Examining the Factor Structure of the Danielson Framework for Teaching, Given Its Intended/Purported Uses

• In Event: 68.029 - Key Teacher Evaluation/Accountability System Components: Using Diverse Methods to Inform Practice and Policy

Abstract

Purpose

Nationally, the Danielson Framework for Teaching (FFT) is one of the most utilized observational approaches used for evaluating the instructional competencies of classroom teachers (Danielson Group, 2015). Its original intent was to provide “the foundation of a school or district's mentoring, coaching, [and] professional development” (Danielson Group, 2013). The researcher in this study examined whether this was “true,” specifically interrogating this framework’s empirical factor structure.

Perspectives

Recent federal and state legislation have effectively repurposed this and other similar observation frameworks as accountability instruments, requiring the quantification of teacher performance/observational ratings for high-stakes, decision-making efforts (White House, 2009; USDOE, 2015). Because of these increased stakes, construct validation of such frameworks becomes critical (AERA, APA, & NCME, 2014). Yet currently, little empirical information has been published regarding the theoretical structures posited by these instruments, including this popular instrument in particular.

Methods

The researcher examined the FFT framework’s empirical factor structure, employing confirmatory (CFA) and exploratory (EFA) factor analytic techniques under conditions of non-multivariate normality and multilevel (nested) data (Muthen, 1991, 1994; Jackson, Gillaspy, & Purc-Stephenson, 2009; Pornprasertmanit, Lee, & Preacher, 2014; Reise, Ventura, Nuechterlein, & Kim, 2005).

The researcher answered three primary research questions: (1) Was the hypothesized four-domain FFT framework supported by empirical evidence? (2) Did the hypothesized domain structure empirically discriminate unique attributes of pedagogical practice? (3) Did attending to violations of normality and hierarchy assumptions impact inferential findings? The researcher derived data for analyses from a large suburban school district (N = 238 self-contained, regular education teachers, grades 3-6).

Results

The researcher estimated single-level CFA models (correcting for non-multivariate normality) for both uncorrelated and correlated factor structures. Uncorrelated CFA models reported poor fit to the posited FFT framework [χ2 (209, N = 238) = 8109.019, p < .001; RMSEA = .399, p < .001; CFI = .155]. In addition, factor score correlations ranged from .70 to .86 suggesting poor domain discrimination. Single-level correlated CFA models reports improved fit statistics, but retained poor discriminate characteristics [χ2(204, N = 231) = 369.691, p < .001; RMSEA = .058, p < .001; CFI = .982]. As before, factor correlations were high ranging from .82 to .92.

Multilevel correlated CFA models (correcting for both non-multivariate normality and multi-level structure) showed additional improvement over single level correlated models [within-level: χ2 (406, N = 238) = 478.68, p < .001; RMSEA = .027; CFI = .986] but still retained poor factor discrimination (r = .829 - .942). Subsequent multi-level EFA models suggested a two-factor (within) latent structure with poor alignment of components loadings to hypothesized domains.

Significance

Results indicate that the posited four-domain FFT structure may not be supported by empirical data. Likewise, there were questionable structural characteristics with, and problematic measurement conditions surrounding the Danielson FFT, which is especially concerning given current conditions of high-stakes accountability and use. Results contribute to the limited empirical literature in this area, while also demonstrating that approaches for applying multilevel latent factor methods can be applied to and used within public policy settings, given policy implications.

Author

• Edward F. Sloat, Arizona State University