- Year 2022
- NSF Noyce Award # 1949530
- First Name John
- Last Name Settlage
- Discipline Other:STEM Teacher Education
Ben Wasserman, University of Connecticut
Children are harmed if their schools suffer from staffing instabilities among the math and science teaching faculty because of exacerbated inequities in educational opportunities – especially in schools serving low-income communities (Ronfeldt et al., 2013). This is acknowledged by Noyce including high teacher turnover as a criterion for schools defined as “high needs.” However, knowing STEM teacher departures contribute to educational equities has yet to translate into effective models for explaining, let alone remedying, the phenomena of science and math teacher departures at rates exceeding any teacher credentialing areas other than special education (Ingersoll & May, 2012). In short, increased recruitment and accelerated preparation of math and science teachers are futile until we can fix the leaking bucket (Ingersoll, 2001). We contend that a conceptual shift is necessary to reframe teacher attrition and retention as something other than individual choices (e.g., Grillo & Kier, 2021; Olitsky et al. 2021) by instead attending to institutional attributes affecting the capacity of schools to keep teachers that other schools seem unable to retain (Han & Hur, 2021). More than simply identify the features of schools corresponding to STEM teachers staying or leaving, this project seeks to describe schools at the departmental, school and district level that would reduce the high rates of STEM teacher attrition.
As mandated by state and federal governments, school systems are required to collect and submit a wide array of data: expenditures, attendance, staffing, college entrance, graduation rates, course offerings, etc. Ordinarily, secondary schools are described – and hence categorized – along few variables, typically student demographics (i.e., distribution of students from difference ethno-racial categories and proportion of students qualifying for free/reduced price meals) and student performance (e.g., graduation completion rates plus standardized test scores). Simplicity may be convenient but loses a great deal in terms of actional information. Further, in terms of investigating variables that may explain STEM teacher attrition and retention, there is not yet a strong theory upon which to build a model. We wanted to know if multivariate ordination analyses, common in studies of ecological diversity and communities, can be used to understand the relationship between a desired school outcomes (disparities in test scores, teacher retention, etc.) and the multivariate ecosystem of potentially predictors. This is especially useful because there are many measures of schools without a clear empirical sense for which among those might relate to science and math teacher mobility.
From the social sciences, we rely on Bronfenbrenner (1999) who recommends looking on education at more than the micro or individual level by also considering the immediate and broader (meso and macro) layers within which the micro-level is nested. From the natural sciences, we borrow from ecological paradigms as well as analytical approaches. More specifically, given the absence of a robust systems theory of teaching movement and mobility, this exploratory study borrows from community ecology’s ordination analyses. The goal of this approach is to use statistical tools to reduce the complexities of a multi-dimensional dataset to those components that maximize the meaningfulness in the data variability (Clapham, 2011). Although not commonly used in STEM education research, reducing large collections of data through ordination is standard practice in community ecology (Digby & Kempton, 2012; Legendre & Legendre, 2012). Such analyses calculate degrees of interschool similarity and dissimilarity (as a quantified “distance”) to generate categories. Rather than describing evolutionary relationships among populations, ordination identifies latent relationships associated with niches, behavior, habitats, abiotic factors, etc. (Adams et al., 2019; Belda et al., 2021; Cao et al., 1997). Further, such analyses promise insights to inform ecological management (Abdelhady, 2021).
We compared three methods of predicting a school’s test scores disaggregated by race: (a) percent of students qualifying for free/reduced priced lunch (FRPL), (b) an unconstrained ordination (non-metric multidimensional scaling) that categorizes schools based on the predictors, and (c) a constrained ordination (partial-least squares regression) which finds the axes of maximum covariation between predictors and response. The ordinations were based on several school/department level predictors. We used 10-fold cross validation to compare linear mixed effects models based on these three. We calculated the expected standard deviation of test scores for the three racial categories (Black, Hispanic/Latino, and White) in a school and show how that correlated to the predictors. We repeat these analyses separately for math and science. For both math and science we found that multivariate models outperformed the FRPL-based models. Math and science ordinations were ultimately similar, reflecting the relative impacts and abundance of school-level factors over department-level factors in our analysis. While FRPL-based models underperformed, it’s important to point out that FRPL loaded heavily on the primary axis in all of our ordinations. As we move ahead to determine how well the school typology explains patterns of teacher transitions between schools. This will involve attending to the varied paths STEM teachers can take within their career paths (Larkin et al., 2022).
These findings present a robust and empirical framework for categorizing schools in a state where no such formal system exists. FRPL remains a useful proxy of a school’s need. However, it is one that is not always actionable. Our analysis offers a typology of schools that state officials, school, and district administrators can use to identify and address those predictors that they can influence in order to reduce disparities in their schools. This analytical approach can be used to address many problems schools face, and we plan to address STEM teacher retention next.Abdelhady, A. A. (2021). Anthropogenic-induced environmental changes in the Nile-delta and their consequences on molluscan biodiversity and community structure. Ecological Indicators, 126, 107654.Adams, B. T., Matthews, S. N., Peters, M. P., Prasad, A., & Iverson, L. R. (2019). Mapping floristic gradients of forest composition using an ordination-regression approach with LANDSAT OLI and terrain data in the Central Hardwoods region. Forest Ecology and Management, 434, 87-98.Belda, I., Gobbi, A., Ruiz. J., de Celis, M., Ortiz-Álvarez, R., Acedo, A., Santos, A. (2021). Microbiomics to define wine terroir. In: A. Cifuentes A, Ed.) Comprehensive foodomics. (pp. 438–451). Elsevier.Bronfenbrenner, U. (1999). Environments in developmental perspective: Theoretical and operational models. In S. L. Friedman & T. D. Wachs (Eds.), Measuring environment across the life span: Emerging methods and concepts (p. 3–28).