What's the issue?
Sampling for quantitative research depends on whether or not we aim for a probabilistic sample from which we would like to draw inferences about the population (i.e. to what extent sample statistics reflect the population parameters). Usually, we have to consider a number of issues (choosing the population, the sampling frame, the way of sampling, and the sample size). When aiming for a representative sample, things get more complicated as we need to have a list of children to sample from. This can be pretty tough. We can sample households or sample children through schools.
When conducting a survey with both children and parents, the household can be used as unit of analysis (Livingstone, 1999).
If financial or time sources do not permit face-to-face surveying at home, we can decide to sample children by schools (e.g. a sample based on clusters), covering different regions of the country. Instead of individual children, we sample groups of children that occur naturally in our population. This is known as cluster sampling.
If we wish various subgroups (e.g. age subgroups or gender subgroups of children) in the sample also to be representative, we can use stratified random sampling, which combines stratified sampling with random sampling. For example, if we wanted to a stratified random sample of boys and girls from the final year of a rimary school, we would first separate the entire population of the last year of the primary school pupils into two groups, one all boys and other all girls. To complete our sampling we would then independently select a random sample from each stratum (a random sample of boys and another one of girls).
We can also do a non-probabilistic sample of children, bearing in mind that no inferences beyond our sample are possible. However, studies with non-probabilistic samples (e.g. quota sample, purposive sample) are still valuable as they can be very informative, and also point to the children beyond our sample which most probably have very similar socio-demographic characteristics to those included in our sample. It is OK to conduct such studies as long as we are not aiming for statistical inferences from the samples to the population. We operate only within descriptive interpretations.
Questions to consider
What size should our sample be? Do we need probabilistic sampling? Can we afford to sample probabilistically? What kind of natural clusters of children are available in our population? Do we also need various subgroups in our sample to be representative?
Kalton, G. (1983). Introduction to survey sampling. Newbury Park, CA: Sage.
A researcher's experience
In designing a national survey for children, as it was too expensive to interview children in their households, it was decided to sample children by schools. This sample was based on clusters covering different regions of the country. After negotiation with the Portuguese Minister of Education, it was agreed that in each of five regions, four elementary schools attended by children (6-15 years) would be selected based on the criteria of urban/rural contexts, children from ethnic minorities and socio-economic status (SES). Based on lists of students in each of the 20 schools, a proportional sample of children by age would be designed and 30 students from each school were then chosen randomly. This way, the sampling would involve 600 students. After parental consent had been obtained, the sample would receive a self-completion questionnaire to be answered at school, outside the classroom. Parents would receive another self-completion questionnaire, given to them by the child in the study. These questionnaires were to be returned to the school, in closed envelopes, and the school would send them to the research project.
This initial design for a national survey proved to be too difficult and time consuming. It involved several factors, starting with the agreement of the schools randomly selected and ending with the parental consent of all the students randomly sampled.
Instead, it was decided to sample children by schools in the greater Lisbon area, which is the leading area for internet penetration in households and the area with more migrant children. The Minister of Education provided us with a list of the public elementary schools covering compulsory education in this area. From this list, 20 schools were selected based on the criteria of urban/rural contexts, children from ethnic minorities and SES. The first 11 schools that accepted the idea were our sample. Each school chose a class per year from the 4th to the 8th grade, providing an average of 90 children as a starting point. Parents were asked for informed consent. In each school, children who had parental consent were presented with the aims of the research and invited to participate, under the assurance of privacy and confidentiality. The self-completion questionnaire was answered at school, in the presence of an assistant, a member of the research team. Parents received another self-completion questionnaire, given to them by the child in the study. These questionnaires were returned to the school, in closed envelopes. In order to provide identification, children and parents' questionnaires had the same code number.
In the end, a total of 810 questionnaires answered by children at school and 630 questionnaires answered by their parents were sent to the research team, which might be considered a quite positive number. Parents who answered this questionnaire differ from the national profile - they are much more info-included and have higher levels of education. Also, parents of younger children (9-11) were overrepresented compared to the parents of older ones (12-14), and this may have different meanings, including the possibility that the older children may have resisted involving their parents. (Cristina Ponte, Portugal)
A researcher's experience
In the TIRO research project (see Annex C) we organised two panels of 20 Dutch and 20 French speaking teenagers (aged 12-18). We interviewed these, had online conversations with them on several occasions and asked them to keep a diary on their everyday life and media use. For sampling those panels we went in different sites where young people are present (schools, youth movements (e.g. scouts) and youth clubs (sport, theatre)) and we used our own social networks, although no close relatives were selected, only casual acquaintances. In order to manage the subjectivity in the sampling process (two researchers were involved and we wanted to avoid discrepancy between the Flemish and Walloon panel), we used a theoretical sampling matrix. First, the hundreds of young people we recruited were asked to provide short information about their social background, ICT use and leisure. Based on a literature review we then decided to sample both panels by means of three criteria that seemed to be distinct for explaining the diversity and heterogeneity of young people's Internet practices: gender, age (aged 12-13; 14-16; 17-18) and SES (reflecting the economic and cultural capital of the parents). Based on these three sociodemographic characteristics we drew a matrix with 18 cells and looked for young people that met the cell criteria that were preconceived (e.g. 1 boy aged 12-13 years with a low SES, 1 girl aged 14-16 years with high SES). To gain insight into future trends in ICT use, we also selected in each panel one teenager that showed an intensive pattern of ICT use. This sampling procedure (in stages and pre-structured) proved to be useful got guaranteeing the diversity of the panel. We wanted especially to avoid assembling a middle-class panel, since many qualitative studies seem to suffer from this bias. Yet, we did not succeed in involving young people with an ethnic minority background in our panel. More specific sampling methods seem to be required for including those groups. (Joke Bauwens, Belgium)