Multilevel analysis is analysing data that are collected from different levels, for instance of schools and their students, companies and their employees, cities and their inhabitants.
Suppose the research question is something like Students in school A perform better than the students in school B, C, D … and so on. The better performance might be due to the children’s background too. Therefore, information should be gathered on the level of schools (for instance city, area, teachers’ education levels, teachers’ age, and so on) and information from the students (knowledge of the subject before training, knowledge after the training, study motivation, parents’ education, and so on). The information must be collected at two levels, that is, the data must be stored in different files: one for the students and one for the school.
To analyse these data a tool for multilevel analysis should be obtained to keep the levels separated. Here is the reason why. If the information from the schools is pasted behind the student data, there would be many more teachers than those from which data was collected; in a class with 25 students, it would seem as if there were 25 different teachers (all with the same scores). This is incorrect. On the other hand, if the data of the students would be aggregated to school level, there would be only one (average) student. This is incorrect as well.
It is hazardous to explain the calculation techniques for multilevel. So I will not do that here. However, I can and will explain what the files should look like before analysing.
For gathering the data two files should be filled with data, one with data from the teachers and one with the data from the students. Make sure a key variable is used in both files. In this case a school-ID is needed. If at some schools more teachers participate in this research, each teacher should get their own Teacher-ID. In the files of the students these ID’s should be present. Now it is possible to see which student got lessons at which school and which teacher. So far so good.
Next is to combine these data in one file. Use the data of the schools as a table look up file. The data of the school should be added after the data of the students.
Now many more types of analysis can be performed, even same level analysis. However, if you want to make use of the multilevel information the variables on the school level should be declared. Now any information on school level is used as a grouping variable (and not as a single variable for every student).
Multilevel analysis should be used more often than it is done nowadays. It is a rather new technique, not so well known and rather difficult to understand and to interpret the results. So it is not a surprise it is hardly used. To give an idea of which research questions for which multilevel analysis should be used, this is a small list:
- Do employees in companies value their boss differently?
- Do consumers in different countries think differently about the price of cars?
- Is there a difference among citizens (of several cities) in their opinions about pollution?