Chromatography Forum

Does any one have any information on how to calculate the uncertainty associated with bias from a forced through zero calibration curve.

We have a dual column headspace GC method for volatiles that uses a 4 point calibration curve with a linear regression that forces through zero using the data from one column. One column is used for quantification and the other for compound verification.

We are in the process of calculating the measurement uncertainty for this method and are a little stuck on how to include the bias from the force through zero.

One way I thought of was to take several batches of data (or our original validation data) and calculate both ways (through zero and not), and see what the difference would be at different levels of the curve. Then adding the noted effect into the budget in some fashion. I am not sure this is a valid approach.

Any information and assistance would be greatly appreciated.

NEVER FORCE THROUGH ZERO! I use 6 concentrations levels and statistically calculate the LOD and LOQ from the sum of the squares of the standard deviation.

I like force through zero just fine. It is a least squares fit to y = mx, rather than y = mx + b. If you are not linear down to the concentration you need, forcing zero will fail. If you are linear, no agonizing over tiny numbers you have no business reporting anyway.

I would use the R or the R squared value to estimate uncertainty.

The merits and otherwise of forcing calibrations through zero have been gone into repeatedly and at length on the forum already. I think that it is fair to summarize the consensus as; you can do it if the intercepts are not significantly different from zero, and if the intercepts are not significantly different from zero, why bother forcing the line through zero ?

Paradox time: you acknowledge that forcing through zero introduces bias, and the GUM requires that sources of bias be identified, and eliminated if possible, so a contribution to uncertainty from a force through zero has no place in a formal uncertainty expression.

Peter

I will add to Peter's comment that the magnitude of your intercept is more of an indicator of the nonlinearity of your data far from zero concentration. If you obtain a large intercept from your calculations, you should probably examine where/why your data is nonlinear.

Here's one of the more recent discussions that I can recall on this subject:

viewtopic.php?f=2&t=32611&p=156841&hili ... on#p156841

Have fun!

Only reason I found to force zero is to estimate background concentration of a contaminant, such as Acetone in the air when you are doing Purge and Trap samples. Otherwise you may get negative concentrations from the linear curve, and I hate negative concentrations since they are so difficult to weigh up