It is seldom the case that we examine the whole population which we have chosen. Instead, we use sampling – that basically means that we take a smaller sample of the population: a study sample. A study sample is often denoted by “n”.

The reasons behind sampling are primarily that it is very costly and time consuming to collect data for the entire population.

However, sometimes you can include the whole population – like if you have small populations, such as one school or one hospital or one company (this is often referred to as a case study). Another example is when you use national registers (then you usually do not have to considered aspects such as time or cost since the data are already available). 

Population (N)		Study sample (n)
	Sampling

There are many different sampling techniques available. Generally, they can be categorised into two types that include several sub types: non-probability sampling and probability sampling.

Non-probability sampling

Non-probability sampling is most common in small-scale studies, marketing research, interview studies, and studies like that.

Snowball

Snowball sampling means finding respondents through already selected respondents.

Example Researchers want to examine health care utilisation among people who are homeless. They go out and find two persons who are currently experiencing homelessness and ask them where to find additional persons with the same experiences to recruit.

Quota

Quota sampling is often used in marketing research. It means adding suitable individuals until a certain quota is achieved.

Example Researchers want to have 100 respondents who have tried a new coffee brand and stands outside the store until they have recruited 100 persons who have bought that specific brand.

Convenience

Convenience sampling is when you pick respondents who are easy to get access to.

Example The researchers work at a university and recruit students taking their class to participate in the study.

Probability sampling

Random

Random probability sampling postulates that every individual in the population should have the equal chance of being selected.

Example The researchers draw a 10% random sample from the entire Swedish population aged 18-45.

Systematic

Systematic sampling means sampling with intervals.

Example Researchers might have a list of individuals who work at a large-scale company. They order the list alphabetically and draw every fifth individual from the list.

Stratified

Stratified sampling is when you draw random samples from some specific groups.

Example Lets assume that researchers want to compare labour market outcomes between native Swedes and immigrants. They may not get a large enough sample of immigrants if drawing a random sample from the entire population living in Sweden. Instead, they draw a larger random sample from the smaller group.

Clustered

Clustered sampling means random sampling of groups, choosing all individuals from these groups.

Example Researchers draw a random sample of schools and then select all students attending ninth grade in these schools.

Representativeness and generalisability

Probability sampling constitutes the foundation of quantitative data analysis. Why is it so important? Well, we want our study sample to be representative. This means that it should have the same characteristics as our population. This is a requirement to be able to draw conclusions about the population based on the study sample (also known as generalisability).

Population (N)		Study sample (n)
	Sampling
	Representativeness

Aspects of generalisability

There are many different aspects of generalisability. It is about whether it is possible to generalise the results from the sample…

To the population

Depends on the type of sampling. Can be assessed by comparing the sample characteristic with the population characteristics. In general: the bigger the sample, the better.

Example Our population consists of all children ages 2-5 in Sweden. We draw a random sample of 100 preschools and choose all children these preschools. Is the sample representative for the population?

Between populations

How unique is the population? Are there similar populations to which the results can be generalised?

Example Our population is defined as unaccompanied minors coming to Sweden from Iraq. Do our results also apply to unaccompanied minors from e.g. Afghanistan? Or to accompanied minors from Iraq?

Between interventions

Does the intervention (e.g. treatment) have to be exactly the same to generate the same results? What happens if we adjust the intervention?

Example Our population is defined as pregnant women who smokes. We include all pregnant women living in a specific Swedish city who reported that they were smoking at the time of enrolment in antenatal care. We randomise them into an intervention group and a control group. The intervention group participants in two hours of motivational interviews per week for two months. At the time of the child’s birth, a higher proportion of women in the intervention group have stopped smoking compared to those in the control group. Would we see the same effect if we reduced the number of interviews?

Between contexts

How unique is the context? Is the study culture specific?

Example Our population is a total sample of all children (0-18) living in joint custody in Sweden. Can the results be applied to other countries, such as the United States?

Over time

Are the results specific for the historical time period for which we collected the data? Are the results and interpretations valid also for today?

Example Our population consists of all children who grew up in societal care in the 1960s. The results from our analysis suggests that these children experience much worse health in adulthood, compared to those who grow up with their biological parents. Would be find the same results if we were to follow-up children in societal care in the 2010s?