Results (
Indonesian) 1: 
[Copy]Copied!
to generalise to a specific subpopulation that is too small to be reliablypicked up in any but the largest of samples. We might, for example, wantto compare the well-being of students in private and state-run schools.Taking a random sample of 1,000 pupils may leave us with only a verysmall group of students in private schools. Therefore, to ensure a suitablylarge number in both, we might want to use stratified random sampling.Doing this involves first dividing the population into the groups wewant to study, in this case private and state-school attendees, and thenrandomly sampling from each group separately, so we would obtain asample of 500 pupils in private and 500 in state-run schools.Sometimes, we may want to ensure that different subgroups are representedin our sample in accordance with their presence in thepopulation. Again, unless you take a very large sample, this will be difficultto achieve for small subgroups. Therefore, we sometimes specify inadvance what proportion of those groups we want to have in our sampleand sample until that quota is fulfilled. For example, we may have a populationin which 10 per cent of pupils are of Afro-Caribbean descent. Inquota sampling, as this method is called, we will sample Afro-Caribbeansuntil we have reached our quota, in this case 10 per cent of 1,000, or 100Afro-Caribbeans.Another reason not to use simple random sampling lies in the problemof being able to draw conclusions about sites in which members of thepopulation are nested. For example, in educational research we are ofteninterested in things happening in schools, or school effects, and how thesemay influence students in those schools. Saying anything about school (orclassroom teaching) effects would be difficult if we used simple randomsampling. Even if we were to have a large sample of 100 students, it islikely that they would be spread over a very large number of schools,meaning that in most cases we would have one pupil or maybe two in anygiven school. Obviously, it would be nonsensical to extrapolate effects ofthe school or teacher from findings on one pupil in that school! Therefore,when we want to look at school effects we will usually sample schools randomly,and then survey all pupils in that school. More generally, usingcluster sampling we will randomly sample higher-level sites in which membersof the population are clustered, and then survey all respondents inthose sites. A related method is multistage sampling in which we firstsample higher-level sites (e.g. local education authorities) at random, thenrandomly sample a lower stage (e.g. schools in those LEAs), and then randomlysample members of the population in that stage (e.g. pupils within
Being translated, please wait..
