Note: This very brief analysis is something fun that I mainly wrote to organize some thoughts and explore somewhat random hypotheses. This is not to be taken as conclusive scientific analysis. See limitations at the end.
I'm revisiting some projects from last year as I look to update them with new waves of data and create drafts for journal submission. One project that has been on the back burner is analysis of the Second International Self-Reported Delinquency Study 2005-2007 (ISRD). Of particular interest is the measurement of student immigrants and perceptions of discrimination. There are several variables of interest that measure neighborhood disorganization, acculturation, delinquency, and substance abuse. ISRD 3 is still underway, so I thought I would put some initial findings here to prepare for the update. This is just a little exploratory analysis, nothing groundbreaking, but I find it interesting.
The study itself was designed to gather information about student delinquency and behavioral risk factors in several countries. The study was deployed in 31 mostly European countries with a standard questionnaire in 128 cities and 1,375 schools. The response rate varied from around 70% to 85% (not wonderful but okay), and some countries did not use population weights making the data somewhat challenging to wrangle properly. Countries that did not use the suggested complex survey design are excluded from exploratory analysis.
We begin with a sample of 56,542 students in grades seven through twelve.
We can see Western Europe has the largest proportion of student immigrants. Given this study is ten years old, the next wave will likely show a greater proportion.
We begin with a logistic regression with a response variable of self-reported discrimination reported in odds ratios. We are most interested in the interaction term DISCRIM><IMM (immigrant students reporting discrimination), leaving Austria, Germany, Italy, Lithuania, Poland, Portugal, Russia, and Spain open for closer analysis for contextual factors that may explain why immigrant students in those countries self-report discrimination.
A bar chart helps visualize this better.
Interestingly, countries that report higher satisfaction with democracy as a system of government report less discrimination.
An explanatory variable of interest is disorganization in schools and neighborhoods. Disorganization is an aggregate of several variables measuring community cohesiveness, access to resources, neighborhood amenities, and types of peer groups students are exposed to, along with the built environment. We now run a multi-level model examining school disorganization predicted by neighborhood disorganization at the school level, country level, and country cluster level.
What we see here is neighborhood disorganization is a strong predictor of school disorganization. However, we have to use caution when making claims of causality. School disorganization is also a predictor of neighborhood disorganization. A question in criminology, and social science in general, is how contextual factors spread from a locus of interest to another. However, often our data does not allow us to point the causal arrow in a particular direction. What we can see, however, is at the country level and at the country cluster level, there is no significance, meaning this is occurring regardless of geography. To examine more closely, we examine fitted values with random effects with our overall mean at xb and random factors as Zu
There is little difference between our grand mean of countries versus country clusters, as far as disorganization in schools or neighborhoods. We now fit a logistic regression with outcome variable having experienced discrimination with demographics and risk factors as predictors.
Unsurprisingly, immigrant students are much more likely to report discrimination than non-immigrants, especially male immigrant students. Immigrant students coming from single parent households where their native language is spoken rather than that of the country they live in is also a significant predictor of experiencing discrimination, along with neighborhood and school measures. The less cohesive a neighborhood is, the more likely immigrant students report discrimination. Risk factors, such as truancy and measures of self-control also are a strong predictor of experiencing discrimination.
A closer look at interaction terms shows neighborhood disorganization affects immigrant students in 10th and 12th grade the most.
We can visualize this to get a better idea of how students experience discrimination by grade level. (This would probably look better as a stacked bar chart.)
We also find that the ethnic composition of students' friends has an impact on self-reporting discrimination. The more friends a student has that are not native to the country they are living in, the probability of self-reporting discrimination increases.
Back to our chicken/egg quandary of causality, what if discrimination is contributing to school disorganization? We now fit an ordinary least squares regression with outcome variable school disorganization with several predictors from our previous model.
One thing we can be certain of is that disorder at the school and neighborhood level are strongly correlated with behavioral risk factors.
Another outcome variable we can look at is probability of a student committing a criminal offense. Interestingly, we see similar results as reported discrimination.
We can see differences by gender, family composition, and whether or not the student is an immigrant. Broken down by native status and grade level, we can see, like above, that neighborhood disorganization has a significant effect on outcome variables.
With these data and exploratory analysis, we see significant variations between gender, family composition, and community disorganization as predictors for both self-reported discrimination and probability that a student is delinquent. We also see that family composition, friendship group composition, discrimination, behavioral risk factors, attitudes toward violence, grade level, and contextual factors of schools, neighborhoods, and countries all contribute to social disorganization.
However, this exploratory analysis is with its limitations. It is cross-sectional, so the causal direction is unclear. Further, there is a lack of detail at the individual level and the study relies on self-reporting, which can be prone to bias and error. When the new round of ISRDS becomes available, longitudinal analysis of discrimination and delinquency may then be possible with causality. Since this study collected data from 2005-2007, it will certainly be interesting to see how the Arab Spring and mass migration into Europe has affected student immigrants from this time period to the next.