Statistics for Censored Environmental Data

NADA Course Outline

Day 1

Types of Reporting Limits
detection and quantitation limits
implications for data analysis

Storing Nondetects in a Database
three standard formats

Plotting Data with Nondetects
censored boxplots
censored probability plots
scatterplots with nondetects
survival curves

Overview: The Three Approaches
ROS (probability plotting)
maximum likelihood
nonparametric methods
why not to substitute (fabricate) data

Estimating Descriptive Statistics
what's the mean of <1, 5 <10 and 17?
parametric methods
Kaplan-Meier and other nonparametric methods
why Cohen and Delta-lognormal are “old technology”

Computing Interval Estimates
how to compute confidence intervals with nondetects
UCLs for censored data
bootstrapping

Tests for paired data
paired Prentice-Wilcoxon test
paired “t-test” for censored data


Day 2

Comparing Two Groups
Generalized Wilcoxon (nonparametric) test
MLE “t-test”
what to do with multiple reporting limits

Comparing Three or More Groups
Generalized Wilcoxon (nonparametric) test
MLE version of ANOVA
familiar tests for one reporting limit

Correlation Methods
censored correlation coefficients
what to do when one or both have multiple detection limits?
Kendall’s tau
likelihood correlation coefficient

Regression and Trends
how to compute a line with nondetects?
what to do with multiple reporting limits?
Akritas-Sen-Theil line
MLE for regression
logistic regression