NRSB RMS Measurement: Statistical Analysis, Control Charts and Results
Statistical analysis helps an RMS distinguish measured variation from a process problem.
- Averages summarize data, spread describes variability, uncertainty qualifies interpretation, and a control chart tracks a quality indicator against established limits over time.
- A point outside a control limit calls for investigation under the QA plan; it is not automatically a client result or proof of one cause.
On this page12 sections
- Why statistical thinking matters to an RMS
- Mean, median and the shape of readings
- Range, spread and repeatability
- Uncertainty is not a mistake
- Control charts track process behavior
- How to read a control signal
- Compare two results without false precision
- Worked calculation: mean and range
- Avoid common statistical traps
- Put quality review into the workflow
- Study statistics through interpretation
- From analysis to a defensible conclusion
The NRSB April 2026 RMS guide places statistical principles in analysis and names measurement error, uncertainty and control charts. A specialist must distinguish a real change in the building from ordinary variation and process instability. Statistics do not make a weak measurement strong; they help describe the evidence, compare observations and identify when a measurement system needs review.
- Exam scope
- NRSB Radon Measurement Specialist (RMS)
- Primary concept
- Statistical Analysis, Control Charts and Results
- Use
- Apply current method and quality plan; verify standards and jurisdictional rules
Why statistical thinking matters to an RMS
Think of three questions: what value summarizes the observations, how variable are they, and how certain is the estimate for the stated purpose? A single reported average can hide a wide range. A detailed trace can show fluctuations but tempt over-interpretation. A statistical summary should fit the measurement design and be accompanied by enough context to interpret it.
For the exam, do not memorize a formula without knowing the question it answers. Identify whether the task is to summarize repeated readings, compare duplicate devices, estimate uncertainty or assess whether a lab process remains stable. The terms may all involve numbers, but they are not interchangeable.
Mean, median and the shape of readings
The arithmetic mean is the sum of values divided by the number of observations. It is familiar and often used to summarize repeated measurements, but extreme values can pull it. The median is the middle value after ordering the observations and is more resistant to a very high or low point. Which statistic is appropriate depends on the method and reporting rule, not personal preference.
Suppose a continuous monitor records mostly similar interval values and then one brief sharp rise. The mean may increase, while the median may remain closer to the typical interval. Neither statistic alone tells you whether the rise was an instrument artifact, a real event or an important exposure feature. Review time stamps, instrument behavior, building context and the objective before deciding how to report.
Do not replace a required arithmetic mean with a median because the result is less concerning. Follow the standard and explain the distribution if it matters. On an exam, watch for a prompt that specifies how to summarize data. When no reporting statistic is specified, answer at the conceptual level and avoid inventing a regulatory calculation.
Range, spread and repeatability
The range is the difference between the highest and lowest values. It is simple but depends entirely on the extremes. Standard deviation describes spread around a mean under its statistical assumptions. A coefficient of variation relates spread to the mean, but it can behave badly when the mean is near zero. Use the statistic required by the method or QA plan rather than picking the one that seems most flattering.
Repeated measurements vary because of instrument response, analysis, placement, temporal conditions and random error. Repeatability asks whether a method gives similar results under specified similar conditions. Reproducibility considers agreement when conditions such as operators, instruments or laboratories vary, depending on the definition used. The current standard’s terms control; casual use can blur important distinctions.
A wide spread should trigger questions. Were readings collected in different rooms or times? Did the test conditions change? Was a device disturbed? Were units or averaging intervals mixed? Before calling the data noisy, inspect design and records. Statistical interpretation starts with data provenance.
Uncertainty is not a mistake
Every measurement has uncertainty. It reflects limits in the instrument, method, sampling period, calibration, analysis and conditions. Reporting uncertainty does not mean the test failed; it communicates how precisely the result is known. A result with many decimal places can still have substantial uncertainty, so formatting precision must not be confused with measurement certainty.
The NRSB guide lists measurement error and uncertainty among the concepts candidates should know. Error is the difference between a measured value and an unknown true value, though that difference is not always directly observable. Uncertainty characterizes the range or dispersion of values reasonably associated with the measurement. In practice, uncertainty is evaluated using defined methods and evidence, not guessed from a single displayed number.
When a measured value is close to a decision benchmark, uncertainty and protocol become especially important. Do not round strategically to push the value above or below a threshold. Follow the reporting rule, retain the reported precision and explain any qualification required by the method. Decision-makers can then apply the applicable guidance honestly.
Control charts track process behavior
A control chart plots a quality indicator in sequence, often with a central line and statistically established upper and lower control limits. It helps a laboratory see whether the process remains stable over time. A point outside a limit, a sustained trend or a nonrandom pattern may signal special-cause variation that deserves investigation under the quality plan.
Control limits are not the same as specification or action limits. Control limits describe the observed process behavior and are derived from statistical rules; specification limits describe what a customer, standard or regulator requires. A process can be statistically stable but fail a specification, or unstable while all current points remain within specification. Keep the concepts separate.
A control chart is a decision aid, not an automatic diagnosis. An out-of-control point may follow an instrument change, maintenance, operator shift, reagent or analysis issue, or data-entry error. Investigate the event and document the response. Do not delete a point merely to restore a tidy chart unless a documented assignable cause and procedure justify exclusion.
How to read a control signal
First confirm the chart type, plotted statistic, time order and limits. Check whether the data are comparable and whether the chart’s assumptions hold. Then identify the signal rule: one point beyond a limit, a run on one side, a sustained trend or another defined pattern. The QA plan or standard establishes which rules trigger action.
Next inspect the operational record for a plausible cause. Was there a calibration, maintenance event, new analyst, new batch, changed method or unusual sample? A chronology can reveal a connection, but correlation is not proof. If the cause is unknown, keep the process under review and follow the plan’s containment and escalation steps.
