Chapter 5

Standardizing Analytical
Methods
Chapter Overview
5A Analytical Standards
5B Calibrating the Signal (Stotal)
5C Determining the Sensitivity (kA)
5D Linear Regression and Calibration Curves
5E Compensating for the Reagent Blank (Sreag)
5F Using Excel and R for a Regression Analysis
5G Key Terms
5H Chapter Summary
5I Problems
5J Solutions to Practice Exercises

The American Chemical Society’s Committee on Environmental Improvement defines

standardization as the process of determining the relationship between the signal and the
amount of analyte in a sample.1 In Chapter 3 we defined this relationship as
S total = k A n A + S reag or S total = k A C A
where Stotal is the signal, nA is the moles of analyte, CA is the analyte’s concentration, kA is the
method’s sensitivity for the analyte, and Sreag is the contribution to Stotal from sources other
than the sample. To standardize a method we must determine values for kA and Sreag. Strategies
for accomplishing this are the subject of this chapter.

1 ACS Committee on Environmental Improvement “Guidelines for Data Acquisition and Data Quality Evaluation in
Environmental Chemistry,” Anal. Chem. 1980, 52, 2242–2249.

147
148 Analytical Chemistry 2.1

5A Analytical Standards
To standardize an analytical method we use standards that contain known
amounts of analyte. The accuracy of a standardization, therefore, depends
on the quality of the reagents and the glassware we use to prepare these
standards. For example, in an acid–base titration the stoichiometry of the
acid–base reaction defines the relationship between the moles of analyte
and the moles of titrant. In turn, the moles of titrant is the product of the
See Chapter 9 for a thorough discussion of titrant’s concentration and the volume of titrant used to reach the equiva-
titrimetric methods of analysis.
lence point. The accuracy of a titrimetric analysis, therefore, is never better
than the accuracy with which we know the titrant’s concentration.

5A.1 Primary and Secondary Standards

There are two categories of analytical standards: primary standards and sec-
ondary standards. A primary standard is a reagent that we can use to
dispense an accurately known amount of analyte. For example, a 0.1250-g
sample of K2Cr2O7 contains 4.249 × 10–4 moles of K2Cr2O7. If we place
this sample in a 250-mL volumetric flask and dilute to volume, the concen-
tration of K2Cr2O7 in the resulting solution is 1.700 × 10–3 M. A primary
standard must have a known stoichiometry, a known purity (or assay), and
it must be stable during long-term storage. Because it is difficult to estab-
lishing accurately the degree of hydration, even after drying, a hydrated
reagent usually is not a primary standard.
NaOH is one example of a secondary Reagents that do not meet these criteria are secondary standards.
standard. Commercially available NaOH
contains impurities of NaCl, Na2CO3,
The concentration of a secondary standard is determined relative to a pri-
and Na2SO4, and readily absorbs H2O mary standard. Lists of acceptable primary standards are available.2 Appen-
from the atmosphere. To determine the dix 8 provides examples of some common primary standards.
concentration of NaOH in a solution, we
titrate it against a primary standard weak
acid, such as potassium hydrogen phthal-
5A.2 Other Reagents
ate, KHC8H4O4.
Preparing a standard often requires additional reagents that are not primary
standards or secondary standards, such as a suitable solvent or reagents
needed to adjust the standard’s matrix. These solvents and reagents are po-
tential sources of additional analyte, which, if not accounted for, produce
a determinate error in the standardization. If available, reagent grade
chemicals that conform to standards set by the American Chemical Society
are used.3 The label on the bottle of a reagent grade chemical (Figure 5.1)
lists either the limits for specific impurities or provides an assay for the
impurities. We can improve the quality of a reagent grade chemical by pu-
rifying it, or by conducting a more accurate assay. As discussed later in the
chapter, we can correct for contributions to Stotal from reagents used in an

2 (a) Smith, B. W.; Parsons, M. L. J. Chem. Educ. 1973, 50, 679–681; (b) Moody, J. R.; Green-
burg, P. R.; Pratt, K. W.; Rains, T. C. Anal. Chem. 1988, 60, 1203A–1218A.
3 Committee on Analytical Reagents, Reagent Chemicals, 8th ed., American Chemical Society:
Washington, D. C., 1993.
 Chapter 5 Standardizing Analytical Methods 149

analysis by including an appropriate blank determination in the analytical

procedure.

