independent random number digits but
chance of being included in the sample. requires the list of population.
Mechanical randomization using Expensive, encourages larger
random numbers generated either by a sample size. Less sampling error
computer programmed to scramble but more non-sampling errors.
numbers or by the table of random
numbers. Manual randomization using
lottery method.
• Beneficial in highly skewed
• Stratified. Divide the population into population. Investigator controls
different strata or groups on the basis the sample size in strata but
of region, gender, age, or income so requires the list of population.
that elements within a stratum are Stratification is difficult in a multi-
homogeneous and between strata are character study.
heterogeneous. Draw randomly a fixed
number of cases (optimum allocation)
or proportionate number of cases
(proportional allocation) from each
stratum.
• (c) Systematic. Selecting an • Simple to design and less
element of the population at the expensive. Periodicity within the
beginning with a random start (i) population may skew the sample
and every kth element following and results
the sampling fraction (sampling
fraction (k) = Total
elements/sample size) (i, i+k,
i+2k… i+ (n-1)k, where i< = k).
• Lowest cost per unit with
• (d) Cluster. Population is divided
geographic clusters but lowest
into groups such that there is
statistical efficiency due to groups
considerable variation within
being homogeneous.
each group but the groups are
essentially similar to each other.
Some groups are randomly
selected for further study.
 Non-Probalilitic
• (a) Convenience: who are available

• (b) Purpose/judgmental: Consciously chooses cases who confirm

certain criteria
• (c) Quota: Pertinent to topic studied, select sample characteristics
conform to the pre-designated population parameters
• (d) Extreme or deviant cases

• (e) Snowball: Relies on people identifying other people

 Methods: Non-Probabilistic Advantages and Disadvantages
• (a) Convenience. Investigator studies • Save time, money, and effort. Poorest
individuals who are available in field rationale and lowest credibility
or cases as they occur.
• (b) Purposive or judgemental. • Adds credibility but without external
Researcher consciously chooses cases check, difficult to ascertain the typical
with sound judgement to confirm cases.
certain criterion that are suitable for
her/ his needs
• (c) Quota. Pertinent to the topic • Researcher may go to the area where
studied, selected sample the chances are good to get a
characteristics conform to the particular type of respondent.
predesignated population Representativeness on certain
parameters. parameters may lead to un-
representativeness on others.
• (d) Extreme or deviant case. Taking • Easy but generalization is restricted.
highly unusual cases such as
outstanding success, notable failures
of class, dropout, and crises.
• (e) Snowball/chain. Relies on people • Particularly useful when there is no
identifying other people or cases to population list or one wants to draw
investigate. As they identify new sample on recommendation.
names on similar characteristics,
the snowball gets bigger.
 Non-respondent Bias
• The non-respondent bias is measured either
– (a) stating sample representativeness,
– (b) comparing those who respond and who do not, or
– (c) early versus late respondents.