Fostering sustained teacher learning: a longitudinal assessment of the influence of vision building and goal interdependence on information sharing

ABSTRACT

To support school improvement, understanding the mechanisms that enhance teachers’ engagement in professional learning activities within schools over time is paramount. The purpose of this three-wave longitudinal study is to examine the role of workplace conditions (school leaders’ vision building and teams’ shared goals), in supporting teachers’ engagement in information sharing over time. To test the directionality of the relationships between the concepts, we analyzed survey data from 655 vocational education and training teachers in the Netherlands using a cross-lagged panel model. Results suggest that teachers’ engagement in information sharing remains stable over time, and the results are indicative of reciprocity between goal interdependence and vision building. Mostly, the results hint at the complexity of the time-based relations involved in teacher learning in support of school improvement. Recommendations for future designs and methodologies to understand this complexity are discussed.

KEYWORDS:

• Transformational leadership vision building
• goal interdependence
• sustained teacher information sharing
• longitudinal reciprocal relations
• cross-lagged panel model

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes on contributors

Arnoud Oude Groote Beverborg works as a post-doc at the Department of Educational Research and the Centre for School Effectiveness and School Improvement of the Johannes Gutenberg-University Mainz. His work concentrates on longitudinal and reciprocal relations between psychological factors, workplace conditions, and leadership, for sustained engagement in professional learning. Additional to his interest in enhancing school change capacity, he is developing dynamic conceptualizations and operationalizations of workplace and organizational learning, and he explores the application of dynamic systems modeling techniques for school contexts.

Peter Sleegers is a senior researcher and consultant at BMC Consultancy. He has published extensively on leadership, innovation, and educational policy in different high-impact journals and several edited books in the field of educational administration and school improvement. Current projects are design studies into professional learning communities and student learning, and longitudinal research on building school-wide capacity for sustained school improvement.

Nienke Moolenaar is a faculty member at the Department of Education at Utrecht University, the Netherlands. In her research, she explores how educators’ social networks change during educational reform. Drawing on complexity theory and literature on dynamic systems, she aims to understand how this network change supports and constrains school improvement in terms of teachers’ instructional practice and student achievement.

Klaas van Veen is professor in educational studies at the University of Groningen and director of the teacher education program. His research concerns the pedagogy of how to organize teacher learning in the context of secondary and higher education.

Notes

1 The missings that are relevant for this study are the missings on time points. Additionally, not all teachers within their teams responded.

2 Many researchers base the interpretation of fit measures on the article from Hu and Bentler (1999), and consider the following values acceptable: a chi-square (Χ²(df)) that is not significant, an RMSEA ≤ .06, a CFI > .95, and an SRMR ≤ .08. Note, however, that these cutoff criteria are to be seen as guidelines that appear to be best applicable for comparing nested models. These values are neither absolute rules nor rules of thumb, and may even be too strict in some cases (Hu & Bentler, 1999; Marsh et al., 2004). For instance, with a sample size larger than 400, Χ² will almost always reach significance (Kenny, 2015), and experiential evidence suggests that path models with multiple factors that are based on a solid number of items will have unacceptable CFI and RSMEA measures in comparison to the criteria (Marsh et al., 2004, 2005). Accordingly, acceptable fit measures could be obtained by returning to simpler models (e.g., by reducing the sample size, the amount of factors, or the amount of items), but such actions would then undermine the validity of the study otherwise. Therefore, the plausibility of the model will have to be assessed with different criteria (see also Byrne, 2001). Instead of values meeting some externally imposed standard, the idea that progress over studies in fit values that are obtained from similar studies represents development seems to be more important, as proposed by different scholars (Bollen, 1989; Marsh et al., 2004).

3 In comparison, a longitudinal study, conducted in the same context, that assessed similar concepts and instruments reported the following fit measures for the measurement model: Χ²(2977) = 6055.275 (p = 0.000), RMSEA = 0.040, CFI = 0.838, SRMR = 0.073 (Oude Groote Beverborg, Sleegers, Endedijk, & van Veen, 2015a).

4 Note that three conditions need to be met in order to infer causality: that cause and effect are related, that the effect follows the cause in time, and that other competing explanations therefore can be ruled out (Eschleman & LaHuis, 2014; Popper, 1959). This longitudinal study meets the first two conditions, but it does not meet the third.

5 Three ΔΧ² tests indicated that a model with invariant autoregressions of information sharing did not worsen the fit to the data as compared to the full model, but that the models with invariant autoregressions of goal interdependence and vision building, respectively, did worsen the fit to the data as compared to the full model. Moreover, because each pair of cross-lagged regressions differed substantially over time, we reckoned that further invariance tests would provide invalid results.

6 We considered the adequacy of the procedure by examining whether following another procedure, in which we restrained each of the pairs of cross-lagged regressions in the full model in separate tests, would lead to different results. We tested six models, which each had the cross-lagged regressions from one construct to one other set to 0, against the full CLP model: no cross-lagged regressions from vision building to goal interdependence (ΔΧ²(2) = 4.634, p = .099) and information sharing (ΔΧ²(2) = 15.212, p = .000), from goal interdependence to vision building (ΔΧ²(2) = 6.099, p = .047) and information sharing (ΔΧ²(2) = 0.219, p = .896), and from information sharing to vision building (ΔΧ²(2) = 0.416, p = .812) and goal interdependence (ΔΧ²(2) = 3.770, p = .152). Then, we combined the results thereof in a new restrained model and tested that against the full CLP model (ΔΧ²(8) = 7.615, p = .472). Then, we trimmed the restrained model by removing the cross-lagged regressions that were nonsignificant to create a parsimonious model and tested that against the restrained model (ΔΧ²(8) = 15.852, p = .000). The series of ΔΧ² tests indicated that the model that fitted best to the data was the restrained model (Χ²(723) = 1478.407, p = .000; RMSEA = .040; CFI = .915; SRMR = .065). This model had two couples of cross-lagged regressions with one regression being significant and the other nonsignificant within each couple (β_{Vision1→Info2}= 0.131, p = .000; β_{Vision2→Info3}= −0.030, p = .573; β_{Goal1→Vision2}= 0.175, p = .012; β_{Goal2→Vision3}= 0.070, p = .350). Notably, the mean of goal interdependence was marginally significant in this model (μ_goal1 = −0.223, p = .060). As such, this procedure also yielded a model with four cross-lagged regressions, but only between two constructs, and two of those regressions were nonsignificant. Because of the unclarity of the results of this procedure, we preferred to present and discuss the results from the parsimonious model that was directly derived from the full model.

Additional information

Funding

This work was supported by the NWO Programming Council for Educational Research (PROO) under Grant number 411-07-302.