Determining the Size of a Sample

Kajal Srivastava
SPM Deptt.
S.N.Medical College,
Contents
 Sample definition
 Characteristics of an ideal sample
 Terminologies
 C.I. method of calculating sample size
 Formulas
 Other methods
What is sample
 A finite set of objects drawn from the
population with an aim is called a sample.

 Probability sampling

 Aim
Why to determine sample size

 Determining sample size is a very important

issue because samples that are too large may
waste time, resources and money, while
samples that are too small may lead to
inaccurate results.
Characters of Ideal Sample

 True representative

 Precision

 Unbiased character
Before calculating sample size one
has to decide on the following:
 Study design

 Types of outcome measure

 Guess at likely result

 Required level of significance

 Required precision / power

 Types of outcome measures- Proportion, rates
and means.

 Sampling error-An estimate of an outcome

measure calculated in an intervention study is
subject to sampling error, because it is based
on a sample of individuals and not on the whole
population of interest.
Confidence Interval

 The confidence interval is a range of plausible

values for the true value of the outcome
measure.
 It is conventional to quote the 95 % confidence

interval (also called 95%confidence limits).

 There is a 95 chances out of 100 to find that

the true value will be with in this range.

 95% of samples drawn from a population will

fall within + 1.96 x Sample error

Significance tests & P value
 It is appropriate to test a specific hypothesis
about the outcome measure.

 Null hypothesis

 Significance tests

 Determine the p – value (probability value) or

‘significance’ of the results.
Types of error
 Type I error(α)- Rejecting a null hypothesis when its
true.
 The probability of this error is referred as P value.
 When P value is small, it is safe to conclude that

groups are different. This threshold, 0.05 is the level

of significance.

 Type II error- The second type of error is failure to

reject null hypothesis when it is actually false. The
complimentary probability of type II error is the
statistical power (1-β). Thus the power of a statistical
test is a probability of correctly rejecting a null
hypothesis when it is false.
Power of study
 Power of the study indicates the probability of
finding a statistically significant difference between
the two groups.
 The power of a study depends on:

1. The value of the true difference between the study

groups (effect size). The greater the effect, the
higher the power to detect the effect as statistically
significant for a study of a given size.
2. The study size; The larger the study size, higher is
the power.
3. The probability level at which a difference will be
regarded as ‘statistically significant’.
One sided and Two sided tests
 Two tailed- when we make hypothesis that
sample statistic is lesser or greater than
population parameter than its called two
tailed.(i.e. in both direction)
 Single tailed- when we make hypothesis that

sample statistic is either lesser or greater

than population parameter than its called one
tailed.(in single direction)
 Usually, it is more important to estimate the
effect of the intervention and to specify a
confidence interval around the estimate to
indicate the likely range, than to test a
specific hypothesis.
 Therefore, in many situations it may be more

appropriate to choose the sample size by

setting the width of the confidence interval,
rather than to rely on power calculations.
 Confidence interval approach: applies the
concepts of accuracy, variability, and
confidence interval to create a “correct”
sample size
 Variability: refers to how similar or dissimilar
responses are to a given question
 P (%): share that “have” or “are” or “will do”
etc.
 Q (%): 100%-P%, share of “have nots” or “are
nots” or “won’t dos” etc.
With Nominal data (i.e. Yes, No), we can
conceptualize answer variability with bar charts…
the highest variability is 50/50
If we conducted our study over and over, e.g.1,000 times,
we would expect our result to fall within a known range (+
1.96 s.d.’s of the mean). Based upon this, there are 95
chances in 100 that the true value of the universe statistic
(proportion, share, mean) falls within this range!
The Confidence Interval Method of
Determining Sample Size
Normal Distribution

1.96 X s.d. defines the endpoints for 95% of the distribution

There is a relationship among:

 The level of confidence we desire that our

results be repeated within some known range
if we were to conduct the study again, and…
 the variability (in responses) in the population

and…
 the amount of acceptable sample error
(desired accuracy) we wish to have and…
 the size of the sample.
Allowances of losses
 The reduced power or precision resulting from
losses may be avoided by increasing the initial
sample size in order to compensate for the
expected number of losses.

 A 5-20% allowance is generally considered

appropriate.

 Practical constraints
Consequences of studies those are
too small
 If the sample is small the confidence interval will
be very wide and even though it will probably
include the null value, it will extend to include
large values of the effect measure.
 In other words, the study will have failed to
establish that the intervention has no
appreciable effect.
 In case the intervention does have an
appreciable effect, a study that is too small will
have low power i.e. it will have little chance of
giving a statistically significant difference.
FORMULAE FOR SAMPLE SIZE
ESTIMATION
 Estimating a population proportion: With
specified absolute precision-
 Required information-

- Anticipated population proportion: P, a rough

estimate of P is sufficient.
- Desired confidence level
- Absolute Precision: (d ) - total percentage
points of the error that can be tolerated on
each side of the figure obtained.
 The estimated sample size is applicable only in
case of SRS.
 If another sampling method is used, a larger

sample size is likely to be needed because of

design effect.
 For cluster sampling strategy, the estimated

sample size as above is multiplied by design

effect, which is defined as the ratio of variance
obtained in cluster survey to the variance for
the same sample size adopting SRS.
 In cluster sampling strategy, a design effect
of 2 is taken.

 This means twice as many individuals would

have to be studied to obtain the same
precision as with SRS.
Estimating a population proportion:
With specified relative precision

Relative Precision:– The sample result should

fall within є % of the true value.
Estimating the difference between two
population proportions with specified absolute
precision (Two – sample situations)

P1, P2 = anticipated value of the

proportions in the two populations.
Hypothesis testing for two population
proportions
 Required information :-
- Anticipated values of the population
proportions: P1 & P2
- Level of significance
- Power of the test: 100 (1-β) %
Estimating a population mean: With
specified absolute precision
Estimating population mean: With
specified relative precision
Estimating difference between means of
two populations with specified precision
Hypothesis testing for two population means
 There are computer programs available that
perform sample size calculations.

 In particular, this facility is available in the

package ‘Epi Info’, though it does not cover
the full range of possibilities.
 Other Methods of Sample Size
Determination
• Arbitrary “percentage rule of thumb” sample
size:
• Arbitrary sample size approaches rely on
erroneous rules of thumb (e.g. “n must be at
least 5% of the population”).
• Arbitrary sample sizes are simple and easy to
apply, but they are neither efficient nor
economical. (e.g. Using the “5 percent rule,”
if the universe is 12 million, n = 600,000 – a
very large and costly result)
 Other Methods of Sample Size
Determination…cont.

• Conventional sample size specification

• Conventional approach follows some
“convention” or number believed somehow to
be the right sample size (e.g. 1,000 – 1,200
used for national opinion polls w/+ 3% error)
• Using conventional sample size can result in a
sample that may be too large or too small.
• Conventional sample sizes ignore the special
circumstances of the survey at hand.
 Special Sample Size Determination
Situations
Sample Size Using Nonprobability Sampling

• When using nonprobability sampling, sample

size is unrelated to accuracy, so cost-benefit
considerations must be used
Refrences
 Sample size determination in health studies- A
practical manual- S.K. Lwanga and S. Lameshow
 Sample size determination in health studies-NTI
Bulletin 2006,42/3&4,55-62, VK Chadha
 Sampling guide- Tulane University School of Public
Health, Robert Magnani
 On validity of assumptions while determing
sample size. IJCM, S.B. Sarmukaddam, S.G. Garad-.
 Essentials of biostatistics- Nishi Agarwal
 Methods in biostatistics- BK Mahajan
Thank You