Evaluation Capacity Development (ECD) project – Uganda

“How To”: Choose Sampling Techniques for Evaluations

By Tom Mwebaze
School of Economics, Makerere University

1. Define Sampling
Sampling is the process of selecting units (e.g. people, organizations, time periods) from a population of
interest, studying these in greater detail, and then drawing conclusions about the larger population. The
evaluator collects data from a subset of individuals – a sample – and uses those observations to make
inferences about the entire population. The characteristics of the sample should reflect the
characteristics of the population targeted by the intervention. In that case, the evaluator’s conclusions
from the sample are probably applicable to the entire population.

2. Techniques of Sampling during Evaluation Process

Sampling techniques can be classified into two major categories: Probability (Statistical
Sampling/Proportional Sampling) and Non-probability (Non-statistical Sampling).

a) Probability/ Statistical Sampling/ Proportional Sampling

In probability sampling, the population to be sampled is defined by the number of items/units. It is usually
used to reveal the presence (or absence) of a defined characteristic in a portfolio of items/units with
similar characteristics. Probability sampling options use random or quasi-random options to select the
sample, and then use statistical generalization to draw inferences about that population. To minimize
bias, these options have specific rules on selection of the sampling frame, size of the sample, and
managing variation within the sample. Consider why you want to study your population of interest and
what you want to do with the information that you have gathered, before you choose your option.

The options include:

 Simple random: drawing a sample from the population completely at random.
th th
 Systematic: selecting every n case from a list (e.g. every 10 client).
 Cluster sampling in which larger clusters are further subdivided into smaller, more targeted
groupings for the purposes of surveying.
 Stratified: splitting the population into strata (sections or segments) in order to ensure distinct
categories are adequately represented before selecting a random sample from each.

Each of the options starts with a sampling frame, which can be thought of as a list of all elements in the
population of interest (e.g. names of individuals, telephone numbers, house addresses, and census
tracts). The sampling frame operationally defines the target population from which the sample is drawn
and to which the sample data will be generalized.

Simple Random
This technique gives all units in the population an equal opportunity of being selected by using a method
that will select units completely at random. To use this technique you will need a complete list of
everyone in the population – a sampling frame. Sampling is done in a single stage with each element
selected independently. You can randomly select by using a random number generator or by using a
random number table, or by pulling out of a hat. It is very effective if you have a small or moderate sized
population.

Systematic sampling
Samples are drawn by starting at a randomly selected element in the sampling frame and then taking
every nth element. Systematic sampling is easier to perform and hence is less subject to interviewer
errors than simple random sampling.

Stratified Random sampling

This technique divides the population into meaningful homogenous or similar groups based on a certain
characteristic (e.g. gender, race, socioeconomic status) called strata, and then selects a random sample
from each group. Stratified sampling ensures greater representativeness on a characteristic of interest

1
 Evaluation Capacity Development (ECD) project – Uganda

within the population. This method allows you to study a wider range of the population without a larger
sample size.

Cluster Sampling
The sample is selected in stages, first selecting groups of elements, or clusters (e.g. city blocks, census
tracts, schools), and then selecting individual elements from each cluster (e.g. randomly or by systematic
sampling). In case there is no list of every member of the population, it is still possible to choose a
random sample by using cluster sampling. Cluster sampling divides your population into groups and a
simple random selection of those groups is made. You then survey everybody within the selected
groups.

Method Best when

Simple random sampling Whole population is available
Stratified sampling (random There are specific sub-groups to investigate (e.g.
within target groups) demographic groupings)
Systematic sampling (every nth When a stream of representative people are available (e.g.
person) in the street)
Cluster sampling (all in limited When population groups are separated and access to all is
groups) difficult, e.g. in many distant cities

b) Non-probability/ Non-statistical Sampling

The difference between non-probability and probability sampling is that non-probability sampling does
not involve random selection and probability sampling does. With non-probability samples, we may or
may not represent the population well, and it will often be hard for us to know how well we've done so. In
general, researchers prefer probabilistic or random sampling methods over non-probabilistic ones, and
consider them to be more accurate and rigorous. Non-probability sampling can be classified into
purposive, snowball, convenience sampling or quota sampling.

Purposive Sampling (or Purposeful)

This is sampling with a purpose in mind. We usually would have one or more specific predefined groups
we are seeking. Purposive sampling can be very useful for situations where you need to reach a
targeted sample quickly and where sampling for proportionality is not the primary concern. The variables
to which the sample is drawn up are linked to the research question. Purposive sampling options study
information-rich cases from a given population to make analytical inferences about the population. Units
are selected based on one or more predetermined characteristics and the sample size can be as small
as one (n=1). To minimize bias, this cluster of options encourages transparency in case selection,
triangulation, and seeking out of disconfirming evidence. Only people with specified characteristics are
selected. These characteristics can represent the range of characteristics in your population of interest,
or you can focus on some characteristics in particular.

For example, if you are evaluating the attitudes of drivers towards speeding, you may want to only
sample those who have got penalty points. Or, you could sample those with extreme characteristics by
only selecting drivers who have been disqualified as a result of multiple offences.

