Page 26 - qp_17

show reason for the outliers, analysed additional samples and use the
data from them to augment (expand) the data set.

7. Visually check for outlier and linear relationship
Plot 4 graphs
a) New method (Y axis) & old method (X axis)

b) Individual value of NEW methodVS Mean of OLD method

c) Different between mean of NEW method-mean of OLD method for each
sample VS Mean of OLD method (horizontal central line has value of zero).
• BUT if the comparative method is NOT a reference method, the third
plot is bias plot where the difference between the mean of NEW and
mean of OLD method against (mean NEW + mean OLD)/2.

d) Difference between individual value of NEW method and Mean of OLD
method VS Mean of OLD method.
• BUT if the comparative method is NOT a reference method, the fourth
plots the difference between individual value of NEW method with
average of(individual value of NEW method - mean OLD method)/2
against (mean NEW+ mean OLD)/2

8. Test for adequate range of sample
a) Regression analysis assume that the variables is known without error,
however if range of data is sufficiently wide, the effect of error on
regression estimates can be considered negligibly small.

b) The correlation coefficient, r

c) rough guide to assess the adequacy of the range of data

d) Adequate sample?
• r ≥ 0.975 or equivalent r² ≥ 0.95
• If r² < 0.95 , add samples n examine again
If range cannot be extended, use partition bias procedure in place of
linear regression to estimate average bias

9. Estimation of Linear Regression (When passed adequate range)
a) There are two ways:
• Computation - for calculation of Y= bX + a
• Alternative regression procedure of which only estimating the slope
and intercept.
√ Deming or
√ Passing-Bablok

10. Visual check for Constant Scatter