quadratic curves in chemstation

[tangaloomaflyer]

A couple of questions:

From what I can gather, when you set chemstation to fit curves by quadratic regression through 0,0 the curve that is generated always takes the form:

Y = ax2 + bx

where y = response ratio; x = conc. ratio; a does not equal zero

This is a vertical parabola. For most analytes, response reaches a maximum at a certain concentration, so the parabola is an 'arch' (as opposed to a 'valley') and the quadratic term (ax2) is negative.

This means there is a maximum response, above which no result can be returned. Is this why chemstation sometimes returns a -2.00 result (which I presume is an error code)?

Also, there are two possible conc. ratios for a given response below the maximum. I presume chemstation always returns the lowest result?

I would have thought that a better curve would be:

x = ay2 + by

which would give a horizontal parabola peaking to the left (c-shaped). this curve would still have a maximum response but only one possible concentration for a given response.

I guess it's swings and roundabouts but is there a way to choose which parabolic function is used? I thought maybe chemstation would select the most appropriate but I entered in data that would give a perfect horizontal parabola it chose the vertical parabola.

[tom jupille]

I suspect that it's because an implicit assumption in regression calculations is that all of the error is associated with the dependent variable (y), so the form is always y = {something}

[lmh]

also, on our copy of LC Chemstation, Chemstation will not extrapolate beyond the highest concentration in the standard curve when using a quadratic curve. This makes sense, as errors become huge very rapidly. In any case, we shouldn't be working outside our curve.

If your copy returns a value of -2, be happy! Ours returns a peak area, but an amount value of zero, which is actually more misleading, as it might equally mean "undetected".

[mbicking]

The ChemStation should generate a "Warning" at the end of the report in some cases. For linear curves, a negative result is set to zero with a warning (which is often not noticed).

For second order polynomials, you are correct about the mathematics of the curve. But you would only use this approach if your data are all on one side of the vertex. If not, then you either have very unusual data or a completely wrong model. This approach is only useful for mild saturation, typically in the detector. The curve is only flattening out to a small degree. If you have more flattening, then it is time to fix the chromatography and/or method.