LE BLOG BY SATISFACTORY
Representativeness and significance of the studies
You regularly hear the terms representative sample and significant results. However, when it comes to analysing your results, these concepts can seem abstract to you. In this article, we’ll review these two statistical notions to help you better interpret your results and put them to use.
Rarely can a survey be taken of all individuals within a population. You must first create a survey sample .
A sample is considered representative when it has the same characteristics as the parent population. The characteristics must be able to differentiate this population from others. These characteristics depend on the sector where the survey is being carried out.
A business that offers B2B services will focus on its clients’ revenue and sector, whereas a B2C business will work with its customers’ characteristics (sex, socio-professional category, age range, etc.).
The sample of respondents is not representative of the parent population, who has 20% customer service contacts. If we conserve the sample without statistical correction, results for “Customer service contact” will have greater weight than in reality. The results will be directly impacted by poor customer service results.
Overall satisfaction level without correction = 39.4%
Overall satisfaction level with correction = 52.6 %
In case of self-administered surveys, whose results depend on the “good will” of individuals, any analysis must begin by verifying the representativeness of the sample obtained. If the sample is not representative, results must be corrected before drawing any conclusions.
Significance and Confidence Interval (CI)
When working with a sample, you will get an “estimation” of phenomena that exist within the parent population. Involuntarily, you agree to work with a margin of error, or a gap between the sample’s level of satisfaction and the levels seen in reality.
The margin of error (or confidence interval) is measured from:
The significance threshold helps you determine the degree of certainty in your results. If your figures are 95% significant, that means that your results are within the confidence interval with 95% certainty. In other words, there is a 5% risk that your results are outside of the calculated confidence interval.
Generally, we consider results significant when they are within the 5% significance threshold. However, for estimating human behaviour or observations, you should not completely ignore results where the significance is between 5% and 10%.
Common statistics rules, used to calculate the significance threshold, apply to large samples. Under 30 responses, the parameters are unusable according to the laws of statistics. That’s why Sharing-Data offers comments to help you understand results and conclusions.
An example of sample size
Let’s say you want to conduct a survey by telephone. For cost reasons, you do not want to call your 1,000 customers. The graph below lets you determine the optimal sample size depending on the margin of error and level of confidence you consider acceptable. The more precise you want your results to be, the larger your sample will need to be.
To obtain a result with 95% confidence and a margin of error of 2%, you will need to survey 706 customers.
The larger the sample, the more precise the results will be.
 Different ways of creating a sample exist, each with their advantages and disadvantages.
 To correct a sample, we assign each response a weight depending on which group they belong to. If 55% of respondents belong to group A compared to 70% of the parent population, we’ll give a weight of 70%/55% to each respondent in group A.
 The formula for calculating the confidence interval is:
where f = measure of the phenomenon
n = sample size