GIST 

Anal Chem 9/26/2016

Testing for a supiciously high (or low) result

1

 To judge whether an extremely, and hence suspiciously high (or low) value in a set of measurements is an
outlier, we have a choice of three different tests:
 If the standard deviation is well known beforehand from N > 20 previous measurements (hence s ≈ σ), we can compare
the difference between the suspected value and the mean of all data in the set with the critical Z-value.
 If the standard deviation is not known beforehand, but rather is computed from the same limited set, the suspicious value
included, we use the Grubbs' test.
 For a small numer of observations (N < 10), we can also apply Dixon's Q-test.
 The t-test can not be used for outlier testing: This is because, with increasing separation of the suspected
outlier, the standard deviation increases even more rapidly, such that the t-value becomes even smaller and
never exceeds a critical test value. It is for this reason that the Grubbs and Dixon tests have been
designed. Because they are based on fewer measurements, Grubbs and Dixon are more severe than the Z-test.
 Although the 5% level is usually applied in statistical tests, 1% is recommended for outlier testing, to avoid rejecting data
too readily.
 Rejection becomes more acceptable if we can propose a possible reason why a particular measurement does not belong to
the same population as the others: have we switched samples, used the wrong reagent…..?

http://www.chromedia.org/chromedia?waxtrapp=ksckzDsHonOvmOlIEcCxBW&subNav=kczkgDsHonOvmOlIEcCxBWV

5. QUALITY ASSURANCE &

CALIBRATION METHODS
GIST Analytical Chemistry by Kim, Tae-Young

1
GIST Anal Chem 9/26/2016

The need for quality assurance

: Measurements of lead in river water
3

181 different laboratories 9 different national institutes

1. Basics of Quality Assurance

4

 Quality assurance: What we do to get the right answer for our purpose.
 The right answer should have sufficient accuracy & precision to support subsequent decisions.
 There is no point in spending extra money to obtain a more accurate or more precise answer if it is not
necessary.

 Raw data: Individual values of a measured quantity.

 Peak areas from a chromatogram, volumes from a buret, …
 Treated data: Concentrations or amounts derived from raw data by use of calibration
method.
 Results: Quantities reported after statistical analysis of treated data.
 The mean, standard deviation, & confidence interval.

2
GIST Anal Chem 9/26/2016

Use objectives & specifications

5

 Use objective: States purpose for which results will be used.

 Specifications: States how good the numbers need to be & what precautions are required in
the analytical procedure.
 Sampling requirements.
 Accuracy & precision.
 Rate of false results.
 Selectivity & sensitivity.
 Acceptable blank values.
 Recovery of fortification.
 Calibration checks.

Terms in specifications
6

 Suppose you must certify that a contaminant in drinking water is below a legal limit.
 False positive: The concentration exceeds the legal limit when, in fact, it is below the limit.
 False negative: The concentration is below the limit when it is actually above the limit.
 In choosing a method,
 Selectivity (specificity): The ability to distinguish analyte from other species in the sample.
 Sensitivity: The capability of responding reliably & measurably to changes in analyte concentration.
 Blanks: Account for interference by other species in the sample & for traces of analyte found
in reagents used for sample preservation, preparation, & analysis.
 Method blank: A sample containing all components except analyte & it is taken through all steps of the
analytical procedure.
 Reagent blank: Similar to a method blank, but it has not been subject to all sample preparation procedures.
 Field blank: Similar to a method blank, but it has been exposed to the site of sampling.

3
GIST Anal Chem 9/26/2016

Spike recovery
: Check the matrix effect on the response to analyte
7

 Matrix: Everything in the sample other than analyte.

 Spike (fortification): A known quantity of analyte added to a sample to test whether the
response to a sample is the same as that expected from a calibration curve.
 Spiked samples are analyzed in the same manner as unknowns.
 If drinking water is found to contain 10.0 μg/L of nitrate, a spike of 5.0 μg/L could be added.
 Ideally, the concentration in the spiked portion found by analysis will be 15.0 μg/L.
 Add a small volume of concentrated standard to avoid changing the volume of the sample significantly.

 A definition of spike recovery

Cspiked sample  Cunspiked sample
% recovery =  100
Cadded
 C: Concentration.

Calibration check & Performance test samples

8

 Calibration check: Analyzes solutions formulated to contain known concentrations of

analyte.
 Solutions for calibration checks should be different from the ones used to prepare the original calibration
curve.

 Performance test samples (quality control samples or blind samples): Samples of known
composition provided to the analyst as unknowns.

 Summary: How to gauge analytical performance?

 Accuracy: Calibration checks, fortification recoveries, quality control samples, blanks, …
 Precision: Replicate samples, replicate portions of same sample, …

4
GIST Anal Chem 9/26/2016

Standard operation procedures & Assessment

9

 Standard operating procedures (SOP): State what steps will be taken & how they will be
carried out.
 Adhering to these procedures guards against the normal human desire to take shortcuts based on false
assumptions.

