Related Experiment Videos
Estimation of error and bias in Bayesian Monte Carlo decision analysis using the bootstrap
C D Linville1, B F Hobbs, B N Venkatesh
1Department of Computer Science and Information Systems, American University, Washington, DC, USA.
Summary
Bayesian Monte Carlo (BMC) decision analysis outputs can be affected by sample error. The bootstrap procedure can estimate this error and bias in BMC analyses, as shown in a Lake Erie case study.
Area of Science:
- Decision Analysis
- Environmental Management
- Statistical Modeling
Background:
- Bayesian Monte Carlo (BMC) decision analysis uses sampling to estimate outcomes and strategy performance.
- BMC outputs, including expected performance and value of information, are susceptible to sample error.
Purpose of the Study:
- To assess the impact of sample error on Bayesian Monte Carlo decision analysis.
- To evaluate the effectiveness of the bootstrap procedure in quantifying BMC analysis uncertainty.
Main Methods:
- The bootstrap procedure was employed, involving resampling from the original BMC sample.
- Decision analysis was re-performed on resampled data to generate output distributions.
- Case studies included a value-of-information calculation and a Lake Erie control structure analysis.
Main Results:
- The bootstrap procedure successfully estimated standard errors and biases of BMC outputs.
- BMC decision analysis outputs demonstrated significant levels of sample error.
- Bias was also identified as a notable issue in BMC analysis outputs.
Conclusions:
- The bootstrap method is a viable approach for quantifying sample error in BMC decision analysis.
- Results highlight the potential for substantial sample error and bias in BMC-derived estimates.
- Careful consideration of sample error is crucial when interpreting BMC analysis results.