High-Traffic Format · Technical Perspective · Published: October 27, 2026
Table of Contents
Key Takeaways
- The mean is highly sensitive to extreme values. Skew and outliers make it misleading — and water quality data during an upset is exactly that shape [1].
- A sensor is not a truth machine. Field instruments get checked against standard solutions of known value, and a probe that reads a known pH 7 buffer as 5 or 8.5 needs recalibration before its data means anything [3].
- Drift is a recognized, ongoing operational problem. Even large utilities report that pH and residual chlorine sensors move over time and that slow drift is hard for operators to catch in time [4].
- Two variables can correlate purely because of a third, unseen factor. The textbook case is Florida’s Alzheimer’s statistics: retirement-age migration, not water, produces the pattern [2].
- Data quality is not one property but several measurable dimensions — accuracy, completeness, consistency, timeliness, validity. More data fixes none of them [5].
- Most of these myths have a procedural fix, not a technical one: record uncertainty, keep maintenance flags attached to the data, and treat every dataset as fit for a specific purpose rather than trustworthy by default.
Water treatment runs on numbers, and numbers feel objective in a way words never do. That feeling is why bad data habits survive for years in plants that are otherwise rigorous. The five myths below are the ones most likely to bend real decisions — chemical budgets, maintenance schedules, compliance strategy, capital plans. Each sounds reasonable. Each fails in a specific, preventable way.
Why Data Myths Survive in Water Operations
Instrumentation multiplied faster than data literacy. Ten years ago a small plant had a handful of gauges. Today it may push dozens of parameters into a historian at high frequency. The measurement layer grew; the reading habits largely did not.
The consequences are also asymmetric. Over-dosing chemicals wastes money quietly. Under-dosing puts you in front of a regulator loudly. Both failures trace back to the same handful of misunderstandings about what data can and cannot tell you. Take them one at a time.
Myth 1: “More Data Is Always Better”
The belief: if some data improves decisions, more data — higher frequency, more parameters, more logs — must improve them further.
Why it fails: volume and decision value are different axes. A second parameter sampled every ten seconds does nothing for a decision that hinges on a daily mass balance. A third decimal place does nothing when the sensor’s uncertainty is ±2%. What governs usefulness is fitness for purpose, and data quality practice breaks that into measurable dimensions — accuracy, completeness, consistency, timeliness, validity — any one of which can be strong while the dataset as a whole still fails your specific decision [5].
And high-frequency logging of a low-value parameter is not merely neutral. It buries the signals that matter, fills storage with trivia, and stretches review attention across charts nobody has time to open.
Better practice: inventory your decisions first. What do you adjust hourly (dosing)? What do you review weekly (trends)? What do you report monthly (compliance)? Assign sampling frequency and parameters to those decisions, not to whatever the hardware can emit. Then aggregate aggressively: minute data exists to build trustworthy hourly summaries, and hourly summaries exist to build trustworthy months.
Myth 2: “The Average Says Everything Is Normal”
The belief: if the monthly mean sits inside the specification band, the month was fine.
Why it fails: the mean is a poor ambassador for skewed, outlier-ridden data. Statistical guidance is explicit — the mean is sensitive to extreme values, which distorts it and makes it misleading when distributions are skewed or outliers are present [1]. Water quality data has exactly that shape. Most readings cluster near the target, and a few excursion events — a storm, a breakthrough, a dosing failure — sit far out in the tail.
Those tail events are the ones that trigger violations, foul membranes, or poison a control model. A comfortable average will never surface them. Two months can share an identical mean and be nothing alike: one stable mid-band, the other flirting with the limit every Friday night.
Better practice: pair every average with its spread. Report percentiles (95th and 99th are the compliance-relevant ones for most permits), maximum excursions, and time above threshold. When the median and the mean disagree by a meaningful margin, believe the gap — it usually points at a tail event somebody should go investigate [1].
Myth 3: “The Sensor Reading Is the True Value”
The belief: the dashboard shows 7.24 mg/L, so the water contains 7.24 mg/L.
Why it fails: every reading is an estimate from a physical device with finite accuracy, exposed to fouling, aging, temperature, and whatever its last calibration did. Standard field practice is to verify instruments against standard solutions of known value. Put a probe in a known pH 7 standard and watch it read 5 or 8.5 — it must be recalibrated before its output can be trusted at all [3].
Between checks, sensors move. Drift is central enough to operations that one of the world’s major water utilities has run open innovation challenges specifically on identifying drift and inaccuracy in sensor readings, noting that sensors for parameters like pH and total residual chlorine drift over time and that slow drift is difficult for operators to detect in a timely way [4].
The operational consequence is subtle. A drifting sensor does not produce obviously wrong numbers. It produces smoothly plausible ones. Decisions calibrated to a drifting baseline — dosing tuned, limits interpreted, models trained — inherit the bias without anyone noticing.
Better practice: read every value as “true value plus uncertainty.” Keep calibration events and verification results as metadata attached to the data record, not in a separate binder. Benchmark online values against lab results on a fixed schedule, and when the gap exceeds your acceptance band, correct the instrument and the interpretation of the historical record — not the chemistry.
Myth 4: “Correlation Proves Cause”
The belief: turbidity tracks chlorine demand, dosing tracks color, so one drives the other. Adjust accordingly.
Why it fails: two variables can move together for reasons that involve neither causing the other. A third, unseen factor — a confounder — can generate the entire relationship. The classic illustration from environmental statistics: Florida has high rates of Alzheimer’s disease, and nothing in Florida’s water causes it. Retirees move to Florida, age drives the disease, and the association is confounded top to bottom [2].
