Drawing causal inferences from a longitudinal cluster randomized experiment with crossovers: a study of a distributed leadership program in urban schools / Nianbo Dong.

Format:

Book

Manuscript

Thesis/Dissertation

Author/Creator:

Dong, Nianbo.

Contributor:

Maynard, Rebecca, advisor.

University of Pennsylvania.

Language:

English

Subjects (All):

Penn dissertations--Education.

Education--Penn dissertations.

Local Subjects:

Penn dissertations--Education.

Education--Penn dissertations.

Physical Description:

xi, 146 pages; 29 cm

Production:

2009.

Summary:

The Distributed Leadership Teacher Training Program (DLT) was implemented to improve student's learning in the United States. However, empirical evidence of the effectiveness of DLT still lacks. This dissertation investigates the effectiveness of DLT using data from the first experimental study of distributed leadership in the United States. The study focuses on the analytic approaches and statistical models to analyze this complex intervention---a longitudinal cluster randomized control trial with cross-overs and unequal probability of treatment assignment.

In 2005 the Distributed Leadership Teacher Training Program (DLT) was launched in the School District of Philadelphia funded by the Annenberg Foundation. This multifaceted professional development program for school leadership teams composed of principals and teacher leaders aims to improve student outcomes through developing effective instructional leaders within elementary and high schools under the framework of distributed leadership perspective.

The evaluation of the DLT is a longitudinal, cluster randomized experiment. In the 2006-07 school year, 19 eligible elementary schools were recruited and four schools were randomly assigned to the treatment group receiving the intervention. In the 2007-08 school year, seven new elementary schools were recruited. Two out of 22 schools were randomly assigned to the treatment group with twice the probabilities assigned to the original 15 control schools.

This dissertation applies 2-level Hierarchical Linear Modeling (HLM) with controlling for the baseline covariates to evaluate the two-year impact of DLT on school leadership teams (Instructional Leadership, and Leadership Data Use), teachers (Teaching Instruction, Interaction with Leader around Instruction, Interaction with Leader around Instruction, Deprivitized Practice/Peer Observation, and Teacher Data Use), and students (school attendance, math, and reading). Except the two-year impact on Deprivitized Practice/Peer Observation , which was significantly positive, no statistical significant results were found in any of the other outcome measures mentioned above at alpha = .05. However, the two-year impact on math was significantly negative at alpha = .10.

In additional analyses, results based on the experimental design and a second-best non-experimental design are compared to assess the estimate bias of the non-experimental analysis. No significant difference was found from the result comparison between these two approaches. In addition, the issues of statistical model selection and specification in analyzing longitudinal cluster randomized controlled trials were discussed, and the suggestions were made.

Notes:

Adviser: Rebecca Maynard.

Thesis (Ph.D. in Education) -- University of Pennsylvania, 2009.

Includes bibliographical references.