5A.3 Preparing a Standard Solution

It often is necessary to prepare a series of standards, each with a different
concentration of analyte. We can prepare these standards in two ways. If the
range of concentrations is limited to one or two orders of magnitude, then
each solution is best prepared by transferring a known mass or volume of
the pure standard to a volumetric flask and diluting to volume.
When working with a larger range of concentrations, particularly a
range that extends over more than three orders of magnitude, standards
are best prepared by a serial dilution from a single stock solution. In a
serial dilution we prepare the most concentrated standard and then dilute
a portion of that solution to prepare the next most concentrated standard.
Next, we dilute a portion of the second standard to prepare a third standard,
continuing this process until we have prepared all of our standards. Serial
dilutions must be prepared with extra care because an error in preparing
one standard is passed on to all succeeding standards.

(a) (b)

Figure 5.1 Two examples of packaging labels for reagent grade chemicals. The label in (a) pro-
vides the manufacturer’s assay for the reagent, NaBr. Note that potassium is flagged with an
asterisk (*) because its assay exceeds the limit established by the American Chemical Society
(ACS). The label in (b) does not provide an assay for impurities; however it indicates that the
reagent meets ACS specifications by providing the maximum limits for impurities. An assay for
the reagent, NaHCO3, is provided.
150 Analytical Chemistry 2.1

5B Calibrating the Signal (Stotal)

The accuracy with which we determine kA and Sreag depends on how accu-
rately we can measure the signal, Stotal. We measure signals using equipment,
such as glassware and balances, and instrumentation, such as spectropho-
tometers and pH meters. To minimize determinate errors that might affect
the signal, we first calibrate our equipment and instrumentation by measur-
ing Stotal for a standard with a known response of Sstd, adjusting Stotal until
S total = S std
Here are two examples of how we calibrate signals; other examples are pro-
vided in later chapters that focus on specific analytical methods.
When the signal is a measurement of mass, we determine Stotal using
an analytical balance. To calibrate the balance’s signal we use a reference
weight that meets standards established by a governing agency, such as the
National Institute for Standards and Technology or the American Society
See Section 2D.1 to review how an elec-
tronic balance works. Calibrating a bal- for Testing and Materials. An electronic balance often includes an internal
ance is important, but it does not elimi- calibration weight for routine calibrations, as well as programs for calibrat-
nate all sources of determinate error when ing with external weights. In either case, the balance automatically adjusts
measuring mass. See Appendix 9 for a
discussion of correcting for the buoyancy Stotal to match Sstd.
of air. We also must calibrate our instruments. For example, we can evaluate
a spectrophotometer’s accuracy by measuring the absorbance of a carefully
Be sure to read and follow carefully the prepared solution of 60.06 mg/L K2Cr2O7 in 0.0050 M H2SO4, using
calibration instructions provided with any 0.0050 M H2SO4 as a reagent blank.4 An absorbance of 0.640 ± 0.010
instrument you use.
absorbance units at a wavelength of 350.0 nm indicates that the spectrom-
eter’s signal is calibrated properly.

5C Determining the Sensitivity (kA)

To standardize an analytical method we also must determine the analyte’s
sensitivity, kA, in equation 5.1 or equation 5.2.
S total = k A n A + S reag 5.1
S total = k A C A + S reag 5.2
In principle, it is possible to derive the value of kA for any analytical method
if we understand fully all the chemical reactions and physical processes re-
sponsible for the signal. Unfortunately, such calculations are not feasible if
we lack a sufficiently developed theoretical model of the physical processes
or if the chemical reaction’s evince non-ideal behavior. In such situations we
must determine the value of kA by analyzing one or more standard solutions,
each of which contains a known amount of analyte. In this section we con-
sider several approaches for determining the value of kA. For simplicity we
assume that Sreag is accounted for by a proper reagent blank, allowing us to
replace Stotal in equation 5.1 and equation 5.2 with the analyte’s signal, SA.

4 Ebel, S. Fresenius J. Anal. Chem. 1992, 342, 769.

 Chapter 5 Standardizing Analytical Methods 151

Equation 5.3 and equation 5.4 essentially

SA = kAnA 5.3 are identical, differing only in whether we
choose to express the amount of analyte
SA = kACA 5.4 in moles or as a concentration. For the re-
mainder of this chapter we will limit our
5C.1 Single-Point versus Multiple-Point Standardizations treatment to equation 5.4. You can extend
this treatment to equation 5.3 by replac-
The simplest way to determine the value of kA in equation 5.4 is to use ing CA with nA.

