Sampling

In a study, we ask a question about an entire population. However, we can’t study everyone. So we pick a small group (the sample) from the larger population. If the sample looks like the population (same number of men and women, same number of old and young, same number of tall and short) then we say that it is a representative sample. How can we guarantee we get a representative sample? We can’t. But we can improve the odds of this happening by taking a random sample.

There are several ways to take a random sample:

Simple Random Samples
Stratified Random Samples
Cluster Samples
Systematic Random Samples

Two points: 1. I sound very angry at the beginning of this video. Apparently sampling gets me upset. 2. There is nothing wrong with hearing more Beatles than a random sampling would dictate. In fact, I won’t mess with my iPod, I’ll just let it be.

Simple Random Sampling

Each sample has equal probability of occurring and each person has an equal probability of being selected. Assign every eligible person (everyone in the sampling frame) a number, then randomly pick numbers and select people for your sample.

Stratified Random Sampling

Divide the population into groups (strata) that are homogenous on a particular characteristic (perhaps gender, age, IQ, etc). Then pick a number of people randomly from each strata.

Cluster Sampling

This is a cost-effective way to sample. First you break the population up into clusters, these are groups that have something in common (perhaps geography). The individuals within the cluster represent the whole population.

Identify the clusters
Randomly pick the clusters
Examine everyone from within each cluster

This allows us to shrink our sampling frame.

Systematic Random Sampling

In systematic sampling, subjects are chosen in a methodical way. The randomization comes in when you choose where to start. If you are going to select every 10th subject who presents, you randomly pick the start point. Say we randomly pick 3, then our subjects would be the 3rd, 13th, 23rd, etc. This becomes problematic when there is some periodicity to the subjects.

The audio and video get out of sync and a bit jumpy; sorry about the technical difficulties but I think it’s still watchable. Let me know if it’s too problematic.

Non-Probability Sampling

There is no random element in this sort of sampling.

Convenience / accidental / haphazard sampling: we pick patients based on a method that is convenient for us, but confounding factors can be introduced.
Snowball sampling: one subject refers another who refers another; obviously since these people know each other, they share characteristics there are obvious confounding factors.
Purposive sampling: the researcher picks who is in the sample, so the researchers preferences can introduce bias into the sample.
Volunteer sampling: subjects volunteer to take part, so those who have self-selected have some reason for being in the study. This is not random.