Populations of interest

A topic in research methodology

In research we are often interested in 'populations' that we can define in various ways. One early task in any research study is to identify the particular population of interest that the research is meant to inform us about.

In the natural sciences the objects of research are often 'natural kinds' where sometimes a single specimen can stand for the kind in general (a pure sample of water at a temperature of 293K and one 10⁵Pa pressure should be much like any other such sample under the same conditions).

In education and the social sciences the populations are 'social kinds' ('students', 'teachers', 'classes', schools', etc.), where membership is determined by social convention and different members of the same kind often vary considerably from each other. Consider the following examples of populations:

• primary schools in Wales
• chemistry teachers in England
• gymnasium physics classes in Germany
• 14-16 year olds who attend courses in field study centres
• teams competing in any International Physics Olympiad
• visits to science centres by family groups

Such populations are often large, and there may be be limited information that allows researchers to identify the members of the population. (Consider how easily a researcher could identify all the secondary school science teachers in the world – or even in, say, Canada.)

Populations in theory-directed research

In theory-directed research, studies are often intended to be generalisable to wide populations. Ideally, we might want out research to tell us about

• science teachers
• students studying biology at university
• school chemistry examination papers
• primary age children
• university geology lectures
• …

but realistically we may have to limit the population that can be sampled (at least within a single national context, for example).

Read about sampling

Read about some examples of populations sampled in research studies

If we want to carry out observations to find out about student interactions in group-work in secondary school science classes (everywhere), but know that resource limitations mean we can only visit schools within easy cycling distance, then if we have good reason to suspect that student interactions will be affected by cultural factors (norms and expectations that may vary between different education systems) then we need to define our population accordingly, to avoid making claims about generalisability that cannot be justified.

Read about generalisation from research

Populations in context-directed research

As opposed to theory-directed research, which is intended to be generalisable, context-dependent research has its primary focus on a specific research context – perhaps a particular school, a particular classroom, a particular lecture course module.

If the focus of research is the teaching staff of a primary school, or the learners in a teacher's own tutor group, or a 12 lectures that make up a unit of a degree course, then in principle it may be possible to collect data from (e.g., interview every teacher in the primary school, observe each of the lectures in a unit of work) each member of the population.

If the population is every students in a large secondary school, it may be feasible to survey the entire population with a questionnaire. However, even if this is feasible, research ethics mean that we rely on the gift of data from others, and they may not consent to us talking to them, or observing them teach, or to filling in an on-line form. So, even when our theoretical sample is the entire population, in practice we may only collect data from a part of it.

Read about research ethics

Representativeness of samples

It may seem that in research the larger the sample, the better.

'All other things being equal', that is reasonable. However, we need to be aware that in research all things are seldom equal! For example, we may sometimes (deepening on the kind of data we need to meet our research puporses) get more useful data by interviewing a sample of a population in depth than having questionnaires completed by everyone in the sample. Of course, even better might be a survey of everyone by questionnaire supplemented by interviews with a sample.

Read about the use of multiple techniques in research

However, we should be aware that while the size of sample may be important, its representativeness may be more important. Consider two hypothetical surveys:

A survey of a large population where a 10% sample was selected randomly, and all responded.

A survey of all the teachers in a large school, where 60% chose to respond.

At first sight, a 60% sample seems much better than a 10% sample. However, whereas the random sample is likely to be representative of the wider population (and there are statistical methods to estimate likely errors introduced by only getting responses from 10% of the population), it may be that the sample of teachers who choose to respond to a survey may be systematically different from those 40% who chose not to respond. Given that concern ,we can be more confident of generalisation from the 10% random sample than the 60% self-selecting sample.

This is not an entirely realistic comparison, as it assumes that everyone in the randomly selected sample replies (and in practice we might only get a third of those selected responding – if we are lucky!)

However, there is a serious issue here, that a sample is useful to the extent we can be confident it is likely to be representative of the population we are sampling. In practice we often know that self-selecting samples are not fully representative in this way.

Read about sampling bias