Water plants are full of the same trap. Season drives raw water temperature and demand patterns together, so any parameter correlated with temperature may simply be riding the calendar. Plant flow moves with production schedules that also shift contaminant loads. A maintenance event changes two traces at once, and the correlation afterwards is bookkeeping, not chemistry.
Better practice: before acting on a correlation, ask what else changed at the same time, and look for the common driver. Where money is attached — chemical reduction programs, control logic changes — verify the causal story with a deliberate test: change one input under controlled conditions and watch the response. Correlation earns you a hypothesis. Only intervention earns you a budget line.
Myth 5: “A Stable History Means a Safe Future”
The belief: the process has been stable for two years, so the limits, dosing curves, and assumptions built on that period still hold.
Why it fails: stability is a snapshot property, and the things that break it rarely announce themselves. Slow sensor drift accumulates until the baseline itself is fiction [4]. Upstream catchments change — new industrial discharges, seasonal algal dynamics, distribution network modifications. Resin and membrane assets age. A model or operating envelope tuned on last year’s stable data can be confidently wrong about next month’s water.
The dataset that was accurate, complete, and timely for its original purpose can lose every one of those qualities without anyone sending a memo. That is why data quality frameworks treat quality as fitness for a purpose, to be re-assessed, not a certificate issued once [5].
Better practice: schedule revalidation. Re-verify calibration against independent references on a fixed cadence, re-fit baselines annually, and re-run limit reviews whenever raw water sources, loads, or equipment change. Stability is evidence, not a guarantee — and its shelf life is shorter than most plants assume.
The Myth-Check Table
| Myth | Why it fails | Better practice |
|---|---|---|
| More data is always better | Volume ≠ decision value; low-value data buries key signals [5] | Assign sampling rate to decisions; aggregate aggressively |
| The average means normal | Mean is distorted by outliers and skew — common in water data [1] | Report percentiles, excursions, time-above-threshold alongside means |
| Sensor reading = true value | Instruments need verification against known standards; drift is ongoing [3][4] | Keep calibration metadata with data; schedule lab cross-checks |
| Correlation = cause | A hidden third factor can generate the whole pattern [2] | Hunt the confounder; verify with one-variable-at-a-time tests |
| Stable history = safe future | Drift and upstream change silently invalidate old baselines [4] | Fixed-cadence revalidation; re-fit limits after any major change |
A Data Quality Baseline Worth Adopting
If the plant adopts one thing from this article, make it a formal data quality baseline. The dimensions are standard, measurable, and cheap to audit: accuracy (does the value represent reality?), completeness (are the expected values there?), consistency (do systems agree?), timeliness (is it current?), validity (does it fall inside defined ranges?) [5].
Reviewing your two or three most decision-critical parameters against those five questions once a quarter takes under an hour and kills most of the errors above. It also produces the traceable, well-annotated record set that makes an audit uneventful — documentation you can hand to an inspector without a week of preparation.
What This Means for Instrument Buying
The myths carry a procurement lesson too. A cheaper sensor that drifts without detection is not cheaper. It is a slow-acting contamination of every decision downstream, which is the opposite of what you bought.
The honest frame is total cost across the instrument’s life: acquisition, calibration labor, verification failures avoided, and the decision quality your team gets out of it. Vendors who publish accuracy specifications, calibration intervals, and uncertainty data openly are telling you how much work they expect you to do. Vendors who do not are asking you to fund the myths.
And when instrumentation already reports standardized digital outputs with status and diagnostics attached, the baseline above becomes nearly free to maintain. The plant stops arguing about whose number is right and starts arguing, productively, about what to do next.
References
- PubMed Central (PMC12401145) — Beyond the average: why data spread matters in clinical studies — peer-reviewed explanation that the mean is sensitive to extreme values and becomes misleading when distributions are skewed or outliers are present. https://pmc.ncbi.nlm.nih.gov/articles/PMC12401145/
- American Council on Science and Health — ACSH Explains “Confounding”: Why Correlation Does Not Mean Causation — the Florida/Alzheimer’s worked example of a confounding variable producing a spurious association. https://www.acsh.org/news/2017/10/18/acsh-explains-confounding-why-correlation-does-not-mean-causation-11981
- IOP Science, Measurement Science and Technology — Measuring imperfections of water quality sensors in water distribution networks — describes verification of field sensors against standard solutions, including recalibration when a sensor misreads a known pH 7 standard. https://iopscience.iop.org/article/10.1088/1361-6501/ab1eeb
- PUB, Singapore’s National Water Agency — Sensor Data Integrity Monitoring — challenge statement documenting that pH and total residual chlorine sensors drift over time and that slow drift is hard for operators to detect in a timely way. https://www.pub.gov.sg/Industry/Events%20and%20Programmes/Global%20Innovation%20Challenge/Challenges/2020/Sensor%20Data%20Integrity%20Monitoring
- Actian — What is Data Quality? — glossary entry defining data quality as fit for intended use across the dimensions of accuracy, completeness, consistency, timeliness, and validity. https://www.actian.com/glossary/data-quality/
- Shanghai ChiMay Tech ltd. — About us — company overview of Shanghai ChiMay’s online water quality analyzer product line referenced in this article. https://www.chimaytech.net/it?p=520
About the author: Written by the Shanghai ChiMay Technical Editorial Team — the editorial group supporting Shanghai ChiMay’s online water quality analyzer product line, with working knowledge of sensor calibration practice, data integrity workflows, and plant-side monitoring programs.
