On the subject in the title: Specificity, linearity, accuracy, precision, range, robustness — with real criteria — a second dataset Working notes rather than a conclusion.
Posting a small dataset on Specificity. It is mine, it is uncontrolled, and the method is stated so it can be discounted appropriately.
What I would like is not agreement but a second dataset collected by someone with no stake in mine. If one exists I would rather read it than argue for this one.
A method that reports the same number for every lot is worth a second look. Real processes vary, and a total absence of variation is a statement about the measurement rather than the process.
Not a strong opinion, just a consistent one.
The opening post and I disagree about the size of the effect, not about the direction.
Range and working range are different things and a certificate rarely distinguishes them. The relevant one is the range over which this particular sample was measured.
That is the shape of it. The detail is where I would expect to be corrected.
Trying to state the Specificity position in a way that someone who disagrees would recognise as fair, because I do not think the version in this thread passes that test.
Building on post #6 rather than restating it.
Robustness testing deliberately varies the parameters most likely to drift — organic percentage, pH, temperature, flow — and shows the result does not. It is the part of validation that predicts whether a method will transfer.
If anyone can point at the primary source I would be grateful.
Robustness: the method gives consistent results when minor parameters vary. Tested by deliberately varying pH, temperature, flow rate, and mobile phase composition within reasonable ranges and demonstrating that results stay within acceptance.
I had written a reply contradicting post #9 and deleted it. Here is what survived.
What I would tell a new member reading about Specificity for the first time: the confident posts are not the reliable ones, and the reliable ones are longer.
The arithmetic in post #11 is right; the assumption feeding it is the part to check.
Specificity: the method can distinguish the intended compound from related impurities and degradation products. Tested by comparing results on pure compounds, mixtures of compounds, and degraded samples.
Reporting a result to more decimal places than the method's precision supports is a small dishonesty that appears everywhere. A method with a two per cent relative standard deviation does not support a figure quoted to a hundredth.
If it helps: the failure mode here is usually boring rather than dramatic.
Linearity across the working range is a routine demonstration and it constrains how far a result can be extrapolated. A method linear from 80 to 120 per cent of nominal says nothing about a sample at ten per cent.
Noting that the question and the thing people usually mean by it are different.
Linearity: the detector response is proportional to compound concentration across the working range. Demonstrated by running standards at multiple concentrations and showing R-squared values typically ≥0.99.
Where I part company with post #16, and it is a narrow parting.
Two things can be true about Specificity at once: the mechanism is plausible and the evidence for the size of the effect is thin. Most of the argument here is people defending the first against attacks on the second.
Post #16 is the version of this I will quote in future. One addition.
I keep a log for Specificity specifically because my memory of it turned out to be systematically wrong in one direction. Six weeks of notes cost nothing and settled it.
I read post #16 twice before replying, because I had assumed the opposite.
The limit of quantitation determines what the impurity table can honestly contain. Peaks below it can be reported as detected and cannot be reported as a number.
Correct me on the arithmetic if it is wrong; I would rather know.
Specificity is the first question: does the method separate the analyte from everything reasonably expected to be present? A method that has not been challenged with its own degradation products has not answered it.
I would call that likely rather than established.
On Specificity the community has more anecdote than the confidence in this thread implies, and I include my own contribution in that.
Picking up post #20: that is the part I would want checked first.
System suitability: injections run at the start of a batch to establish that the instrument and column are performing. Acceptance criteria typically include replicate precision (RSD ≤2%), peak tailing (0.8–1.5), theoretical plates (>2000), and resolution (>1.5).
That is all the detail I have. Someone else will have more.
On post #22 — agreed on the reasoning, with one qualification.
Accuracy: the method measures what you intend to measure. For purity methods, this is tested by spike-and-recover experiments: add a known amount of impurity to a sample and measure whether you recover the added amount.
Someone will know this better than I do and I hope they say so.
Precision and repeatability: within-run and between-run variability of the method. Acceptance criterion is typically a relative standard deviation of ≤2% for area measurements.
Posting my Specificity numbers with the method attached so they can be discounted properly. Uncontrolled, unblinded, and collected by someone who wanted a particular answer.
Range: the concentration range over which the method has been validated. Going outside the validated range is going outside the method's demonstrated performance.
Limits of detection and quantitation: LOD is the lowest concentration that produces a signal above background. LOQ is the lowest concentration at which the method meets precision and accuracy acceptance criteria. Both are determined empirically.
Worth saying I have only my own numbers here, and n is small.
Where I part company with post #26, and it is a narrow parting.
Precision has two levels worth distinguishing: repeatability within a run and intermediate precision across days, analysts and instruments. The second is where most methods lose the numbers people quote.
I keep a log of this specifically because memory is unreliable about it.