Getting negative concentration when quantifying results

[duarte]

I am using HP 7694 headspace analzyer and a HP 6890 GC. I am using Chemstation software.

Not sure if this is the right forum, but I am having trouble getting non-negative numbers when I generate a report after running a sample through the GCFID. I just finished setting up AutoQuant following this: http://www.chem.agilent.com/Library/Sup ... a05214.pdf . But whenever I generate a report, the calculated concentration for the peak area comes out negative. I think changing the integration parameters might help, but I am quite lost in that respect.

Any help would be appreciated!

[Steve Reimer]

How does your peak area compare with the area in your standards? If the area is not within the calibration range then you can have some interesting concentration calculations if you aren't using average fit or forcing zero.
How good of fit is your calibration model?

[duarte]

The peak area is within the area of my standards. I am getting r^2 values of 0.983 - 0.994. I don't know what you mean by average fit or forcing zero, could you elaborate?

[kubowicz.tomasz]

Hello

Negative concentration you've got is result of calibration curve setting - forcing the curve through zero is not the same as including the origin as a fictitious point in the calibration.

Regards

Tomasz Kubowicz

[rb6banjo]

Duarte, when you analyze your samples, are the peak areas you measure closer to the low-end of your calibrated range? If you take the y-intercept (peak area side) of your calibration and divide it by the slope of the line, you can get an idea of your minimum quantitation limit. It's actually a confidence interval around that intercept meaning that your measured response must be a certain amount larger than the intercept to give you the ability to say that it's actually real. A good analytical chemistry text book can take you through the explanations if you're not quite following what I'm saying here.

Anything lower than the calculated positive intercept will give you a negative concentration because when you allow for an intercept (b) in the calibration model, the intercept must first be subtracted from the response (R) before division by the slope (m) to calculate the concentration:

R = m*C + b

C = (R - b)/m

If R ~ b, you'll get a number close to zero as the answer. If R < b, it will go negative. Likewise, a negative intercept will make your calculated concentrations larger (subtracting a negative is adding a positive) than they are in reality.

Try to make sure that you get your samples to lie as close to the middle of your calibrated range as you can. This will get you as far from this sort of behavior as you can get.

[duarte]

Thanks everyone!

I re did the calibration with lower concentrations and that seemed to do the trick. Banjo your explanation was fantastic thank you.

Duarte

[MSCHemist]

Another issue I've come across is that a simply linear regression assumes homoscedacity of the errors. That is you have the same absolute uncertainty at all calibration levels when more often it is somewhat relative. Put simply to generate the line you are making Y axis corrections and you are just as likely to make corrections equal to 1ppm at 1ppm as at 1000ppm because all the residuals are given the same weight. It can often be beneficial to apply a weighing factor to the calibration when you have a range greater than say 20 fold such as weight = inverse to concentration. Some chemists will force or include the origin which can have a similar effect but is not as valid.

[tangaloomaflyer]

You might also see a -1.00 result which indicates an error. If you have a quadratic calibration curve that intercepts the x-axis twice where x>0 then a concentration cannot be calculated for certain responses. If you imagine a bell curve that crosses the x-axis twice to the right of the y-axis, then a response will either have two possible concentrations, or none if it is above the peak of the curve. Another situation would be a 'u' shaped curve that intercepts the y-axis but not the x-axis. For this curve a response below a certain level will have no corresponding concentration.

Of course, if your cal curve looks like that then there I probably something wrong with your calibration....