Science proficiency levels in PISA refer to the various different competencies of students in science subjects. (OECD, 2007). The scores of students are divided into six proficiency levels, with levels 5 and 6 at the top (meaning that students have those competencies that are essential for the creation of innovative new technology) and Level 1 and below at the bottom (referring to students who do not possess even the most basic science skills and knowledge). 15 year-old students at levels 5-6 and at levels 0-1 are therefore the “best” and the “worst” in Europe as regards their future possibilities to participate effectively in life situations related to science and technology. The aim of this paper is to identify, for the European PISA 2006 data, the pattern of variables which are distinguishing features of those students performing at the lowest science proficiency levels compared to those students performing at the highest science proficiency levels. The theoretical framework of the study is consistent with the PISA 2006 assessment framework (OECD, 2006) that considers a large number of contextual factors which could hypothetically influence students’ performance in science. The contextual factors include: - various aspects of schools, such as the quality of human and material resources, the nature of public or private control and funding, decision-making processes; - the context of instruction, including institutional types and structures, class size; - students and their family economic, social and cultural background and resources; - students’ attitudes such as students self-efficacy, self-regulated learning, types of motivation and goals; - various aspects of students’ lives, such as their attitudes towards learning, their behavior and lifestyle at school. All contextual variables are central (OECD, 2006) to the analysis of results within the theoretical framework referring to a range of student and school characteristics. One should note that the theoretical framework does not identify a precise pattern of interactions between these variables and an explorative approach is therefore appropriate. The explorative aim of the paper includes the identification of interaction effects, not completely predictable a priori, among a large number of variables of different types (Nominal, Ordinal, Interval).

Data is based on the answers of those European students, as well as those of principals who took part in the PISA 2006 study in Europe. The dependent variable of the analysis is a dichotomous variable whose values represent the two different groups of students and the explanatory variables are derived from the contextual questionnaires (single items and/or indices. A complete list and a description of the analysed variables will be provided in the paper). The analysis will be based on classification and regression trees (CART) (Williams, Lee, Fisher, & Dickerman, 1999), a full non-parametric method suited to detecting and interpreting complex interactions that most traditional means of regression and classification analysis might ignore or find difficult to estimate and interpret (Allore, Tinetti, Araujo, Hardy, & Peduzzi, 2005). The accuracy of the final model was estimated using cross validation techniques (Breinman, 1984).

The expected outcomes include the individuation of the variables, within those included in the contextual questionnaires in PISA 2006, that have an important role (in terms of principal effects and interaction effects) in predicting who will have the highest or the lowest science proficiency levels in PISA in the European Countries. Furthermore, since classification trees derive their predictions from a few “if-then” conditions, we expect a straightforward interpretation of results in terms of the distinctive characteristics of the different groups. In other terms the results we expect will have a simple form of the following type: If the school characteristic is A (e.g. urban area), the respective student group will be most likely classified as high performer; if the school characteristic is B (e.g. rural area) and the student characteristic is Z (e.g. low self-efficacy in science), then the respective student group will be most likely classified as low performer.