therefore give information to calculate an optimum sample size; second, fi eldwork may yield early results that can be used and the study can end early; conversely, there may be a need to seek more views and therefore increase the number of respondents.
Qualitative versus quantitative research
In qualitative research, sample size is far more subjective than in quantitative approaches; it is also complex. In theory, qualitative sample sizes should not be fi xed fi rmly at the start of the project. The overriding idea is that new cases should be selected until the data bring nothing new. In practice, a methodological compromise is made and most proposals set a certain
Part 2 Data collection
176
number of groups and depths (see Chapter 7). This then allows budgets to be controlled and
timetables to be developed. As a guide to typical sizes, it is useful to refer to recent studies;
Table 5.11 shows the sample sizes used for a study of newspaper delivery.
In quantitative research, the main ways of deciding on sample size are: by calculation, by
using ‘accepted’ industry standards, by budget (time or money available), and by ‘building’
analysis cells.
The optimum sample size can be determined by a series of calculations. The ‘calculation
method’ implies that the sample will be selected by probability means. It takes account of the
population size and the expected accuracy of results. In theory, this is the best way to arrive at
a sample size; in practice, it is only used with government or cross-industry studies. For very
large populations, the size of the sample is entirely independent of the size of the population.
We can discover what the UK population thinks by posing questions to a sample of just 1000;
we can know what the US population thinks also by taking a sample of just 1000. Remember
there are 60 million people in the UK and over 290 million in the USA. For newcomers,
this can be a surprise: the common misconception is that the sample should be taken in
proportion to the population. The important thing is how homogeneous the population is,
rather than its full size. To make this point, let us say that if one million people were exactly
the same as each other, we would only need to take a sample of one person. Similarly, if there
were 200 very diff erent people, we would need to interview all 200.
There are two main formulae: one is used for studies that involve the estimation of the
average (mean) value in a population; the second is used for studies that involve proportions.
The formulae change, depending on the type of probability method used. This book does not
give details of the formulae used in sample size estimation. The reader is advised to consult
statistical texts and to search the Internet, where ‘sample size estimation software’ can be found
relatively easily: for example, http://www.surveysystem.com/sscalc.htm (Creative Research
Systems 2009) and Lenth (2009).
The use of ‘accepted’ industry standards is more common than we might imagine. At some
point, people working in a specifi c fi eld have drawn conclusions from studies using a certain
sample size. When repeat studies seem to be stable, then there is no reason to increase sample
sizes. This method is also used ‘as a proxy’ for non-probability methods (where calculations
cannot take place); it is as if a ‘past’ probability sample composition is replicated for a non-
probability sample. Rather than performing calculations, the researcher seeks comparable
studies and examines the methodology. As a guide to such standards, look at the following
sizes used in some recent UK studies (Table 5.12). Remember that the detailed composition of
each sample is not reproduced.
Many sample sizes for research studies are decided by researchers on what is feasible within
time or money available: these are ‘budget’ limitations (see Table 5.13). The fi nal method of
Qualitative sample sizes
Qualitative studies Size (people)
Five groups with readers Eight at each
Depths with newsagents Ten
Table
5.11
Chapter 5 Sampling
177
‘building’ analysis cells is associated with non-probability methods. Consider the fi nal results, and then think backwards. If you want precise results, you need a large sample; if you are happy with ‘indicative’ data, then you can get away with a smaller sample. You are likely to provide table breakdowns on some standard demographics: male and female, for example.
It is reasonable to think that percentages will be applied. A percentage is based on 100, so
a minimum of 100 people might be chosen. The sample size can be built up in this way.
We may assume that we want to see cross-analyses of the standard demographics (sex, age, region, social grade) and perhaps of other aspects important to the subject area (high, low,
or medium consumption; high or low awareness). If we accept that an analysis cell should have a minimum of 100 people, then the sample will be at least 600 (there are six social grades) and even higher if the other cells are not satisfi ed by this selection. In part, this relies
on forecasting likely incidence; in part, it is based on known characteristics of the marketplace. Some practitioners say there should be a minimum of 50 in a cell; others say 100. In fact, these choices are based on their own experiences and knowledge of the marketplace in question.
Quantitative sample sizes
Surveys Size (people)
UK National Readership Research Survey 36,000 p.a.
Omnibus Survey 1000
FMCG usage study 500
Product test in-home 200
Table
5.12
Summary of sample size determination approaches
Method Comment
Calculation The ‘ideal’. For quantitative studies, based on
probability, common in large studies
‘Accepted’ industry standards Past experience can identify ‘safe’ sizes, unknowingly Money available Common reason for using quotas
Time available A very poor reason, and a good reason to cancel
research
Building analysis cells Commonly used, particularly in non-consumer
studies Researcher’s judgement The best method for qualitative research
Combination of above Commonly used for both quantitative and qualitative
commercial research
Table
5.13
Part 2 Data collection
178