Data Analysis (est-data-analysis)
ES&T reviewers scrutinize the analytical chain: blanks, recoveries, detection limits, replicates, and
whether the numbers add up. This skill covers execution and reporting; design decisions live in
est-study-design, and deposit/reproducibility in est-reporting-and-reproducibility.
When to trigger
- Reducing raw instrument/field data into results
- Building the results section and the QA/QC reporting
- A reviewer asked about detection limits, recoveries, replicates, or statistics
- Closing a mass/energy balance or fitting kinetics/dose–response
Analysis norms ES&T expects
- Report QA/QC explicitly. Method/field blanks, matrix-spike recoveries, CRM results, LOD/LOQ,
calibration range and R², surrogate/internal-standard recoveries, and how non-detects were handled.
- Honest uncertainty. Report replicates with measures of dispersion (SD/SE/CI), not single
values; propagate uncertainty through derived quantities; state n every time.
- Right statistics for environmental data. Handle left-censored (below-LOD) data correctly
(e.g., substitution caveats, MLE/ROS, Kaplan–Meier); check distributional assumptions; use
nonparametric or transformed analyses for skewed/heteroscedastic data; correct for multiple
comparisons.
- Mass / energy balance. Account for products, sorbed and volatilized fractions, and losses;
report closure (%) and explain gaps.
- Kinetics / dose–response. Report rate constants/half-lives with CIs and goodness of fit;
EC/IC/LC values with confidence bounds; state the model fitted.
- Effect size & significance. Give magnitudes and intervals, not p-values alone; relate results
to environmentally meaningful thresholds.
Reproducibility while you work (not at the end)
- A master script/workflow regenerates every figure, table, and SI exhibit from processed data.
- Set and report seeds for any stochastic step (bootstrap, Monte Carlo, simulation).
- Pin software/package versions; record instrument settings and integration parameters.
- Keep figure/table numbers matched to script outputs (see
est-reporting-and-reproducibility).
QA/QC reporting table reviewers expect to see
ES&T referees often work down a mental analytical checklist. Reporting each item pre-empts the most
common "rigor not established" objection. The minimum set, with what a reviewer reads when it is
absent:
| Element |
What to report |
If omitted, the reviewer assumes |
| Method/field blanks |
blank levels vs. sample levels; subtraction approach |
contamination is uncontrolled |
| Recoveries |
matrix-spike % and CRM agreement |
quantitation is biased/unknown |
| LOD/LOQ |
derivation (e.g., 3σ/10σ or S/N) and per-analyte values |
"detections" may be noise |
| Calibration |
range, R², whether samples fall in range |
extrapolation beyond standards |
| Surrogate/IS recoveries |
per-sample recovery correction |
run-to-run drift hidden |
| Non-detects |
censoring method (ROS/MLE/KM), not bare substitution |
summary statistics distorted |
Worked micro-example (illustrative — PFAS quantitation with censored data)
A river-PFAS dataset (illustrative numbers) shows how the rules combine into a reportable result:
- 24 samples, triplicate injection; LOQ for PFHxA = 0.5 ng/L (illustrative, derived at 10×S/N).
- Matrix-spike recovery 92% (RSD 7%, n=6); field blank < LOQ; surrogate-corrected.
- 9 of 24 below LOQ — left-censored. Naive half-LOQ substitution would report a mean of
3.1 ng/L; regression-on-order-statistics (ROS) gives 2.4 ng/L (illustrative), because substitution
inflated the low tail. Report the ROS mean with its CI and state the method.
- Reported result: "PFHxA = 2.4 ng/L (95% CI 1.7–3.3, n=24, 38% < LOQ; ROS), recovery 92±7%." That
single line carries magnitude, uncertainty, n, censoring handling, and recovery — the form a
reviewer can sign off without a query.
Referee-pushback patterns and the venue-specific fix
- "Detection limits and QA/QC are not reported." → Add the blank/recovery/LOD/LOQ table to the SI
and cite it from Methods; never leave it implicit.
- "Below-detect data handled by substitution." → Re-analyze with ROS/MLE/Kaplan–Meier; show the
result is robust to the censoring choice.
- "The mass balance does not close." → Report closure %, name the unaccounted fraction (sorbed,
volatilized, mineralized), and bound it rather than ignoring the gap.
