Lessons Learned from Causal Analysis from Army Project Data
Abstract
Why Model Causal Structure. Depending on causal structure, factor loadings may or may not be identifiable by conventional adjustments. Bias can be introduced by: Failure to adjust for Common Causes (Confounders); Adjusting on a Common Outcome (Colliders); Common sources of measurement error; Treatment confounder feedback. Therefore, causal structural assumptions are necessary to: Correct (adjustment) for bias; Interpreting covariate loadings in regression models (anova and ancova); Identify appropriate analysis methods (e.g. stratification, g-methods, and so forth).
Document Details
- Document Type
- Technical Report
- Publication Date
- Jan 01, 2020
- Accession Number
- AD1090438
Entities
Organizations
- Carnegie Mellon University