The short answer: investigate first, delete last

An outlier in polyphosphate dissolution test data is not automatically a bad measurement. In most lab investigations, it is one of three things: an error in analysis or sample handling, a real but unplanned change in test conditions, or genuine product variability. The correct handling sequence is to identify which one applies before running any statistical exclusion test. Statistical tests such as Grubbs’ or Dixon’s Q only help you confirm what investigation already suggests. Using them as a first step removes real information and weakens the batch record.

A workable rule: never exclude a data point from a polyphosphate (siliphos-type) dissolution study until a documented assignable cause exists, or until a statistical test has cleared the assumptions of the dataset. The sections below explain how to do that in practice.

Why polyphosphate dissolution data produces outliers more easily than many other tests

Dissolution testing of polyphosphate crystal is inherently more variable than, say, a finished-product assay. Several factors interact:

• Time-dependent release. Polyphosphate dissolves progressively. Any difference in stirring, flow, or aliquot timing changes the measured concentration at that point in the curve.
• Temperature sensitivity. Dissolution rate rises with temperature. A small drift in a water bath or recirculation loop can shift a point well outside the rest of the curve.
• Particle size and surface area. Fines dissolve faster than coarse fractions. If a sample aliquot carries undissolved fines through a filter, the measured value reads high. If fines are lost during handling, it reads low.
• Hydrolysis of polyphosphate to orthophosphate. If the method calculates polyphosphate by difference (total phosphate minus orthophosphate), a sample held too long or too warm will show reduced polyphosphate and increased orthophosphate. That shift looks like an outlier but is a holding-time artifact.
• Colorimetric sensitivity. Standard molybdenum-blue methods are precise but sensitive to reagent age, acid strength, timing, and cuvette cleanliness.
• Unit and species conversions. Reporting as P, PO4, or P2O5 without a consistent conversion factor is a frequent source of “outliers” that are simply arithmetic errors.

Because these sources overlap, the same numeric value can mean very different things. Classification matters more than the test statistic.

Classify the outlier before you test it

1. Analytical causes

Check the instrument log, calibration curve, reagent lot, and dilution sequence for the run in question. Recalculate the value from raw absorbance or raw titration data. Confirm that the dilution factor and the P/PO4/P2O5 conversion used for that sample match the rest of the dataset.

2. Sampling and physical causes

Review aliquot timing, filtration step, filter pore size, sample temperature at the moment of quenching, and whether the vessel was stirred consistently. If the dissolution rig uses a recirculation loop, check for dead volume or partial blockage. These causes usually affect one point or one replicate rather than the whole curve.

3. Genuine process or material variation

If no analytical or sampling fault is found, the value may be real. A single high value can indicate a fines-rich fraction; a low value can indicate a coarse fraction, a partially wetted sample, or unexpected hydrolysis. This is the category that should never be deleted, because it carries product information.

A practical outlier workflow

1. Freeze the raw data. Record the value, timestamp, analyst, instrument, reagent lot, temperature, aliquot volume, dilution factor, and any observations. Do not overwrite it later.
2. Verify the calculation independently — units, conversion factor, dilution, and blank subtraction.
3. Look for an assignable cause using the three categories above. If one is found and documented, the point can be excluded with justification.
4. Re-measure if a retained sample exists. A repeat measurement on the same aliquot or a retained portion is the strongest evidence for or against an analytical fault.
5. Only then apply a statistical test, and only if the dataset satisfies that test’s assumptions.
6. Report the decision. State whether the value was retained, excluded, or flagged, and why. Report results both with and without the excluded point if the exclusion changes the conclusion.

Choosing a statistical test for polyphosphate dissolution data

The table below summarizes common options. None of them replace an investigation; they formalize it.

Note the pattern: every parametric option assumes an approximately normal distribution. Cumulative dissolution data rarely satisfies that assumption across time, because concentration rises with time by design. Applying Grubbs’ test to raw time points will flag the late, higher-concentration points as outliers in perfectly valid runs.

Handling outliers in a dissolution curve rather than a single value

If your study produces a release curve, test the residuals, not the raw values:

1. Fit the appropriate model to the run — a first-order, Higuchi, or simple plateau model, depending on your method and how the release mechanism behaves.
2. Calculate residuals for each time point.
3. Apply an outlier test to the residuals, which are centered near zero and far closer to normal.
4. Where a model is not justified, analyze the derived parameters (initial rate, plateau value) across replicates instead of the individual time points.

This approach also separates two different problems: a single bad time point (usually analytical) versus a whole curve that sits off-target (usually material or method).

What to do with the outlier once it is confirmed

• Retain and flag: appropriate when the value is real and belongs to the population of interest.
• Exclude with documented cause: appropriate when an identified analytical or sampling fault is proven.
• Winsorize or substitute the median: occasionally used in statistical modeling, but rarely appropriate for a regulated batch record. If used, state it explicitly and keep the original value.
• Use robust summary statistics: median and median absolute deviation describe skewed dissolution data better than mean and standard deviation.
• Increase replication: if outliers keep appearing without assignable cause, the problem is method repeatability, not the individual data point.

Documentation that survives review

An outlier decision should be traceable from raw data to conclusion. A defensible record includes the raw value, the investigation performed, the assignable cause if found, the statistical test and its pre-defined criterion, the identity of the reviewer, and the impact on the reported result. Pre-define the acceptable outlier procedure in the test method or SOP. Deciding the rule after seeing the data is the fastest way to lose an audit argument.

Common mistakes that create false outliers

• Using P, PO4, and P2O5 interchangeably across a dataset.
• Applying a normality-based test to a time series without detrending.
• Running Grubbs’ test repeatedly on the same dataset, which masks multiple outliers.
• Deleting a point because it “looks wrong” without any recorded cause.
• Ignoring holding time and temperature when polyphosphate is measured by difference against orthophosphate.
• Comparing dissolution results between products, particle size grades, or methods as if they were interchangeable.

Frequently asked questions

Should I remove an outlier if the measurement was obviously faulty?

Yes, but only with a documented cause such as a spilled aliquot, a mis-pipetted dilution, or an instrument fault. “Obviously faulty” needs an evidence trail, not an opinion.

How many replicates do I need before a statistical outlier test is meaningful?

Grubbs’ test is generally used with seven or more replicates. With three to five replicates, statistical tests have limited power, and an investigation plus a repeat measurement is usually more informative than a formal test.

Does an outlier mean the polyphosphate material is defective?

Not by itself. It may indicate fines, incomplete wetting, a particle-size shift, or an analytical fault. Material conclusions require a pattern across replicates or batches, not a single point.

Why does my polyphosphate dissolution test show high variability even with good technique?

Common contributors are temperature drift, inconsistent stirring or flow, filtration of undissolved fines, and hydrolysis during sample holding. Review these before changing the product specification.

Can I compare my results with another lab’s dissolution data directly?

Only if the method, sample preparation, temperature, filtration, and calculation basis are the same. Comparing results measured as different phosphate species or under different hydrodynamic conditions produces differences that are procedural, not material.

Discuss your polyphosphate dissolution test data

If you are setting up or troubleshooting a dissolution method for siliphos-type polyphosphate crystal, ENVO CHEMICAL can review the technical details with you. To make the discussion specific and useful, please share the test method you are using, the source-water conditions, the equipment type and flow arrangement, the specification or particle-size range you need, the quantity under consideration, and the destination market. With those details, we can talk about the material and method factors that most often drive outliers in this type of testing.