 Assessment: The process of (1) collecting data to show that analytical procedures are
operating within specified limits & (2) verifying that final results meet use objectives.
 Documentation is critical for assessment.
 Standard protocols: Provide directions for what must be documented & how the documentation is to be
done.
 Control chart: A visual representation of confidence intervals for a Gaussian distribution (Box 5-1).

2. Method Validation
10

 Method validation: The process of proving that an analytical method is acceptable for its
intended purpose.
 Specificity, linearity, accuracy, precision, range, limit of detection, limit of quantification, & robustness.
 Linearity: How well a calibration curve follows a straight line, showing that response is
proportional to the quantity of analyte.
2
 Square of correlation coefficient, R2:   ( xi  x)( yi  y ) 
R = 
2 
 ( xi  x)2  ( yi  y)2
 The y-intercept of the calibration curve (after subtraction of the blank) should be close 0.
 Range: The concentration over which linearity, accuracy, & precision are all acceptable.
 Robustness: The ability of an analytical method to be unaffected by small, deliberate
changes in operating parameters (e.g., solvent composition, pH, buffer concentration,
temperature, injection volume, & detector wavelength in a chromatographic method).

5
GIST Anal Chem 9/26/2016

Limit of Detection (LOD)

11

 Detection limit (lower limit of detection): Student’s t for 6 DF Student’s t for 6 DF

The smallest quantity of analyte that is

“significantly different” from the blank.

 False positive: There is 1% chance of

concluding that a blank has analyte above
the detection limit.

 False negative: There is 50% chance of

concluding that analyte is absent because
the signal is below the detection limit.
Distribution of measurements for a blank and a
sample whose conc. is at the detection limit.

Procedure producing a detection limit with ~99%

chance of being greater than the blank
12

 Assume that ssample (near the detection limit) = sblank

 After estimating the detection limit from previous experience with the method, prepare a sample whose
conc. is ~1 to 5 times the detection limit.
 Measure the signal from n replicate samples (n ≥ 7).
 Compute the standard deviation (s) of the n measurements.
 Measure the signal from n blanks (containing no analyte) & find the mean value, yblank.
 The minimum detectable signal, ydl, is defined as
Signal detection limit: ydl = yblank + 3s
 The corrected signal, ysample – yblank, is proportional to sample conc.:
Calibration line: ysample – yblank = m × sample conc.
(ysample: The signal observed for the sample, m: The slope of the linear calibration curve)
 Detection limit: Minimum detectable conc. = 3s/m

6
GIST Anal Chem 9/26/2016

Example
: Detection limit
13

 An estimated signal detection limit from previous measurements: low nA range.

 Signals from 7 replicate samples with a conc. about 3× the detection limit:
5.0, 5.0, 5.2, 4.2, 4.6, 6.0, & 4.9 nA
 Signals from reagent blanks: 1.4, 2.2, 1.7, 0.9, 0.4, 1.5, & 0.7 nA
 The slope of the calibration curve for higher concentrations: m = 0.229 nA/μM
 Blank: Average = yblank = 1.26 nA
 Sample: Standard deviation = s = 0.56 nA
 The signal detection limit: ydl = yblank + 3s = 1.26 nA + (3)(0.56 nA) = 2.94 nA
 The minimum detectable conc.:
3s (3)(0.56 nA)
Detection limit    7.3 μM
m 0.229 nA/μM

Limit of quantitation & reporting limit

14

 The standard deviation is a measure of the noise (random

variation) in a blank or a small signal.
 When a signal is 3× greater than the noise, it is detectable, but
still too small for accurate measurement.
 Lower limit of quantitation: The smallest amount that can be
measured with reasonable accuracy.
10 s
Lower limit of quantitation 
m
 Reporting limit: The conc. below which regulations say that
a given analyte is reported as “not detected.” which does not
mean that analyte is not observed.
 At least 5 to 10× higher than the detection limit.

7
GIST Anal Chem 9/26/2016

3. Standard Addition
: Matrix & matrix effect
15

 Matrix: Everything in the unknown, other than analyte.

 Matrix effect: A change in the analytical signal caused
by anything in the sample other than analyte.

 The figure on the right shows a strong matrix effect in

the analysis of perchlorate by mass spectrometry.
 Perchlorate > 18 μg/L in drinking water can reduce thyroid
hormone production.
 The slope in the lower curve for standard solutions in
groundwater is 15× less.
 Reduction of the perchlorate signal is a matrix effect attributed
to other anions present in the groundwater. Calibration curves for perchlorate
in pure water & in groundwater.

Standard addition
16

 Standard addition (“Spiking” of the sample): Known quantities of analyte are added to the
unknown.
 From the increase in signal, we deduce how much analyte was in the original unknown.
 Requires a linear response to analyte.
 As in titrations, higher precision can be achieved when standards are added by mass, instead of volume.

 Standard addition especially appropriate when the sample composition is unknown or

complex & affects the analytical signal.
 When we add a small volume of concentrated standard to an existing unknown, we do not change the
concentration of the matrix very much.