a single-point standardization in which we measure the signal for a

standard, Sstd, that contains a known concentration of analyte, Cstd. Substi-
tuting these values into equation 5.4
kA = CS std
std
5.5
gives us the value for kA. Having determined kA, we can calculate the con-
centration of analyte in a sample by measuring its signal, Ssamp, and calculat-
ing CA using equation 5.6.
S samp
CA = 5.6
kA
A single-point standardization is the least desirable method for stan-
dardizing a method. There are two reasons for this. First, any error in our
determination of kA carries over into our calculation of CA. Second, our
experimental value for kA is based on a single concentration of analyte. To
extend this value of kA to other concentrations of analyte requires that
we assume a linear relationship between the signal and the analyte’s con-
centration, an assumption that often is not true.5 Figure 5.2 shows how
assuming a constant value of kA leads to a determinate error in CA if kA be-
comes smaller at higher concentrations of analyte. Despite these limitations,
single-point standardizations find routine use when the expected range for
the analyte’s concentrations is small. Under these conditions it often is safe
5 Cardone, M. J.; Palmero, P. J.; Sybrandt, L. B. Anal. Chem. 1980, 52, 1187–1191.

assumed relationship

actual relationship
Ssamp

Figure 5.2 Example showing how a single-point standard-

ization leads to a determinate error in an analyte’s reported
Sstd concentration if we incorrectly assume that kA is constant.
The assumed relationship between Ssamp and CA is based on
a single standard and is a straight-line; the actual relationship
Cstd (CA)reported (CA)actual between Ssamp and CA becomes curved for larger concentra-
tions of analyte.
152 Analytical Chemistry 2.1

to assume that kA is constant (although you should verify this assumption

experimentally). This is the case, for example, in clinical labs where many
automated analyzers use only a single standard.
The better way to standardize a method is to prepare a series of standards,
each of which contains a different concentration of analyte. Standards are
chosen such that they bracket the expected range for the analyte’s concen-
tration. A multiple-point standardization should include at least three
Linear regression, which also is known as standards, although more are preferable. A plot of Sstd versus Cstd is called a
the method of least squares, is one such al- calibration curve. The exact standardization, or calibration relationship,
gorithm. Its use is covered in Section 5D.
is determined by an appropriate curve-fitting algorithm.
There are two advantages to a multiple-point standardization. First, al-
though a determinate error in one standard introduces a determinate error,
its effect is minimized by the remaining standards. Second, because we
measure the signal for several concentrations of analyte, we no longer must
assume kA is independent of the analyte’s concentration. Instead, we can
construct a calibration curve similar to the “actual relationship” in Figure
5.2.

5C.2 External Standards

Appending the adjective “external” to
the noun “standard” might strike you as The most common method of standardization uses one or more external
odd at this point, as it seems reasonable
to assume that standards and samples are standards, each of which contains a known concentration of analyte. We
analyzed separately. As we will soon learn, call these standards “external” because they are prepared and analyzed sepa-
however, we can add standards to our rate from the samples.
samples and analyze both simultaneously.

Single External Standard

With a single external standard we determine kA using equation 5.5 and
then calculate the concentration of analyte, CA, using equation 5.6.

Example 5.1
A spectrophotometric method for the quantitative analysis of Pb2+ in
blood yields an Sstd of 0.474 for a single standard for which the concentra-
tion of lead is 1.75 ppb. What is the concentration of Pb2+ in a sample of
blood for which Ssamp is 0.361?
Solution
Equation 5.5 allows us to calculate the value of kA using the data for the
single external standard.
kA = CS std = 0.474 = 0.2709 ppm -1
std 1.75 ppb
Having determined the value of kA, we calculate the concentration of Pb2+
in the sample of blood is calculated using equation 5.6.
S samp 0.361
CA = = = 1.33 ppb
kA 0.2709 ppm -1
 Chapter 5 Standardizing Analytical Methods 153

Multiple External Standards

Figure 5.3 shows a typical multiple-point external standardization. The
volumetric flask on the left contains a reagent blank and the remaining
volumetric flasks contain increasing concentrations of Cu2+. Shown be-
low the volumetric flasks is the resulting calibration curve. Because this is
the most common method of standardization, the resulting relationship is
called a normal calibration curve.
When a calibration curve is a straight-line, as it is in Figure 5.3, the
slope of the line gives the value of kA. This is the most desirable situation
because the method’s sensitivity remains constant throughout the analyte’s
concentration range. When the calibration curve is not a straight-line, the
method’s sensitivity is a function of the analyte’s concentration. In Figure
5.2, for example, the value of kA is greatest when the analyte’s concentration
is small and it decreases continuously for higher concentrations of analyte.
The value of kA at any point along the calibration curve in Figure 5.2 is the
slope at that point. In either case, a calibration curve allows to relate Ssamp
to the analyte’s concentration.

