The size of your population of interest does not usually affect the size of your sample.
I know – I know! When I say this to our clients they often look like they want to fire me. When I’m working with a client on preparing to take a sample, the first thing (and sometimes the only thing) they tell me is the size of the population.
This makes a lot of intuitive sense.
However, unless your population size is several hundred or more, it isn’t important in deciding what size of a sample you need. For example, if you want to estimate a population average at plus or minus 3% with a 95% confidence interval, about 1000 people in your random sample is going to do the trick – whether your population is a school of 2000 or a country of 1 million.
I know. I know…..
Instead of size there are actually two different characteristics about your population that you need to know in order to figure out what size of a sample to use in researching your learning agenda. These two population attributes are:
- What the indicator that you’re interested in looks like at the beginning of your study.
- How similar or different most people in the population are to the average in terms of your indicator of interest.
Once you start to look at different samples and explore different ways to collect the data, that you need to answer your research questions, it often starts to become intuitive that these two characteristics are more important than the sample size of your population. In fact, even for very large national studies, a well-selected sample of 1000 people is often sufficient.
I’m going to give you one quick example of how this works below, along with a handy table that you can use to quickly get a ballpark estimate of a good sample size. And over the next few weeks, we’re going to dig a little bit deeper into other factors that you can consider and use to help you calculate an appropriate sample size. We’ll also talk about things to do in the situation that lots of us are often in – where we calculate a lovely sample size…that is way outside of our project’s budget. There are better and worse solutions to that situation, and we’ll talk about how to handle it with grace and effectiveness.
So, let’s take for example a study where we’re interested in establishing trends in households linked to the formal banking sector.
There are the two attributes of your population that you do need in order to determine a good sample size for your study.
- The first is the starting proportion of your population that is experiencing your indicator of interest.
You need a rough estimate of the proportion of your population that is linked to the formal banking sector. You don’t need to know this exactly, just an estimate. If you can’t find an estimate from a similar study or in wisdom from the field, use an estimate of 50% in your sample size calculations.
- The second thing you need to know about your population is the level of variability in the population.
The level of variability is the range of values in the indicators of interest. For example, in a study where you’re exploring herd sizes and most people have either 1 or 2 cows in their herd, these is low variability in the population. If you’re exploring changes in income level and the annual income in your population ranges from $1000 to $2.5 million you have a population with high variability. The higher the variability of your indicator within your population, the larger your sample size is going to need to be in order to get an accurate picture of your indicators of interest.
- We will also include the acceptable error rate in the calculations.
The acceptable error rate is also called the confidence level. The confidence level is how likely it is that the results obtained from the sample fall within the associated precision. The higher the confidence level, that is, the more certain you want the results to not be atypical, the larger the sample size. We normally use 95 per cent confidence, however this is largely tradition rather than science. If you are only seeking an indication of likely population value, then a lower level, such as 90 per cent, often is quite good for your purposes.
Of course, there are other factors to consider, and we will get into these in detail in the coming weeks, such as whether you want to make estimates about specific sub-populations, whether certain groups are more likely to be highly similar or diverse, what types of clusters exist in your population. All of these things need to be added in to a more sophisticated sample size calculation to be certain you really get the research results you’re looking for.
However, if you have an idea about the size of the three attributes we discussed here, you can use the table below to get a good idea of your necessary sample size.
For example, for 5% precision with a population proportion of 40%, a sample size of 369 is required at the 95% confidence level.
As I said, along with the above handy table that you can use to quickly get a ballpark estimate of a good sample size, over the next few weeks we’re going to dig a little bit deeper into other factors that you can consider and use to help you calculate an appropriate sample size, along with addressing some other survey challenges. So stay tuned to Datassist.
…So take time right now to: