How many surveys do I need for my social or market research study?
​

In designing a sampling frame for a social or market research survey (i.e., an approach to sample selection), it is important to consider a range of factors. 

​

It is important to consider the volume of sample required to ensure that the selected sample provides an accurate estimate of the entire 'population' (i.e., All the respondents of interest) – if too few respondents are interviewed, 'random sampling error' may be high and imply a more biased estimate.

​

Tips for success in sampling for social and market research studies

 

The following tips are useful in determining the size of the sample you need for a survey or other data collection exercise:
 

• Consider the groups or segments of interest – to ensure that there is sufficient data to permit reliable analysis of data following data collection, the sampling frame should include all groups or segments of interest

​

• Use random sampling where possible – random sampling is the best technique to avoid sampling bias
   

• If there are groups of interest, it is also possible to sample randomly within groups of interest (called stratified sampling) – e.g., surveying people who drink alcohol and those who do not. These are often referred to sampling 'quotas'
   

• ​​​​The following table may be used to work out appropriate sample sizes for surveys. While many other more complex factors can influence sampling design, the table provides a 'ballpark' of sampling confidence intervals - or margins or error - for different sample sizes.
  
  It assumes an arbitrary population size of 5,000 (i.e., you're taking a sample of people from the total population that is estimated to around 5000 in total).
  
  It should be noted that population sizes in excess of thousands will generally not have a large effect on sample sizes required.
  
  So the following can also be considered as indicative for large populations of even 100,000 or more!
  
  This implies that there is a 95% chance that the given result is within +/- X% margin of error (confidence interval) of the obtained result.
   

  • N=357 achieves +/-5%​

  • N=536 achieves +/-4%

  • N=1622 achieves +/- 3%

​

​​

• Working out confidence intervals can also assist with research budgeting. 
  
  For example, if there is only a small improvement in sampling confidence intervals, it is arguably not worth spending additional budget on a larger sample, unless you need the sample for analysis. 
   

​

• If there are important sub-groups in the population of interest, remember that the above sample sizes should also be applied to any segments of interest.
  
  For instance, for roughly +/-5% confidence within each segment, you could sample 536 each for 3 segments (N=1608 in total) and you would achieve roughly +/- 3% for the total sample.