Snowball sampling
A type of purpose sampling where existing participants recruit future subjects from among their
acquaintances. Thus the sample group appears to grow like a rolling snowball. Using this approach, a
few potential respondents are contacted and asked whether they know of anybody with the
characteristics that you are looking for in your evaluation. For example, if you wanted to interview a
sample of vegetarians/cyclists/people with a particular disability/people that support a particular political
party or the homeless, your initial contacts may well have knowledge of other. In these cases you are not
likely to be able to find good lists of people of these characteristics within a specific geographical area.
However, if you go to that area and identify one or two, you may find that they know very well who the
others in their vicinity are and how you can find them. This makes it easier to contact people you might

2
 Evaluation Capacity Development (ECD) project – Uganda

not otherwise have any way of getting in touch with. However, it does introduce a lot of bias into your
results because your samples are all likely to know each other and to have similar opinions. With this
method you survey whoever you happen to have access to. It is also called 'opportunity sampling'. This
is commonly used in market research.

Quota
This is another method commonly used in market research. With quota sampling, you divide the
population (e.g. older drivers) into distinct parts (strata). You then decide how many of each stratum you
want to have in the total sample. For example, you could stand outside an out-of-town supermarket in
the day time. You have decided that you want to survey 50 male drivers aged over 65. Every time you
see a male older driver enter the store you ask if he is aged over 65 and you continue until you reach
your quota of 50 completed surveys. This is the non-random form of stratified sampling.

Method Best when

Quota sampling (get only as many as you
You have access to a wide population, including sub-groups
need)
Proportionate quota sampling (in You know the population distribution across groups, and when
proportion to population sub-groups) normal sampling may not give enough in minority groups
Non-proportionate quota sampling There is likely to be a wide variation in the studied
(minimum number from each sub-group) characteristic within minority groups
Purposive sampling (based on intent) You are studying particular groups
Expert sampling (seeking 'experts') You want expert opinion
Snowball sampling (ask for
You seek similar subjects (eg. young drinkers)
recommendations)
Modal instances sampling (focus on When sought 'typical' opinion may get lost in a wider study,
'typical' people) and when you are able to identify the 'typical' group
Diversity sampling (deliberately seeking You are specifically seeking differences, e.g. to identify sub-
variation) groups or potential conflicts

3. Sampling contributes to a quality evaluation

Sample size
Decisions on sample size depend on whether your data is quantitative or qualitative. For quantitative
data, you use statistical calculations to determine how large your sample should be.

Response rates
Once a sample is selected, an attempt is made to collect data (e.g. through interviews or questionnaires)
from all of its members. In practice, researchers never obtain responses from 100% of the sample. Some
sample members inevitably are travelling, hospitalized, incarcerated, away at school, or in the military.
Others cannot be contacted because of their work schedule, community involvement, or social life.
Others simply refuse to participate in the study, even after the best efforts of the researcher to persuade
them.

Unless every person in a sample is interviewed, the sample will not be a true random sample. Non
response must therefore be kept to negligible proportions if the results of the survey are to be valid. If a
respondent is not interviewed when first called upon, it will be necessary to follow-up with a second call.
The simplest way of dealing with non-response is to treat the non-respondents as being similar to the
respondents. This treats the people who do not respond at all as if they are similar to the initial non-
respondents who subsequently respond, but takes into account differences between those who respond
initially and those who do not.

3
 Evaluation Capacity Development (ECD) project – Uganda

It is useless to substitute the person next door. Suppose, for example, an enquiry is made as to the size
of family. Then people without children are more likely to be out than those who have children. If the
people without children are not interviewed because they are not at home, and the interviewer turns to
other people who are at home, the results will obviously be biased and inaccurate.

Sampling Errors
It has already been mentioned that even when a correct process is employed in choosing a sample, the
sample cannot be exactly representative of the aggregate. Thus errors will arise. These inevitable errors
arising from the fact that the sample will vary from the aggregate are called random sampling errors. The
amount of the error will depend upon:

(a) the size of the sample;

(b) the extent of the variability of the material;
(c) the sampling procedure employed, and;
(d) the way in which the results are calculated.

However, given a proper method of selecting the sample, the probability of errors of any given size can
be calculated from the detailed results of the actual sample. By calculating the errors of different
sampling methods which can be used on the same material, it is possible to plan further surveys more
efficiently. The use of appropriate sampling methods and an adequate response rate are necessary for a
representative sample, but not sufficient. In addition, the sample size must be evaluated. The sampling
error is a number that describes the precision of an estimate from any one of those samples. It is usually
expressed as a margin of error associated with a statistical level of confidence.

4. Conclusion
In conclusion, the following items need to be considered when selecting the sampling technique:
 A combination of different sampling methodologies at different stages in evaluation.
 The representation of different stakeholders.
 Correction of inherent biases.

References
- Cochran W. G. (1963). Sampling Techniques. (2nd ed.). New York: John Wiley and Sons, Inc.
- Fowler, F. J. (1988). Survey Research Methods. (rev. ed.). Applied Social Science Research
Methods Series, Volume 1. Newbury Park, CA: Sage Publications.
- Fruend, J.E. and Walpole, R.E., Mathematical Statistics, Prentice Hall, Inc.
- Mendenhall W. et al, Mathematical Statistics with Applications. PWS-Kent Publishing Company.
- Morris H. (1991). Statistical Analysis for Decision Making, (5th Ed).
http://changingminds.org/explanations/research/sampling/choosing_sampling.htm