Skip Navigation
U.S. flag

An official website of the United States government

Dot gov

The .gov means it’s official.
Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

Https

The site is secure.
The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

Internet Explorer is no longer a supported browser.

This website may not display properly with Internet Explorer. For the best experience, please use a more recent browser such as the latest versions of Google Chrome, Microsoft Edge, and/or Mozilla Firefox. Thank you.

Your Environment. Your Health.

Comparison of Results Across Statistical Methods Applied to Simulated Datasets

Table 1. Statistical methods applied in the workshop using one or more of the datasets. Some methods were used by more than one research group, representing different abstracts (11MB).

Abstract # Method(s)
1 Visualization, Structural Equation Models (SEM), and Principal Components Analysis (PCA)
2 Building Bayesian Networks
3 Novel approach + Least Angle Regression (LARS)
4 Bayesian Kernel Machine Regression (BKMR)
5 Machine Learning
6 Two step variable selection and Least Absolute Shrinkage and Selection Operator (LASSO)
7 Bayesian Estimation of Weighted Sum
8 Two step shrinkage-based regression
9 Informed sparse PCA + segmented regression
10 Bayesian g-formula
11 PCA (a) and Classification and Regression Tree (CART, b, evaluated separately)
12 PCA
13a Single Chemical Analysis
13b Multiple Regression
13c PCA
14 Bayesian profile regression
15 Conformal predictions
16 Shrinkage methods (LASSO/LARS)
17 Modes of action (results presented for Z=0 strata)
18 PCA
19 Random Forest
20 Feasible Solution Algorithm (FSA)
21 Factor Mixture Models
22 CART
23 Subset and bootstrap
24 Exposure Space Smoothing (ESS)
25a Variable Selection Regression (VSR)
25b Multivariate Adaptive Regression Splines (MARS)
26 Bayesian non-parametric regression
27 Weighted quantile sum regression (WQS)
28 LASSO
29 Exploratory data analysis (EDA)
30 Weighted Quantile Sum (WQS) Regression
31 Bayesian Additive Regression Trees (BART) and Non-negative Sparse Principal Component Analysis (NSPCA) (PC1)
32 PCA
33 ESS

Comparable results for the dataset of  N=500 individuals, 7 chemical exposures

Figure 1. Results for exposure-outcome associations across the statistical methods applied to Simulated Dataset 1. Results were most comparable in this simple dataset of N=500 individuals, 7 chemical exposures, and one potential confounder, relative to the other more complex datasets used for the workshop method applications. Abstracts/methods in Table 1 with insufficient data or qualitative metrics are not displayed


Dataset of N=500 individuals, 14 exposures, and three potential cofounders
Figure 2. Results for exposure-outcome associations across the statistical methods applied to Simulated Dataset 2. Relative to Simulated Dataset 1, greater variability in results was observed in this slightly more complex dataset of N=500 individuals, 14 exposures, and three potential confounders. Abstracts/methods in Table 1 with insufficient data or qualitative metrics are not displayed.

Notes: Results for the real world dataset are not shown. Figures generated by Dr. Bonnie Joubert at the NIEHS based on information from the publicly available NIEHS Epi Stats Workshop Final Abstracts (11MB). Replication of and expansion of results using larger and more detailed datasets is encouraged. Corrections or questions can be addressed to Bonnie.

Back
to Top