Anti-patterns
- Results with no blanks, recoveries, or detection limits reported
- Single measurements with no replication or dispersion
- Naive zero/half-LOD substitution for heavily censored data with no caveat
- A transformation/treatment study whose mass balance never closes (or is never reported)
- p-values without effect sizes or environmental thresholds
- Over-fitting kinetics/dose–response with too few points
Output format
【Main result】magnitude + uncertainty (n, SD/CI) + units
【QA/QC】blanks / recoveries / CRM / LOD-LOQ / calibration reported? [Y/N]
【Censored data】handled how
【Mass/energy balance】closure % + explanation
【Statistics】appropriate test + assumptions checked? [Y/N]
【Reproducible】master script + seeds + pinned versions? [Y/N]
【Next】est-figures-and-tables
Supplementary resources
Source: brycewang-stanford/Awesome-Journal-Skills → Environmental-Science-and-Technology-Skills/skills/est-data-analysis/SKILL.md
1---2name: est-data-analysis3description: Use when executing and reporting the analysis for an Environmental Science & Technology (ES&T) manuscript so it survives expert review — analytical QA/QC, honest uncertainty, statistics appropriate to environmental data, and closed mass balances. It guides analysis and reporting norms; it does not fabricate results.4---567# Data Analysis (est-data-analysis)89ES&T reviewers scrutinize the analytical chain: blanks, recoveries, detection limits, replicates, and10whether the numbers add up. This skill covers execution and reporting; design decisions live in11`est-study-design`, and deposit/reproducibility in `est-reporting-and-reproducibility`.1213## When to trigger1415- Reducing raw instrument/field data into results16- Building the results section and the QA/QC reporting17- A reviewer asked about detection limits, recoveries, replicates, or statistics18- Closing a mass/energy balance or fitting kinetics/dose–response1920## Analysis norms ES&T expects21221. **Report QA/QC explicitly.** Method/field blanks, matrix-spike recoveries, CRM results, **LOD/LOQ**,23 calibration range and R², surrogate/internal-standard recoveries, and how non-detects were handled.242. **Honest uncertainty.** Report replicates with measures of dispersion (SD/SE/CI), not single25 values; propagate uncertainty through derived quantities; state n every time.263. **Right statistics for environmental data.** Handle left-censored (below-LOD) data correctly27 (e.g., substitution caveats, MLE/ROS, Kaplan–Meier); check distributional assumptions; use28 nonparametric or transformed analyses for skewed/heteroscedastic data; correct for multiple29 comparisons.304. **Mass / energy balance.** Account for products, sorbed and volatilized fractions, and losses;31 report closure (%) and explain gaps.325. **Kinetics / dose–response.** Report rate constants/half-lives with CIs and goodness of fit;33 EC/IC/LC values with confidence bounds; state the model fitted.346. **Effect size & significance.** Give magnitudes and intervals, not p-values alone; relate results35 to environmentally meaningful thresholds.3637## Reproducibility while you work (not at the end)3839- A **master script/workflow** regenerates every figure, table, and SI exhibit from processed data.40- **Set and report seeds** for any stochastic step (bootstrap, Monte Carlo, simulation).41- Pin software/package versions; record instrument settings and integration parameters.42- Keep figure/table numbers matched to script outputs (see `est-reporting-and-reproducibility`).4344## QA/QC reporting table reviewers expect to see4546ES&T referees often work down a mental analytical checklist. Reporting each item pre-empts the most47common "rigor not established" objection. The minimum set, with what a reviewer reads when it is48absent:4950| Element | What to report | If omitted, the reviewer assumes |51|---------|----------------|----------------------------------|52| Method/field blanks | blank levels vs. sample levels; subtraction approach | contamination is uncontrolled |53| Recoveries | matrix-spike % and CRM agreement | quantitation is biased/unknown |54| LOD/LOQ | derivation (e.g., 3σ/10σ or S/N) and per-analyte values | "detections" may be noise |55| Calibration | range, R², whether samples fall in range | extrapolation beyond standards |56| Surrogate/IS recoveries | per-sample recovery correction | run-to-run drift hidden |57| Non-detects | censoring method (ROS/MLE/KM), not bare substitution | summary statistics distorted |5859## Worked micro-example (illustrative — PFAS quantitation with censored data)6061A river-PFAS dataset (illustrative numbers) shows how the rules combine into a reportable result:6263- 24 samples, triplicate injection; LOQ for PFHxA = 0.5 ng/L (illustrative, derived at 10×S/N).64- Matrix-spike recovery 92% (RSD 7%, n=6); field blank < LOQ; surrogate-corrected.65- 9 of 24 below LOQ — **left-censored**. Naive half-LOQ substitution would report a mean of66 3.1 ng/L; regression-on-order-statistics (ROS) gives 2.4 ng/L (illustrative), because substitution67 inflated the low tail. Report the ROS mean with its CI and *state the method*.68- Reported result: "PFHxA = 2.4 ng/L (95% CI 1.7–3.3, n=24, 38% < LOQ; ROS), recovery 92±7%." That69 single line carries magnitude, uncertainty, n, censoring handling, and recovery — the form a70 reviewer can sign off without a query.7172## Referee-pushback patterns and the venue-specific fix7374- *"Detection limits and QA/QC are not reported."* → Add the blank/recovery/LOD/LOQ table to the SI75 and cite it from Methods; never leave it implicit.76- *"Below-detect data handled by substitution."* → Re-analyze with ROS/MLE/Kaplan–Meier; show the77 result is robust to the censoring choice.78- *"The mass balance does not close."* → Report closure %, name the unaccounted fraction (sorbed,79 volatilized, mineralized), and bound it rather than ignoring the gap.8081## Anti-patterns8283- Results with no blanks, recoveries, or detection limits reported84- Single measurements with no replication or dispersion85- Naive zero/half-LOD substitution for heavily censored data with no caveat86- A transformation/treatment study whose mass balance never closes (or is never reported)87- p-values without effect sizes or environmental thresholds88- Over-fitting kinetics/dose–response with too few points8990## Output format9192```93【Main result】magnitude + uncertainty (n, SD/CI) + units94【QA/QC】blanks / recoveries / CRM / LOD-LOQ / calibration reported? [Y/N]95【Censored data】handled how96【Mass/energy balance】closure % + explanation97【Statistics】appropriate test + assumptions checked? [Y/N]98【Reproducible】master script + seeds + pinned versions? [Y/N]99【Next】est-figures-and-tables100```101102## Supplementary resources103104- [`../../resources/external_tools.md`](../../resources/external_tools.md) — analytical QA/QC, statistics, and modeling packages105- [`../../resources/official-source-map.md`](../../resources/official-source-map.md) — data-availability and reproducibility expectations106107---108109**Source:** [`brycewang-stanford/Awesome-Journal-Skills`](https://github.com/brycewang-stanford/Awesome-Journal-Skills) → `Environmental-Science-and-Technology-Skills/skills/est-data-analysis/SKILL.md`
thedixitjain