Application of the Levenshtein Distance Metric for the Construction of Longitudinal Data Files

The analysis of longitudinal data in education is becoming more prevalent given the nature of testing systems constructed for No Child Left Behind Act (NCLB). However, constructing the longitudinal data files remains a significant challenge. Students move into new schools, but in many cases the unique identifiers (ID) that should remain constant for each student change. As a result, different students frequently share the same ID, and merging records for an ID that is erroneously assigned to different students clearly becomes problematic. In small data sets, quality assurance of the merge can proceed through human reviews of the data to ensure all merged records are properly joined. However, in data sets with hundreds of thousands of cases, quality assurance via human review is impossible. While the record linkage literature has many applications in other disciplines, the educational measurement literature lacks details of formal protocols that can be used for quality assurance procedures for longitudinal data files. This article presents an empirical quality assurance procedure that may be used to verify the integrity of the merges performed for longitudinal analysis. We also discuss possible extensions that would permit merges to occur even when unique identifiers are not available.     

Educational ethology and experiments in vacuo: Some implications for research and education 1

Course memories and careful observation in the field are both important if researchers are to come to grips with the problems in the real world of the public schools. We should keep in mind that our goal is the production of causal statements which potentially make some educationally significant difference. Using “experiments in vacuo”; a la Galileo is one way for researchers to keep themselves oriented toward the production of causal statements. “Experiments in vacuo” are also a means to avoid overlooking obvious simplicities or appraising nonevents. The examples given in this article illustrate the operation of causal events operating in chains and in parallel and some pitfalls of not examining, at least hypothetically, such causal chains or of distinguishing between causal events operating in series or in parallel. Avoiding such pitfalls is one way to prevent the pessimistic view that research tells us nothing and education makes no difference.