August 2001 Number 2

Describing a sample
Contents unrepresentative as neither the age group nor
• Samples and populations indication are representative of the population in which
• Intersubject variation the drug will be used.
• Standard deviation
• Use of graphs to illustrate data This is why the inclusion and exclusion criteria of a trial
• Centiles and ranges should be specified to allow the reader to judge whether
a suitable sample has been selected
Samples and populations
Intersubject variation
What is a “sample”?
A sample is a group of individuals and/or observations
selected from a larger group for purposes of analysis. What does intersubject variation mean?
Opinion polls are an example. The views of a selected It is unlikely that all patients in a sample will respond
group are analysed in order to predict those of the similarly to an intervention. Some will respond better
population. than others, and some may fail to respond at all. This
is known as intersubject variation.
What is a “population”?
In statistical terms, the population is an entire group of How relevant is intersubject variation?
individuals and/or observations about which information The degree of such variation will influence the reliability
is sought. It does not necessarily refer to people; a of measurements and subsequent analysis of results.
population may be a collection of blood pressure or For example, for antihypertensive drugs, if the range of
height measurements. blood pressure responses is wide and the sample size
small, the mean blood pressure reading may be an
unreliable index of response.
In clinical trials, what do these terms mean and how
are they used?
Selected groups of patients (samples) are exposed to Although differences in individual responses to a drug
various treatments to assess their response. From the may relate to recognised patient characteristics, most
results, conclusions are drawn regarding the efficacy of differences cannot be explained and are therefore
such treatments in general use (i.e. within the attributed to random variation. This causes problems,
population from which the samples are drawn). as it is a factor that may significantly influence results.
To overcome this, we use statistical analyses to give
the reader an indication of the potential influence of
Why is choice of sample important?
random variation on results. This will be discussed
Not surprisingly, a sample representative of its
further in later issues.
population is more likely to yield results that predict the
response of that population than an unrepresentative
sample will. How is the spread of values, or variability, within a
sample described?
An obvious measure is the range, the difference
For example, if a new antihypertensive drug were to be
between the highest and lowest values. This was
used in an elderly population with moderate
discussed in the first issue of this series of bulletins.
hypertension, a trial with a sufficiently large sample of
However, the range is influenced by extreme values
elderly patients with this condition would constitute a
and will vary from sample to sample. One way to
representative sample. A trial involving a sample of
overcome this problem is to use the standard deviation.
young patients with malignant hypertension would be

Produced for healthcare professionals by North West Medicines Information Service,

The Pharmacy Practice Unit, 70 Pembroke Place, Liverpool, L69 3GF.
Editor: Frank Leach. Telephone: 0151 794 8117 E-mail: [email protected]
However, don’t be misled, there is nothing standard
about it!

Standard deviation

How do we calculate the standard deviation of a

sample?
The formula for calculating the standard deviation is:

√
2
SD = ∑(x-x)
n-1

SD = standard deviation
x = the individual values within the sample
x = the mean value
n = the number of values
∑ = sum
The histogram plots the heights (in 5cm intervals)
For example, imagine that we have measured the against the number of men who had a particular height
heights of 20 men to the nearest centimetre. The i.e. the height frequency.
respective heights are:
165, 170, 170, 172, 175, 167, 170, 167, 150, 155, 170, What does the histogram describe?
172, 187, 180, 177, 167, 167, 172, 162, 180. A histogram describes frequency distribution. This
shows the relationship between individual values in a
The mean value ( x = 170) is derived from adding all the sample and the frequency with which those values
values together (to give 3395) and dividing by the occur. Thus, figure 1 illustrates that 7 men had a height
number of values (n = 20). between 170 and 174cm while only one had a height in
the range 155 to 159cm.
If we subtract the mean value from each value in turn,
we would get both negative and positive numbers i.e. - What if the sample size is larger?
2, 0, +1 etc. and if we added these together we would As the sample size increases, the distribution of the
get 0! But if each of the numbers is squared, we get a samples will approximate more closely to that of the
positive number. parent population.
2 2 2
Hence, Σ [(165-170) , (170-170) , (170-170) , (172-
2
170) ….etc], gives Σ [25, 0, 0, 4…etc] which equals Does the shape of the frequency distribution
1341. matter?
In figure 1, the values are roughly symmetrical around
What is the variance of a sample? the mean value of 170cm. If the sample size is larger
The variance is a statistical term which describes the the histogram will become even more symmetrical. This
variability within a sample. Using the above example, type of frequency distribution within a population is
to calculate the variance, we divide 1341 by [n-1] i.e. known as Normal (or Gaussian) distribution. It follows a
19, to give 70.6 (the variance). curve which is symmetrical around the mean with a
characteristic bell-shape. This is illustrated in figure 2.
Why is n-1 used rather than n in the formula for
standard deviation?
The use of n-1 involves a rather elusive statistical
concept known as degrees of freedom. We shall return
to this in a future bulletin. By using n-1 we obtain a
closer estimate of the variability around the mean within
the population from which the sample is taken.
Therefore, as n (the sample) gets larger, the difference
between n and n-1 is reduced.

