Place Matters, Missouri, and the Conundrum of Similarity: The Utility of Spatial Mapping in Understanding Variable Relationships Across School Districts

• In Event: Cultural-Ecological Frameworks for Understanding and Impacting Opportunity to Learn

Abstract

The concept of place in the study of local context as a part of ecological studies of education is important because specific location variable values tend to be more similar, which produces a clustering effect. Most researchers are now aware of the need to account for correlations in data that originate from the same group or cluster. Data that come from the same group tend to be correlated in that the factors unique to a specific group (e.g., local context or place) influence all group members in a similar way. When data points within groups are dependent, then statistical methods assume independent observations underestimate standard errors and increase Type I error rates. In cases where observations are nested within groups, multi-level models (MLM) are appropriate to account for correlation among group members (Raudenbush & Bryk, 2002). However, what method provides appropriate estimates about situations where the grouping or clustering is based on spatial proximity such as adjacent schools or districts? Often students within a district tend to be similar. If the spatial unit of analysis is a school district, then at what point do students in one district become different from students in another district? If similarity is defined by a district boundary such as a street, then it is unlikely that students on one side of the boundary are much different that those on the other side. Yet, they may be different from students in a district twenty miles away. The major takeaway is that similarity based on spatial proximity is better understood as a continuum which does not start and stop at artificial lines such as district boundaries. Thus, in the case of spatial clustering, multilevel modeling is not appropriate because it focuses on within group correlations and ignores similarity between adjacent clusters (Fotheringham, Brunsdon, & Charlton, 2002). In several studies, Hogrebe and Tate illustrate how geographically weighted regression (GWR) provides more ecologically valid estimates by accounting for the spatial clustering of school districts (Hogrebe & Tate, 2012 [b]; Hogrebe & Tate, in press). Like MLM, GWR allows relationships to vary across school districts, but also takes into account the underlying spatial continuum that MLM ignores. Mapping the GWR results with GIS shows that variable relationships are not constant across districts and the underlying spatial continuum. One implication for state policymakers is that reform must take place into consideration when considering policy and programmatic solutions across a state. One size fit all solutions may not be the right course of action. One implication for researchers is that the careful study of the local context is warranted including anthropological approaches that seek to describe place in greater detail. Moreover, the transferability of findings to new groups (places) must be considered with great caution. Empirical examples from the state of Missouri will be used to illustrate the problem space.

Author

• William F. Tate, Washington University in St. Louis