EDA Studio

Complete Exposure Decision Analysis suite · IHSTAT + Bayesian + Expostats Tools 1–4 · IHMOD · SkinPerm · SIHM · control banding

Frequentist core validated against AIHA IHSTAT v8
Bayesian + suite engines aligned with Expostats / AIHA tools

Exposure Decision Analysis Report

EDA Studio · IHSTAT-parity statistics + Bayesian Decision Analysis · AIHA 5-category framework · generated 6/25/2026, 1:10:46 AM

OEL 5 mg/m³ · n = 15

Key decision statistics

95th percentile (X₀.₉₅)	4.84 mg/m³ (97% of OEL)	UTL₉₅,₉₅ (K=2.566)	7.05 mg/m³
Exceedance fraction	4.24% (95% CrI 0.58–18.23%)	MVUE arithmetic mean	2.68 mg/m³
Land 95% LCL / UCL on AM	2.20 / 3.56	Shapiro–Wilk W (log)	0.974 → lognormal? Yes

Descriptive statistics

n	15	GM	2.48 mg/m³	GSD	1.50
Min / Max	1.20 / 5.50	Mean	2.68 mg/m³	Median	2.50
Mean ln(x)	0.9079	SD ln(x)	0.4071	% > OEL (obs)	6.7%

Bayesian Decision Analysis (50,000 posterior draws)

Category	Cat 0 <1%	Cat 1 1–10%	Cat 2 10–50%	Cat 3 50–100%	Cat 4 >100%
Posterior probability	0.0%	0.0%	0.0%	54.0%	46.0%

X₀.₉₅ posterior median 4.91 (95% CrI 3.76–7.70) · P(F>5%) 46.0% · P(X₀.₉₅>OEL) 46.0%

Methods

Descriptive and lognormal parametric statistics replicate AIHA IHSTAT v8 (one-sided 95/95 Natrella tolerance factor; Finney MVUE; Royston Shapiro–Wilk). Land's limits use the generalized-confidence-interval pivot. Bayesian Decision Analysis samples the (µ,σ) posterior under a non-informative Jeffreys prior, reporting probabilities across the AIHA exposure-control categories (Expostats Tool 1 methodology). Certainty follows the Hewett rule-of-thumb. This report is a decision aid; professional judgement and SEG validity remain the assessor's responsibility.

EXPOSURE DATA

How to use this tool

Name the SEG, then paste your sampling results in the box — one number per measurement, any separator.
For non-detects, type the limit with a '<' prefix (e.g. <10); pick how to handle them below (MLE is recommended).
Enter the OEL and unit. Results update live — no 'run' button.
Tick 'Grouped by worker' to test SEG homogeneity; tick 'informed prior' to combine judgement with the data.
Use Download CSV / Print PDF (top right) to save the result.

Reading the result: The big number is the AIHA category (0–4) from the 95th-percentile vs OEL, with a certainty rating. Cat 3–4 = act; Cat 0–1 = acceptable; in between = confirm. The Bayesian chart shows how sure the data make that call.

SEG / data descriptionGrouped by worker — one worker per line (runs SEG homogeneity test)Samples — space/comma/newline separated. Prefix non-detects with < (e.g. <10)

OEL

Unit

Non-detect treatment Use informed Bayesian prior (IHDA / WebExpo style)

Quick load

15 values parsed

IHSTAT validation set

Category 3 — 50–100% of OEL

Greater than acceptable / uncertain — control or gather data

Decision certainty

Medium

X₀.₉₅ (95th pct)

4.84 mg/m³

97% of OEL

UTL₉₅,₉₅

7.05 mg/m³

Category 4

Exceedance fraction

4.24%

≤ 5% rule

P(exposure > OEL band)

46%

posterior, Cat 4

The exposure profile sits in the 50–100% band — close to the OEL. Treat as a priority SEG: tighten controls and collect additional samples to sharpen the decision.

BAYESIAN DECISION CHART

50,000 draws · Jeffreys

Posterior probability that the true exposure profile (X₀.₉₅ ÷ OEL) falls in each AIHA category. The decision category is the one carrying the most posterior weight.

Cat 0 < 1% of OEL

0.0%

Cat 1 1–10% of OEL

0.0%

Cat 2 10–50% of OEL

0.0%

Cat 3 50–100% of OEL

54.0%

Cat 4 > 100% of OEL

46.0%

P(F > 5%)

46.0%

P(X₀.₉₅ > OEL)

46.0%

P(AM > OEL)

0.0%

Decision weight

54.0%

DESCRIPTIVE STATISTICS

IHSTAT parity

n15

Min / Max1.20 / 5.50 mg/m³

Arithmetic mean2.68 mg/m³

Median2.50 mg/m³

Std deviation1.14

Geometric mean (GM)2.48 mg/m³

Geometric SD (GSD)1.50

Mean of ln(x)0.9079

SD of ln(x)0.4071

% samples > OEL(observed)6.7%

LOGNORMAL PARAMETRIC + FIT

Land's exact · UTL · W-test

Shapiro–Wilk W (log-data)0.9743 → lognormal? Yes

Shapiro–Wilk W (raw data)0.9040 → normal? Yes

MVUE arithmetic mean2.68 mg/m³

Land's 95% LCL / UCL on AM(GCI)2.20 / 3.56

95th percentile (X₀.₉₅)4.84 mg/m³

UTL₉₅,₉₅ (K=2.566)7.05 mg/m³

% > OEL (parametric)4.24%

% > OEL, 95% credible interval0.58 – 18.23%

BAYESIAN POSTERIOR ESTIMATES

Jeffreys prior · complete data

X₀.₉₅ posterior median4.91 mg/m³

X₀.₉₅ 95% credible interval3.76 – 7.70

Exceedance F posterior median4.58%

F 95% credible interval0.58 – 18.23%

Arithmetic mean (median)2.71 mg/m³

AM 95% upper credible limit≈ Land UCL3.36 mg/m³

GSD posterior median1.52

LOG-PROBABILITY PLOT

lognormal fit check

Points hugging the line support the lognormal assumption (W = 0.974, fit accepted at α = 0.05).

Methods. Descriptive and lognormal/normal parametric statistics replicate AIHA IHSTAT v8 (Mulhausen, Drolet, Lavoué). Tolerance factor K is the one-sided 95%/95% Natrella factor. The exceedance fraction uses the parametric estimator F = 1 − Φ((ln OEL − ln-mean)/ln-SD). Land's confidence limits on the arithmetic mean use the generalized-confidence-interval pivot (Krishnamoorthy & Mathew). The Bayesian Decision Analysis samples the posterior of (µ, σ) under a non-informative Jeffreys prior — exact conjugate draws for complete data — and reports posterior probabilities across the five AIHA exposure-control categories, the methodology of Expostats Tool 1 (Lavoué et al. 2019) and IHSTAT_Bayes. Category certainty follows the Hewett rule-of-thumb (point estimate sets the category; the upper tolerance limit sets confidence). Non-detects are handled by maximum likelihood, robust ROS (Helsel), or substitution; the Bayesian engine switches to data-augmentation Gibbs when censored. SEG homogeneity(grouped mode) uses a random-effects one-way ANOVA on log-data to split within- vs between-worker variance, with Rappaport's 2-fold criterion and the F-test for a worker effect (Expostats Tool 2).