8
GIST Anal Chem 9/26/2016

Standard addition
17

 Before standard addition:

 Initial concentration of standard: [S]i
 Unknown initial concentration of analyte: [X]i
 Signal intensity of unknown analyte: IX = k[X]i (k: proportionality constant)
 After standard addition:
 Concentration of standard in the final solution: [S]f
 Diluted concentration of analyte: [X]f
 Signal intensity after addition of standard to analyte: IS+X = k([S]f + [X]f)
 Signal is directly proportional to analyte concentration,
 Standard addition equation:
[X]i I
 X
[S]f  [X]f I S X

Dilution factor
: Ratio of final concentration to initial concentration
18

 Initial volume of unknown: VO

 Added volume of standard with initial concentration [S]i: VS
 The total volume after standard addition: V = VO + VS

 The concentrations in the standard addition equation are

V  V 
[X]f  [X]i  O  [S]f  [S]i  S 
V  V 

Dilution factor Dilution factor

9
GIST Anal Chem 9/26/2016

Example
: Standard addition
19

 Find the original concentration of Na+ in the serum.

 Signal of serum containing Na+: 4.27 mV
 Signal after addition of 5.00 mL of 2.08 M NaCl to 95.0 mL of serum: 7.98 mV

V   5.00 
[S]f  [S]i  s   (2.08 M)    0.104 M
V   100.0 

V   95.0 
[X]f  [X]i  O   [X]i    0.950[X]i
V   100.0 

[X]i I [X]i 4.27 mV

 X  [X]i = 0.113 M
[S]f  [X]f I S X (0.104 M)  0.950[X]i 7.98 mV

Graphical procedure
: Successive standard addition to a single solution
20

 In case that the analysis does NOT consume solution.

 Measure the analytical signal of an unknown soln.
 Add a small vol. of conc. standard & measure the signal.
 Add several more small vol. of standard & measure the
signal after each addition.
 Added standard should increase the analytical signal by a
factor of 1.5 to 3.

V  I V 
I S X    I X  X [S]i  S 
 VO [X]i  VO
 
Function to plot Function to plot
on y -axis on x -axis
x-intercept: –[X]i

10
GIST Anal Chem 9/26/2016

Uncertainty & confidence interval

in the x-intercept
21

 Standard deviation of x-intercept:

2
sy 1 y
 2
m n m  ( xi  x) 2
 sy: The standard deviation of y (Eqn. 4-20).
 : The absolute value of the slope of the least-squares line (Eqn. 4-16).
 n: The number of data points.
 : The mean value of y.
 xi: The individual values of x.
 ̅ : The mean value of x.
 Confidence interval: ±t × (standard deviation of x-intercept)
 t: Student’s t for n − 2 DF.

Graphical procedure
: Standard addition with constant total volume
22

 In case that the analysis consumes some of the solution.

 IS+X = k([S]f + [X]f)

 [X]f : Constant

 Plot of IS+X vs. [S]f

 x-intercept: –[X]f

Standard addition experiment with constant total volume.

11
GIST Anal Chem 9/26/2016

4. Internal Standards
23

 Internal standard
 A known amount of a compound, different from analyte, that is added to the unknown.
 Signal from analyte is compared with signal from the internal standard to find out how much analyte is
present.
 Assumption: Relative response to analyte & standard remains constant over a range of concentrations.

 Where to use?
 Analyses in which the quantity of sample analyzed or the instrument response varies slightly from run to
run.
 Sample loss can occur during sample preparation steps prior to analysis: Add to the unknown prior to any
manipulations, then the ratio of standard to analyte remains constant because the same fraction of each is
lost in any operation.

Response factor, F
24

 Response factor: Relative response of a detector to analyte (X) & internal standard (S).
 If the detector responds equally to standard & analyte: F = 1.
 If the detector responds twice as much to analyte as to standard: F = 2.
 If the detector responds half as much to analyte as to standard: F = 0.5.

Area of analyte signal  Area of standard signal 

 F 
Concentration of analyte  Concentration of standard 

AX A 
F S 
[X]  [S] 
Chromatographic separation of
 [X] & [S]: The concentrations of analyte & standard after mixing. unknown (X) and internal standard (S).

12
GIST Anal Chem 9/26/2016

Example
: Using an internal standard
25

 Preliminary experiment:
 A solution containing 0.083 7 M X & 0.066 6 M S: AX = 423 & AS = 347.
 Analysis of the unknown:
 10.0 mL of 0.146 M S were added to 10.0 mL of unknown, & the mixture was diluted to 25.0 mL:
 AX = 533 & AS = 582.

 Unknown?
 Find the F first.

AX A  423  347 
 F  S   F  F = 0.9700
[X]  [S]  0.083 7  0.066 6 

Example
: Using an internal standard (cont.)
26

 Then, find the concentration of S

 10.0 
[S]  (0.146 M)   0.058 4 M
 25.0 
 If we substitute this value into the response factor equation;

AX A  533  582 
 F  S   0.0970 0   [X] = 0.057 21 M
[X]  [S]  [X]  0.058 4 
 Because X was diluted from 10.0 to 25.0 mL, the original concentration of X in the unknown:

 25 .0 
(0.057 21 M)   0.143 M
 10.0 

13