What are statistical non-detects and how should they be treated in environmental data analyses?

• A. Values reported as below detection or quantitation limits and can be treated via substitution (LOD/2), maximum likelihood, regression on censored data, or specialized software.
• B. They are random outliers that should be discarded.
• C. They are data entries that should be replaced with the mean.
• D. They indicate a need to rerun the entire sampling program.

Answer: (A)

Statistical non-detects are measurements reported as below the instrument’s detection or quantitation limit. In environmental data, that means the true concentration is somewhere between zero and the detection limit; the exact value isn’t known, it’s left-censored.

Treating them this way is important because they carry information about the distribution of concentrations. Substituting each non-detect with a single number (like the detection limit or its half) or discarding them can bias estimates of averages, variability, and relationships. More robust approaches use the fact that these values are below a threshold: maximum likelihood estimation for left-censored data, regression methods that account for censoring, or specialized software designed for censored data. These methods use all available information and provide less biased estimates and better uncertainty quantification.

So non-detects aren’t random outliers to drop, nor should they be replaced with the mean or treated as evidence to rerun sampling by default. They are censored data that should be analyzed with methods that incorporate the detection limit.