The analysis is based on data collected in a large scale data survey in secondary schools in the Dutch-speaking northern part of Belgium: the Student Employment Survey (SES). The SES aims for a representative picture of the work done by adolescent students who are in the second and third grade of secondary education (i.e. year 3 to 6) and attending school full-time. Most of these pupils are between 14 and 18 years old (except for some who repeated a class). Students in part-time education thus fall outside the sample. Adolescents in special needs programmes or in the very small Arts track (2.2% of the school population) have not been included in the sampling frame.
In line with most educational research, respondents were indirectly sampled through their school. Practically and economically, it is more feasible to follow an indirect sampling strategy. An extra quality of indirect samples in schools is that unit non-response (of the higher unit) is not dependent on the characteristics of the respondent. The definite disadvantage is that respondents are clustered in schools. This problem becomes more serious when schools are larger than average (Lavallee 2007). We reduced clustering by decreasing the basic unit size in the sampling frame by dividing schools into smaller parts and making those the elementary units for sampling. The sampling frame was provided by the Flemish educational administration, and is based on data from the 2009-10 school year. Initially 60 units and 60 replacements were selected with a “probability proportional per size” sampling. The odds of a unit being selected thus depend on the number of students, so that every student has the same chance of being sampled. Of the sample, 26 initial units and 10 replacements agreed to participate.
The data were collected through written questionnaires from November 2010 to January 2011. Usually, the questionnaires were filled out in the classroom under the supervision of a school teacher, but some schools chose to assemble all the students in one place. Eventually 4,018 respondents filled out the questionnaire. Because the distribution of this response does not entirely reflect the population, a weighting was carried out on the two central variables of class year and educational track (for more information about these matters, see Adriaenssens et al. 2014).