Abstract:

As Educational Analytics come to age, there is an increasing need to define standardized and agreed-upon processes and representations that allow data from educational settings to be made available in forms suitable for micro-, meso- and macro-level analyses. Micro-analysis is concerned about what goes on in a classroom or a school. Meso-level analysis, one the other hand, is concerned with analyzing behaviors, patterns and trends across schools. Finally, the macro-level analysis concerns itself with issues across school districts, provinces or states, for example. One common approach towards educational analytics is to first map the raw educational data into an online analytical processing (OLPC) platform using some variant of dimensional analysis yielding an initial data mart or data warehouse. Using this approach has the immediate advantage of the capabilities of mature data-warehousing technologies for slicing and dicing data into various forms. The warehouse can later be utilized for advanced predictive analytics. Dimensional analyses often rely on star schemas and typically employ principles that consist of viewing data as measurements called “facts” and context descriptors called “dimensions.” For example, in assessment-based analytics, the primary facts are scores on test items that can be aggregated along various dimensions like levels of a curriculum (e.g., unit, topic and student learning outcome etc.), geography (e.g., school, cluster, district, province, state etc.), time (e.g., days, months, years etc.), difficulty (e.g., using item response analysis, for example), etc. A star schema represents relationships between facts and a rich set of dimensions. While what constitute facts in an assessment-based star schema is pretty obvious, the design of a rationalized set of dimensions is not so trivial. This paper proposes the use of Bronfenbrenner’s Ecological Systems Theory (BEST) as the basis of dimensional design for educational analytics. BEST assumes that a child’s development is affected by a host of systems operating at different layers of aggregation. Microsystem layer consists of a child’s relationships with peers, school, family neighborhood etc. The mesosystem layer, on the other hand, consists of relationship between a child and her parent, the relationship between the parent and the teacher, the relationship of teacher to the community etc. The exosystem layer defines the larger social system in which the child lives; parent’s work practices and schedules, and available learning resources and opportunities within a community may have an indirect impact on a child’s learning. The macrosystem is the outermost layer in a child’s learning environment consisting of cultural values, local customs, community attitudes and laws. Finally, the chronosystem is about time and how time relates to a child’s environments and includes events such timing of a parent’s illness as well as internal developmental changes in a child. From a modelling perspective, each of these systems defines a natural set of contextual variables that may act to define and refine dimensions in dimensional analysis. This paper shows the design and evaluation of a number of star schemas based on this approach in the context of large early grade literacy and numeracy assessments in a developing country.

Keywords:

Bronfenbrenner’s Ecological Systems Theory, star schema, educational analytics.