Finally assess scope. Determine which results were produced since the last acceptable control point and whether they need hold, reanalysis, qualification or client notification. The time period and response depend on the quality system. A control-chart alert should be handled consistently, not only when it changes an inconvenient client result.
Compare two results without false precision
When two results differ, compare the sampling periods, locations, device methods, labs, units and building conditions. An arithmetic difference is not automatically statistically meaningful. If the method defines duplicate precision or an acceptance band, use it. If no comparable basis exists, state that the results answer different questions rather than forcing them into a single average.
Do not pool results across rooms or time intervals unless the protocol supports that aggregation. A basement and a bedroom may have different concentrations; averaging them can erase a meaningful spatial difference. Likewise, measurements taken during different building operations may not estimate the same target. The statistic cannot repair a mismatched study design.
If a candidate must explain an apparent conflict, the best answer is often to verify comparability, review QA records and repeat under controlled conditions if necessary. Avoid selecting the lower value, taking an unweighted mean or assuming that the latest test is always best. Evidence and method determine the next step.
Worked calculation: mean and range
Consider a deliberately simple practice set of three hypothetical interval values: 2, 3 and 4 pCi/L. The arithmetic mean is 3 pCi/L, and the range is 2 pCi/L because the maximum minus minimum is 4 minus 2. This example demonstrates operations only; it does not represent an official test form, a validated protocol or a health-risk calculation.
The unit remains pCi/L because all input values use that unit and the operations are addition, division and subtraction among concentrations. If one value were in Bq/m³, convert using the supplied relation before combining values. If the observations cover different rooms or unequal intervals, a simple mean may not represent the intended exposure or reporting quantity.
This is the pattern to use in an exam: state the requested statistic, show the operation, preserve units and explain whether the observations are comparable. Do not add a confidence interval or uncertainty percentage unless the question supplies the method or inputs.
Avoid common statistical traps
Averages do not remove bias. Three devices with the same calibration error can yield a very consistent mean that is still systematically off. A small standard deviation does not prove accuracy. A wide range does not prove tampering. Statistical descriptions are evidence about patterns; they must be interpreted with method and quality records.
Do not treat every threshold as a statistical control limit. EPA’s action level is a health-protection recommendation for homes; a lab control limit monitors process stability; a device acceptance specification evaluates performance under defined conditions. The numbers may all look like cutoffs, but their purposes differ. Name the authority and decision each number governs.
Avoid over-rounding, selective data removal and unsupported extrapolation. A short test cannot automatically predict annual average conditions; a single room cannot represent every floor; and a process chart cannot diagnose an individual building. Be precise about the sample and period represented.
Put quality review into the workflow
Statistical analysis should be planned before data arrive. The lab or professional defines which summaries and control rules apply, who reviews exceptions, how corrections are recorded and when clients are notified. A control chart that nobody checks is decoration. A duplicate result without an acceptance criterion is just another number.
Review the method’s input and output. Confirm sample identifiers, units, decimal conventions, missing values and dates before calculating. A misplaced decimal or a mix of pCi/L and Bq/m³ can create a mathematically correct but meaningless chart. Software can reduce arithmetic errors, but the analyst remains responsible for checking that the data are appropriate.
When a process changes, document the change and consider whether the chart needs a new baseline under the approved quality procedure. Do not reset limits simply to make a process look stable. The record should explain the technical reason and authorization for a baseline change.
Study statistics through interpretation
Make a glossary in your own words: mean, median, range, standard deviation, uncertainty, repeatability, control limit and specification limit. For each term, write one sentence saying what it can tell you and one sentence saying what it cannot. That second sentence helps prevent overclaiming.
Practice short scenarios rather than long formula drills. Ask whether two tests are comparable, whether a control signal requires investigation, whether a mean is meaningful across different rooms, or whether an action level is the same as a control limit. Use formula practice only where the study guide and current standard identify the calculation.
The official source list names ANSI/AARST MS-QA and the current standards portal. Use those materials for exact chart rules and acceptance criteria. The public outline establishes that statistical analysis is in scope; it does not publish every laboratory algorithm. Knowing that boundary is part of research literacy as well as exam preparation.
From analysis to a defensible conclusion
A sound conclusion states the statistic, the data set and interval it summarizes, the method used and any relevant quality limitation. It does not imply that a mean is a guarantee about every future moment. If a control failure affects confidence, disclose the scope and corrective step rather than hiding it in a footnote.
Explain uncertainty in plain language so the client can use the result. State that it estimates conditions during a defined period and explain when follow-up may be appropriate. Connect each statistical method to the practical question it answers.
For the RMS exam, the safest sequence is: validate records, choose the prescribed calculation, inspect variability, check controls, interpret in context and report limits. When data cannot support the requested conclusion, say so and recommend a valid next step.
Common questions
What is the difference between a control limit and an action level?
A control limit is a statistical boundary used to monitor whether a process remains stable. EPA’s radon action level is a public-health recommendation used to guide action in homes. They answer different questions and should not be treated as interchangeable.
Does a small spread prove that radon results are accurate?
No. A small spread can show that repeated results agree, but they can share a systematic bias. Accuracy requires comparison with an appropriate reference or validated performance evidence.
What should a lab do when a control-chart point is out of limits?
Follow its quality plan: preserve the data, investigate the event, assess affected results, take corrective action and document the outcome. The exact hold, repeat or notification steps depend on the current method and plan.
Can results from different rooms be averaged?
Only when the applicable method supports that summary and the readings represent a comparable target. Different rooms may have distinct concentrations, so a simple average can hide spatial variation.