Meta-Analysis and Benefit Transfer at Different Levels of Aggregation: Comparing Group-Averaged and Individual-Level Models Using Hierarchical Bayesian MethodsEPA Grant Number: R833460
Title: Meta-Analysis and Benefit Transfer at Different Levels of Aggregation: Comparing Group-Averaged and Individual-Level Models Using Hierarchical Bayesian Methods
Investigators: Moeltner, Klaus , Johnston, Robert J. , Rosenberger, Randall S.
Institution: Lehigh University
EPA Project Officer: Hahn, Intaek
Project Period: March 1, 2007 through August 31, 2008
Project Amount: $198,675
RFA: Methodological Advances in Benefit Transfer Methods (2006) RFA Text | Recipients Lists
Research Category: Environmental Justice
Under pronounced individual heterogeneity in preferences for environmental quality or individual-specific components in policy baseline conditions and changes, benefit transfer (BT) predictions based on aggregate data can be inaccurate and invalid. Standard applications of BT methods, especially those flowing from meta-regression models (MRMs), rely on highly aggregated information on people’s preferences and values, as these are typically reported as averages over individuals in original studies. This project aims to reduce the resulting uncertainties and potential biases affecting BTs through methods that enrich and disaggregate data underlying MRMs. Specifically, we will examine two types of data enrichment for MRMs: Partial enrichment, i.e the use of raw data from original studies to augment the set of regressors in aggregate-level MRMs, and full enrichment, i.e. the use of raw data and possibly re-computation of original models to derive individual-level welfare measures for original studies. This leads to an individual-level MRM. We will compare these enriched models to the basic MRM using multiple performance criteria such as the accuracy and efficiency of in-sample predictions, and the efficiency and variability of out-of-sample forecasts for a variety of hypothetical BT contexts. Throughout this analysis we will document and contrast both the improvements and costs associated with various approaches to MRM enrichment, and the applicability of these methods to typical BT contexts.
Aggregate-level and raw data will be gathered from existing primary studies on hiking and sports-fishing, with focus on the impact of changes in site access and quality. A novel multi-layer hierarchical Bayesian estimation framework will be used to efficiently accommodate the different levels of aggregation in the meta-data. BT results will be simulated for sites that differ in population characteristics as well as resource quality baselines and changes, mirroring the situation in real-world policy applications.
We anticipate that the enriched MRMs will offer significant, cost-effective improvements in both the efficiency and reliability of functional BT in typical valuation contexts. Moreover, if individual heterogeneity is pronounced in original studies, the individual-level model will generate the most reliable confidence intervals for BT estimates. Overall, this project will illustrate novel yet practical means to capitalize on existing data to improve BT, applicable to a wide variety of typical environmental policy scenarios. We also expect our results to lend rigorous and strong support to recent calls for consistency in data reporting, improvement of data access, and creation of data repositories to facilitate environmental valuation.