220919P - SAMPLING IN RESEARCH

By Professor Omar Hasan Kasule Sr. MB ChB (MUK), MPH (Harvard), DrPH (Harvard) Consultant on Research Almaarefa University

Empirical scientific research is observing and measuring objects and events to be able to reach generalizable knowledge. Generalization would be easily achieved if all objects and events are observed or measured but this is logistically impossible. For example, if we wanted to reach generalizable knowledge on distribution of blood pressure in the Saudi population we would have to locate and measure each one of the over 40 million citizens and residents which is an impossible task. Even if for argument’s sake we say we can reach all the 40+ million the data will be of low quality because it is impossible to recruit enough researchers and travel to all cities, valleys and mountains of the country and measure blood pressure in the same standard way. Data collected on so many will have many mistakes and will be of low quality. We therefore have no choice but to study a small group of people let us say 5000 in the hope that data from them can be correctly collected so that it is generalizable to apply to the whole country.

If we choose the sample from the whole population randomly i.e., each member has an equal chance of being in the sample, we say that we have a representative or scientific sample. All researchers, therefore, aim at random samples. Statistical formulas were developed and worked most efficiently with random samples. Selection of a random sample is not easy in practice. We may use a computer to generate a sample from a database of the population if we have one. We may also choose a stratified random sample which is selecting a random sample separately in each group let us males and females and then combining. We may also select the sample in stages (multi-stage) for example we pick 10,000 and then pick randomly pick 500 from them. There are situations in which we cannot select individuals and we resort to selecting groups or clusters. For example, we may select 10 houses at random and form a cluster of 5 neighboring houses around each index house. Our sample will then be the inhabitants of the index and surrounding houses.

In many situations, it is not possible to choose a random sample and we resort to the less accurate non-random or non-scientific sampling. We shall discuss these nonrandom samples in the next article.