Intercept in Linearity

[pat_val]

Hi!

Right now i am validating one method. For Linearity i have r2 is 0.9999 but my intercept is in negative. Does it matter? if yes than how?

Thank you...

[tom jupille]

You should be able to burp your statistics program and get an estimate of the standard error of the y-intercept. If that interval around your intercept includes zero, then your intercept is statistically indistinguishable from zero.

If your intercept is really non-zero by the above test, the consequences depend on what you are doing. For pharmaceutical, major-component assay, I don't think a zero intercept is required, so long as you have standard bracketing your expected value. If you are measuring over a wide range, then you may want to look at a weighted least-squares fit. There is a fairly extensive discussion of some of these issues going on in another thread:
http://www.sepsci.com/chromforum/viewtopic.php?t=7465

[adam]

It does not matter so much whether your intercept is negative or positive. what is important is its magnitude: the smaller the better.

One test for this is the 95% confidence interval test (which Tom mentioned). But with the very high precision of moder injection systems, I think that approach can get you in trouble. what we generally do is state that the magnitude of the intercept should be less than or equal to 3% of the area count values that you get for the analyte at 100%.

[rnelson]

For method validation, we also use correlation coefficient, but to evaluate intercept, we use the ASTM procedure of calculating the response factor rsd of your five standards. From my experience, there appears to be a direct correlation between the magnitude of the intercept and this rsd.

There are some HPLC vendors that also use this method.

I would like to use this same approach for LC calibration and I'm currently performing some data analysis to determine what impact it might have.

I'd be curious to know what companies do to evaluate intercept for LC calibration.

[JGK]

As mentioned previously you can test the standard error (SE) of the intercept to determine if your and detrmine if 0 is within the intercept ± SE range.

However, in regulatory work a zero intercept is of little practical use since the regulations/guidelines do not allow for calculations outside the calibrated range (determined by the non-zero STDs) of the assay.

Is there any particular reason for a zero intercept requirement in your assay?

[Stanley110]

When preparing calibration lines by regression, the line should be corrected for "constant error" by application of the Total Youden Blank, and for proportional error.

After correcting for these errors, the signals or responses are true analyte signals and analysis from that point forward is correct.

As a general rule, where linearity has been established and quantitation is performed by external standard, error is tiny and acceptable when the signal or area count from the standard is essentially the same as that of the sample.

Stan Alekman