A Guide to Margin of Error, Confidence, and the Reliability vs Validity Debate

What is margin of error? 

When presenting statistics many larger organisations, such as Government departments and charities, like to understand the margin of error so that they feel confident that their findings are robust. The margin of error  informs the reader how accurately the results will represent the whole population, and when illustrated with a confidence level, will tell how often your margin of error will be accurate (the industry standard is 95%).

In practice, a 95%  confidence level with a 6% margin of error means that your statistic will be within 6% points of the real population value, 95% of the time. In simple terms by calculating a confidence level and interval you can predict that your sample will yield results that are 95% (confidence level) representative of the mean (average) answer, with a + / –  6% range. Hence, the margin of error can be used to demonstrate the reliability of research findings. 

You may have seen predictive margin of error calculators elsewhere. These are calculators that ask you to supply your sample size and desired confidence level (and potentially your population size), and then provide you with a predictive margin of error for that research. For example, you may enter a population size of 100, a sample size of 80, and a desired confidence level of 95%, to which the calculator will happily tell you your margin of error is just 3%. However, if you head over to another margin of error calculator and enter the same details, it may tell you that your margin of error will be 10%.  

The difficulty is that margin of error cannot be calculated without knowing the results, as you need to be able to calculate the mean and standard deviation; this means that the predictive calculators are guessing your standard deviation to calculate the margin of error. Predictive margin of error calculators are estimations, and will not accurately represent the margin of error you will see from your survey results. 

Don’t get hung up on numbers (the reliability vs validity debate) 

If you are seeking a highly empirical, over-time comparison of changes in your customers' perceptions of your business, via benchmarking or Brand Tracking surveys, then ideally you will be looking to achieve the lowest feasible margin of error. However, applying a margin of error does not always deliver valuable insight. If your organisation's goal is to identify specific areas of improvement for your business, to increase customer loyalty and satisfaction, then you will find much greater benefit in conducting a deeper, qualitative survey with a smaller sample. In fact, some of the most influential ethnographic studies contain data relating to just 15 respondents. 

Long responses and open-ended questions will tell you far more about how you can improve your product or customer service, even if your sample is considerably smaller, than you could have achieved with a ‘catch-all’ NPS survey. Knowing, to a confidence level of 95% and a margin of error of 2%, that the vast majority of your customers think your communication could be improved, does not necessarily tell you how to improve your communication. If you instead discovered that 17 of 25 people surveyed liked to be communicated to via weekly emails, and explained which content would be helpful, you would be able to create targeted and engaging campaigns.  

Larger samples, with applied margins of error, help us to identify themes whilst smaller samples will reveal detailed information (why and how), yet it can be difficult to determine which approach is appropriate. Our expert team will advise on the most appropriate approach and will recommend a research method based upon your projects aims and objectives to ensure that your research project delivers the absolute best insight.

If this sounds like something your company might benefit from, you can contact us here.