Using Bayesian statistics for statistical inferences with non-representative samples (based on an example of the research of the dropout bachelor students in the Faculty of Philosophy of the Sofia University “St. Kliment Ohridski”)

Kaloyan Haralampiev

Sofia University “St. Kliment Ohrdski”


Abstract: Very often in the course of fieldwork, there are difficulties which lead to a situation in which a sample that has been planned to be representative is actually non-representative. The most common problem is the high proportion of undiscovered respondents and/or non-respondents. In this case, it is not correct to use the classic statistical methods for confidence intervals and/or hypotheses testing. There is a need to use Bayesian statistics, which allows confidence intervals to be constructed and statistical hypotheses to be tested based on non-representative sampling data. This is exactly the case in the study of the dropout Bachelor students in the Faculty of Philosophy of Sofia University “St. Kliment Ohridski”. The dropout study had been planned to be exhaustive, but due to the low percentage of filled-in questionnaires, it was implemented as a non-representative sample. The study of the control group of students continuing their education had been planned as a representative sample, but due to a high percentage of undiscovered and/or non-responded students, it was again implemented as a non-representative sample. This required the comparison between the two groups to be made by use of Bayesian statistics.

Keywords: representative samples, non-representative samples, statistical inferences, confidence intervals, hypotheses testing, Bayesian statistics.

Rhetoric and Communications E-journal, Issue 36, September 2018,,, ISSN 1314-4464

Read the original of the text (in Bulgarian)