Random Samples & Demographics

Those that are new to research often loathe the idea that a random sample is necessary to generate useful results. Getting a random sample is far more difficult than convenience sampling, because it involves not only randomization algorithms, but also the ability to randomly sample your population – something that a lot of companies do not yet have. A common concern about random sampling is that when you randomize your sample, you run the risk of collecting a sample that doesn't match directly with your demographics.

However, while it's possible that you received a sample that wasn't accurately representative of the population as a whole, that does not mean that it isn't still representative of what you need to know. Consider the following idea: – Your store has 65% female clientele, 35% male clientele. You run a survey with a random sample, and you get an end result of 50% male and 50% female.

You assume that the sample isn't representative of the population. However, it may still be. Despite the demographics of your store, the men that shop at your store may have something in common with the women that shop there, so although your results are leaning towards males, the survey data may be the same as if the demographics were more accurate. – On the other hand, if you had picked out 65% women and 35% men, you may have used a convenience selection criteria for collecting those samples that cuts out a personality trait.

For example, if you collected your sample weekday mornings of those that came into your store, you are neglecting all of those that come in on weekends or in the evening – all of whom may have something in common that makes them different from the convenience population (for example, they may all be employed workers that cannot come in weekdays or mornings). Convenience sampling is far more likely to provide you with inaccurate results, because there is no surefire method of collecting a convenience sample that doesn't risk losing out on a subgroup of the population. Random sampling does have some risks, because like any random sequence it's possible to get a statistical improbability.

You can flip a coin 100 times and it's possible that you land with 100 straight heads. But the likelihood of that is slim, and by getting a large random sample you greatly reduce the chances of having that type of bias in the data.