Calibration (The Good Curve) - EnviroCompliance

C a libratio n (The G o o d C u rve)

Greg Hudson EnviroCompliance Labs, Inc. greghudson@

A b s tra c t

n It should come as no surprise that the correlation coefficient is only one way to judge the quality of a calibration curve.

n The fact is that there are numerous tools available to evaluate how good a calibration curve represents a given data set.

n With a basic understanding of how to use regression analysis, some simple methods can be applied to determine both the quality of necessary to improve calibration technique.

O v erv iew

n The black box versus thinking outside the box.

n The black box is the regression algorithm.

n Thinking outside the box is evaluating analytical processes or biases which may effect the quality of the curve.

W h at is a C u rv e ?

n STANDARD CURVE:

A plot of

concentrations of known analyte

standards versus the instrument

response to the analyte. Calibration

standards

are

prepared

by

successively diluting a standard

solution to produce working

standards which cover the working

range of the instrument. Standards

should be prepared at the frequency

specified in the appropriate

section....

SW-846

OK, what do we really mean by curve?

n The ability to predict behavior by establishing known outcome.

n Simply put, getting a result with reasonable certainty.

W h at' so important about a g o o d C u rv e ?

n Easier to correct problems if the calibration is bad

n Higher confidence in analytical results

A Peek inside the Black Box

n Basic linear regression by least squares

n LINEAR Eq.: x=(y-b)/m or y=mx+b

n where m = slope or rise over run

n

b = intercept (curve crosses the y axis)

n

(xy - (x * y /n))

n Slope = ---------------------------

n

(x^2 - (( x)^2 / n))

n intercept = (y/n) - (m *x / n)

T h e C o rrelatio n C o e ffic ient

n An estimate of the remainders derived from the regression.

F How do we calculate it?

n

nr

n

(xy - (x * y / n))^2 = ----------------------------------------------

((x^2 - ((x)^2 / n)))*(y^2 - ((y)^2 / n))

u What does it mean?

u For our purpose it must be >0.995

u The correlation coefficient only indicates variance from the averages; therefore it is not the best indicator of curve quality.

 ................
................

In order to avoid copyright disputes, this page is only a partial summary.

Google Online Preview   Download