9780521516358book.pdf

test is based on anull hypothesis, namely that there is no difference in the values

being compared. If the hypothesis is proved to be correct then the suspect value

cannot be rejected. The suspect value is used to calculate anexperimental rejection

quotient,Qexp.Qexpis then compared with tabulatedcritical rejection quotients,

Qtable, for a given confidence level and the number of experimental results (Table

1.8). IfQexpis less thanQtablethe null hypothesis is confirmed and the suspect value

should not be rejected, but if it is greater then the value can be rejected. The basis of

the test is the fact that in a normal distribution 95.5% of the values are within the

range of two standard deviations of the mean. In setting limits for the acceptability or

rejection of data, a compromise has to be made on the confidence level chosen. If a

high confidence level is chosen the limits of acceptability are set wide and therefore

there is a risk of accepting values that are subject to error. If the confidence level is set

too low, the acceptability limits will be too narrow and therefore there will be a risk of

rejecting legitimate data. In practice a confidence level of 90% or 95% is most

commonly applied. TheQtablevalues in Table 1.8 are based on a 95% confidence level.

The calculation ofQexpis based upon equation 1.14 that requires the calculation of

the separation of the questionable value from the nearest acceptable value (gap)

coupled with knowledge of therangecovered by the data set:

Qexp¼

xnxn 1

xnx 1 ¼

gap

range ð^1 :^14 Þ

wherexis the value under investigation in the seriesx 1 ,x 2 ,x 3 ,...xn–1,xn.

1.4.5 Validation of an analytical method – the use oft-tests

At-test in general is used to address the question as to whether or not two data sets

have the same mean. Both data sets need to have a normal distribution and equal

variances. There are three types:

• Unpairedt-test: Used to test whether two data sets have the same mean.


• Pairedt-test: Used to test whether two data sets have the same mean where each value
  in one set is paired with a value in the other set.


• One-samplet-test: Used to test whether the mean of a data set is equal to a particular
  value.

Table 1.8Values ofQfor the rejection of outliers

Number of observations Q(95% confidence)

4 0.83

5 0.72

6 0.62

7 0.57

8 0.52

27 1.4 Quantitative biochemical measurements