Use of graphs to illustrate data

How can the spread of values in a sample be

illustrated graphically ? What is meant by the term Normal distribution’?
One method is to construct a histogram as in Figure 1 The Normal distribution is a key concept and many
which is based on our fictitious height sample. statistical tests are based upon it. It is described by two

The information contained in this bulletin is the best available from the resources at our disposal at the time of publication. Any
opinions reflected are the author’s own and may not necessarily reflect those of the Health Authority
parameters – the mean and the standard deviation. Its
curves are always symmetrical and bell-shaped - the
extent to which the bell is flattened or compressed
depends on the standard deviation of the population.

If measurements are ‘Normally distributed’ around the

mean, you can be confident that use of the standard
deviation, as an estimate of distribution, is likely to be
valid. However, be careful. Just because the
distribution of the sample is not symmetrical, you
cannot assume the parent population does not exhibit
Normal distribution. It may be that the sample size is
too small.

What does the standard deviation tell us about a

sample from a population with a Normal
distribution? Centiles and ranges
In such a sample, or one in which there is reasonably
symmetrical distribution around the mean, one standard What does ‘centile’ mean in statistical terms?
deviation value either side of the mean will represent A centile is an alternative way to describe a frequency
th
about 68% of observations within the sample. In distribution. For example, if a child is on the 60 centile
addition, 2 SDs either side of the mean will represent for height it can be assumed 60% of children, at the
approximately 95% of measurements within the sample same age, have the same or a lower height than the
(represented by points A and C in figure 2). Going back individual child does.
to our fictitious height example, we can assume that
approximately 95% of height measurements within our What do the terms ‘interquartile range’ and ‘central
sample lie between the limits of 170 - (2 x 8.4) and 170 range’ mean?
+ (2 x 8.4) cm (i.e. approximately 153 to 187cm). The The interquartile range describes the range of values
th th
mean ± 3 SDs includes 99.7% of the observations. between the 25 and 75 centiles. It will be quite
variable from sample to sample but is useful to describe
It is important to remember that in a Normal distribution, measurements that have an asymmetrical distribution.
the mean and median values will be identical. The central range is that within which 90% of values lie.

What does a negative standard deviation value

imply?
When calculating the range of values representing 95%
of the sample you would get negative and positive
values if the mean and standard deviation values are of
a similar size. This would suggest that the data are
skewed (see below). In this case, use of standard
deviation to calculate sample range would be invalid.

Do all frequency distributions exhibit Normal

distribution?
No, there are a number of different frequency
distributions (e.g. Poisson and Binomial). Skewed
distributions are those in which the distribution of values
is weighted towards one extreme. They can be
described as negative or positive depending on the
shape of the curve (Figure 3). Positive is more
common than negative skewing. Different statistical
tests are used for analysis of skewed distribution,
compared to those used for Normally distributed data.
These will be covered in future issues of Statistics in
Small Doses.

The information contained in this bulletin is the best available from the resources at our disposal at the time of publication. Any
opinions reflected are the author’s own and may not necessarily reflect those of the Health Authority