 |
Box 13. 4 the Glueck study. Cross-sectional and longitudinal research:the main differences. People
|
|
|
|
BOX 13. 4 THE GLUECK STUDY
In the 1940s, Sheldon and Eleanor Glueck at Harvard University designed and con- ducted one of the most impressive longitudinal studies to date, published as Unraveling Juvenile Delinquency (1950). In the 1980s, two researchers, Sampson and Laub stumbled upon the original data and research material in a basement at Harvard. They first re-analysed the material, collected new crime record and mortal- ity data, and published it as Crime in the Making: Pathways and Turning Points through Life (1993). Having collected this new data, Laub and Sampson conduct a quantitative analysis and explore the men’s recorded criminal careers all the way up to the present day. Having done so, they use this data to select 52 cases from five groups of offenders: (1) those who persistently engaged in violence and theft across the life course; (2) non-violent juvenile offenders who desisted in adulthood; (3) vio- lent juvenile offenders who desisted in adulthood; (4) intermittent offenders with an onset of violence in adulthood; and (5) intermittent offenders with an onset of vio- lence in young adulthood.
Then, through close analyses of these 52 life histories, they delve deeper into the lived experiences of the men in the various groups. By doing so, they find the under- lying processes and mechanisms at work in their age-graded theory of informal social control, but also realize the importance of human agency. Indeed, they go so far as to term it the missing link in understanding both persistence and desistance, and as such human agency provides a crucial piece of the puzzle of continuity and change in crime across the life course. This piece would not have been discovered were it not for Laub and Sampson’s qualitative data analysis (in traditional, quantita- tive analysis, human agency may have constituted the portion of the variance in the data that cannot be explained).
Let us leave you with an additional example. Let us say we find a quantitative, statistical relationship between being diagnosed with some form of cancer and desist- ance from crime. That is, cancer diagnosis is a predictor of desistance; when people get the diagnosis, they are more likely to desist than before. This can be a simple co-variation between the two variables, so we have to create additional statistical tests where we control for various factors – such as age – to make sure that the two changes (in health and criminal offending) indeed are connected to each other. Now, depending on what kind of quantitative data we have access to, we might not be able to say much more than this: other things controlled for (or ‘held constant’), cancer diagnosis is a predictor of desistance from crime.
But why is cancer diagnosis a predictor of desistance? It could be due to a number of things. First, it could be that the simple ‘shock’ of the diagnosis makes the indi- vidual turn his life around. Or, it could be that the cancer diagnosis is an indicator
|
|
|
of the individual’s health problems, and people who have severe health problems are often not capable of committing crimes simply because their bodies are not up to the task. Or, third, it could be that the everyday life of the offender now changes in important ways: s/he must go through treatment, counselling sessions, and so on. These things can entail a strengthening social control which inhibits future criminal offending. Any of these three explanations – or a combination of all three and pos- sibly additional ones! – are possible, but they all have one thing in common: the only way to find out is to undertake a qualitative analysis. You take a sub-sample of those you have studied quantitatively and ask to interview them, and ask them about it.
CROSS-SECTIONAL AND LONGITUDINAL RESEARCH: THE MAIN DIFFERENCES
A cross-sectional study usually has two dimensions: individuals and variables. In a longitudinal study, as you now know, a third dimension is added: time. In practice, this means that for every individual, not only do we collect data on traits, circum- stances and events, but also data on when these occur. This three-dimensionality in longitudinal data is described by Biljeveld and van der Kamp (1998) as a data box where every dimension is an axle: the box has the axles people, variables, and moment or time.
People
Every study that strives towards making generalizable conclusions has to handle the question about the representativity of the material. In quantitative studies, the sam- ple is usually drawn from a well-defined sample population. Additionally, the sample must have a certain size for us to be able to get stable, statistically significant findings (the larger the sample, the better, for larger samples give us stronger, statistical power, which, in turn, is necessary for statistical significance).
In so-called cohort studies, such as the Swedish Stockholm Birth Cohort (Stenberg, 2013), you choose to study a whole population of people, in this case every person born in Stockholm in 1953, and who resided in the city in 1963 (when the project was launched). In such cases, given that the attrition and missing data in the study are relatively small and random, it is possible to make claims about the studied population.
Now, a problem here is the issue of time: a cohort born in 1953 turned 60 in 2013 and only then could the researchers get a sense of how the cohort members’ lives unfolded. To answer questions regarding, for example, the connection between juve- nile delinquency and the risk of dying from different kinds of diseases, you may have to wait even longer. Thus, after many years of research, we may finally have an
answer to our question about the relationship between crime and morbidity but do these findings apply to today’s youth, or are they by now no more than historical documents?
When it comes to research about relatively rare phenomena, such as serious and persistent criminal offending, you either have to work with very large samples, or stratify your sample, so that those individuals who are especially interesting to the researcher (in our case, those with serious criminal offending) are overrepresented. The Pittsburgh Youth Study (e. g. Loeber et al., 1998) is an example of such a study (Box 13. 5).
|
|
|
 |
A difficulty for those of us engaged in prospective longitudinal studies is that seri- ous and persistent criminal offending distinguishes itself relatively late in life (e. g. after the age of 20). If we want to initiate our study before that, we have to over- sample the group of individuals who we believe will have a high risk of developing serious and criminal offending. It is not certain either that our choice of risk variables will be adequate in the long run.
Another problem is the question of attrition. Attrition is a problem for every type of research that, in one way or another, is dependent on the respondents’ consent. The whole idea with longitudinal studies is that we, in some way, study individuals through time and place. We thus repeatedly measure the same respondents. So, a respondent may answer our questions at one time, but may change his/her mind (or tire, or die, leave the country, or move to an address we for some reason cannot find) the second or third time. We must remember that, for ethical reasons, we should not
attempt to contact those who have already declined to participate. Thus, in longitu- dinal studies, the problem with attrition is usually larger than in cross-sectional ones, and, at the same time, many of the longitudinal analyses we want to make are sensi- tive to attrition and missing data (Biljeveld and van der Kamp, 1998). There are, however, various methodological and statistical techniques for handling this issue.
|
|
|
