|
Event History Analysis: Modeling Occurrences of Events Over Time
|
| Presenter(s):
|
| Blair Stephenson, Los Alamos National Laboratory, blairs@lanl.gov
|
| Christine Starr, Los Alamos National Laboratory, cstarr@lanl.gov
|
| Melissa Schaum-Nguyen, Los Alamos National Laboratory, mschaum@lanl.gov
|
| Abstract:
The present study demonstrates the use of Event History Analysis (EHA). Originally developed in the biostatistics arena (as survival analysis), EHA offers a viable methodology for understanding both the timing and etiology of qualitative outcomes in the social sciences. We demonstrate the use of a related set of techniques in the context of modeling the voluntary (non-retirement) attrition of scientists over two decades. In contrast to traditional methods (e.g., logistic regression), EHA models time to the event of interest, considers information from censored observations, and allows for the inclusion of both invariant (e.g., gender) and time varying covariates (e.g., salary). We discuss basic dataset design, exploratory techniques, popular approaches such as (Cox regression), along with assumptions and alternatives such as Competing Risks Regression (CRR), which allows for an accounting of multiple possible outcomes in competition with the primary outcome of interest.
|
|
The Use of Piecewise Growth Models to Estimate a Staggered Interrupted Time Series
|
| Presenter(s):
|
| Keith Zvoch, University of Oregon, kzvoch@uoregon.edu
|
| Joseph Stevens, University of Oregon, stevensj@uoregon.edu
|
| Drew Braun, Bethel School District, dbraun@bethel.k12.or.us
|
| Abstract:
The proposed paper describes the use of piecewise growth models as a means for estimating intervention outcomes associated with a complex interrupted time series (ITS) design. The demonstration utilizes literacy data obtained on elementary school students in the Pacific Northwest. During the course of one school year, weekly literacy assessments were administered and supplemental instructional interventions were delivered to students as a means to facilitate the attainment of literacy benchmark goals. However, the timing of treatment was not constant as the onset and duration of particular instructional supplements were purposely differentiated by student. To illustrate the challenges and opportunities associated with the evaluation of staggered ITS designs, a series of multilevel growth models are presented. The demonstration shows that multilevel modeling techniques provide a flexible and powerful approach for capturing the complex structure of individualized treatment regimes while simultaneously documenting the immediate and more distal responses to intervention.
|
|
The Z-Kids: What Happens to Individual Clients Over Time? Outcomes With Clinical, Program, and Evaluation Salience
|
| Presenter(s):
|
| Richard Wood, Pima Prevention Partnership, rwood@thepartnership.us
|
| Judith Francis, Pima Prevention Partnership, jfrancis@thepartnership.us
|
| Abstract:
This paper describes the use of a single subject design (SSD n=1) for measuring individual client outcomes over time using the ipsative Z test developed by Mauser, Yarnold and Foy (1991) for autocorrelated data. This measure was tested using Global Appraisal of Individual Need(GAIN) data for 613 youth in outpatient drug treatment. The study examined changes over time (baseline, 3,6,12 months) for substance use, self reported criminal behavior and emotional problems. The result was identification of individual clients who significantly improved, significantly deteriorated, or displayed no significant change over time for these GAIN outcomes. This approach yields two advantages. First it can be used to test causation between an intervention and individual client change. Second it provides clinicians with information during an intervention that can be used to modify programs to better serve client needs. The paper includes the SPSS syntax used to compute the ipsative Z score per individuals.
|
| | |