In representative studies, you want to generalise sample results for the whole population. In order to achieve this, you often need to create an analytical weight, which may, depending on its type, compensate for:
- the uneven probability of inclusion of units in the sample (inclusion weight);
- effects of a failure to obtain data about some people randomly selected for the sample (non-response rate weight);
- differences between the structure of the sample and the structure of the population (post-stratification weight).
This is an example of a survey where the latter, post-stratification weight, was used as the analytical weight.
After the vacation season, a travel agency wanted to gain some insight into the satisfaction of its customers with their services. The survey involved random customers who went on holiday in July–August 2017 with the agency (to keep things simple, we will refer to them as ‘all customers’). It was important to make sure that the results reflect opinions of all customers in the whole population. Using data in the agency's database such as age, sex, and whether customers travelled with children, weights could be designed to make the structure of the sample similar to the structure of all customers of the agency. The general customer satisfaction results were then presented in a table.
Travel agency customer satisfaction
Let’s focus on the weighted data first. According to the survey, over 66% of customers were very satisfied. There was, however, also a group of dissatisfied customers. It can be estimated at 10% (including 5.6% of very dissatisfied customers). It is quite a number but if you look at the ‘Unweighted percentage’ column, you can see that the analytical weight adjusts the results for the population slightly to favour satisfied customers.
Why did the agency decide to present unweighted results as well? As it happens, the marketing department has an idea how to use them in practice. They are going to reach out to very satisfied customers to ask them to share their opinions with friends on social media. They will also attempt to convert dissatisfied customers into satisfied ones by offering them some kind of compensation by phone. The unweighted column gives them the number of customers the marketing department has to contact.
As you can see, the right statistics in a table can provide valuable information to various departments. Information from a single table can be used for strategic planning and for operational purposes.