Occupational exposure analysis for Similar Exposure Groups (SEGs)
Enter your OEL value and units, then paste measurements one per line. Prefix non-detect results with < followed by the detection limit (e.g. <5.5). The tool requires at least 3 values.
Both Bayesian (using an uninformative prior on logσ) and frequentist (maximum likelihood lognormal fit) analyses are run simultaneously. The AIHA risk band classification is applied using the 95th percentile of the exposure distribution relative to the OEL. Charts update on each run.
Understanding your results
AIHA Risk Bands classify the SEG into one of four categories based on the probability that the 95th percentile of the exposure distribution exceeds the OEL. Band 0 (properly controlled) means the 95th percentile is very likely below 10% of the OEL. Band 1 (likely controlled): 95th percentile probably below the OEL. Band 2 (data suggest inadequate control): 95th percentile likely above the OEL. Band 3 (unacceptable): 95th percentile almost certainly above the OEL. The higher the band, the more urgent the need for controls review.
Lognormal assumption: occupational exposure data are typically lognormally distributed — the log-transformed values follow a normal distribution. The geometric mean (GM) and geometric standard deviation (GSD) describe this distribution. A GSD > 3 indicates high variability and may reflect mixed exposure conditions within the SEG; consider whether the SEG should be split.
Bayesian vs frequentist: both use the same underlying lognormal model. The Bayesian analysis uses an uninformative prior on logσ and yields a posterior distribution for the 95th percentile, allowing explicit probability statements (e.g. “87% probability that the 95th percentile exceeds the OEL”). The frequentist analysis uses maximum likelihood estimation and Wald confidence intervals, which are appropriate for larger datasets (≥15 samples). With fewer than 6 samples, both methods are highly uncertain — results should be interpreted cautiously.
Exceedance fraction is the estimated proportion of individual 8-hour TWA exposures in the SEG that exceed the OEL. Even Band 1 SEGs can have non-trivial exceedance fractions if the GSD is high.
Non-detects: beta substitution replaces each non-detect with its maximum likelihood estimate from the lognormal fit, iteratively. LOD/√2 and LOD/2 are simpler substitution methods. With >30% non-detects, all methods become unreliable; consider lowering the LOD or using specialist censored-data software.
Enter repeated measurements for individual workers. Each row needs a worker identifier and a measured value. Log-transformed one-way ANOVA partitions total variance into within-worker (day-to-day) and between-worker components. Requires at least 2 workers with at least 2 measurements each.
Reference: Kromhout et al. (1993) Ann. Occup. Hyg. 37:121–138. AIHA IH Statistics guidance.
Enter one row per measurement: WorkerID, value (one per line). Worker IDs can be any text. Values must be positive numbers. Prefix non-detects with < (e.g. <2.5).
| Worker | n | GM | GSD | Min | Max |
|---|
Enter measurements with up to two categorical determinants (e.g., task type, department, control measure). Log-transformed one-way ANOVA is run for each determinant to test whether categories have significantly different geometric mean exposures. The tool ranks determinants by the proportion of total log-scale variance they explain (η²).
Requires at least 2 categories per factor and at least 2 measurements per category. Reference: Tielemans et al. (1999); AIHA Exposure Assessment Strategies Committee guidelines.
| Category | n | GM | GSD | 95th %ile | % OEL |
|---|
| Category | n | GM | GSD | 95th %ile | % OEL |
|---|
η² (eta-squared) is the proportion of total log-scale variance explained by each factor. Values > 0.14 indicate large effect; 0.06–0.14 moderate; < 0.06 small (Cohen, 1988). A factor with high η² is a strong driver of exposure variability in this dataset.