Example 5.2
A second spectrophotometric method for the quantitative analysis of Pb2+
in blood has a normal calibration curve for which
S std = (0.296 ppb -1) # C std + 0.003
What is the concentration of Pb2+ in a sample of blood if Ssamp is 0.397?

0.25

0.20

0.15
Sstd Figure 5.3 The photo at the top of the figure shows
0.10 a reagent blank (far left) and a set of five external
standards for Cu2+ with concentrations that in-
0.05 crease from left-to-right. Shown below the external
0 standards is the resulting normal calibration curve.
0 0.0020 0.0040 0.0060 0.0080 The absorbance of each standard, Sstd, is shown by
Cstd (M) the filled circles.
154 Analytical Chemistry 2.1

Solution
To determine the concentration of Pb2+ in the sample of blood, we replace
Sstd in the calibration equation with Ssamp and solve for CA.
S samp - 0.003
CA = = 0.397 - 0.003 = 1.33 ppb
0.296 ppb -1 0.296 ppb -1
It is worth noting that the calibration equation in this problem includes
an extra term that does not appear in equation 5.6. Ideally we expect
our calibration curve to have a signal of zero when CA is zero. This is the
purpose of using a reagent blank to correct the measured signal. The extra
term of +0.003 in our calibration equation results from the uncertainty
in measuring the signal for the reagent blank and the standards.

Practice Exercise 5.1

Figure 5.3 shows a normal calibration curve for the quantitative analysis
of Cu2+. The equation for the calibration curve is
Sstd = 29.59 M–1 × Cstd + 0.0015
What is the concentration of Cu2+ in a sample whose absorbance, Ssamp,
The one-point standardization in this ex-
ercise uses data from the third volumetric is 0.114? Compare your answer to a one-point standardization where a
flask in Figure 5.3. standard of 3.16 × 10–3 M Cu2+ gives a signal of 0.0931.
Click here to review your answer to this exercise.

An external standardization allows us to analyze a series of samples

using a single calibration curve. This is an important advantage when we
have many samples to analyze. Not surprisingly, many of the most common
quantitative analytical methods use an external standardization.
The matrix for the external standards in
There is a serious limitation, however, to an external standardization.
Figure 5.3, for example, is dilute ammo- When we determine the value of kA using equation 5.5, the analyte is pres-
2+
nia. Because the Cu (NH 3) 4 complex ent in the external standard’s matrix, which usually is a much simpler ma-
absorbs more strongly than Cu2+, adding trix than that of our samples. When we use an external standardization we
ammonia increases the signal’s magnitude.
If we fail to add the same amount of am- assume the matrix does not affect the value of kA. If this is not true, then
monia to our samples, then we will in- we introduce a proportional determinate error into our analysis. This is not
troduce a proportional determinate error the case in Figure 5.4, for instance, where we show calibration curves for
into our analysis.
an analyte in the sample’s matrix and in the standard’s matrix. In this case,
using the calibration curve for the external standards leads to a negative de-
terminate error in analyte’s reported concentration. If we expect that matrix
effects are important, then we try to match the standard’s matrix to that of
the sample, a process known as matrix matching. If we are unsure of the
sample’s matrix, then we must show that matrix effects are negligible or use
an alternative method of standardization. Both approaches are discussed in
the following section.
 Chapter 5 Standardizing Analytical Methods 155

standard’s
matrix
sample’s
Ssamp
matrix

Figure 5.4 Calibration curves for an analyte in the

standard’s matrix and in the sample’s matrix. If the
matrix affects the value of kA, as is the case here, then
we introduce a proportional determinate error into
(CA)reported (CA)actual our analysis if we use a normal calibration curve.

5C.3 Standard Additions

We can avoid the complication of matching the matrix of the standards to
the matrix of the sample if we carry out the standardization in the sample.
This is known as the method of standard additions.

Single Standard Addition

The simplest version of a standard addition is shown in Figure 5.5. First we
add a portion of the sample, Vo, to a volumetric flask, dilute it to volume,
Vf, and measure its signal, Ssamp. Next, we add a second identical portion
of sample to an equivalent volumetric flask along with a spike, Vstd, of an
external standard whose concentration is Cstd. After we dilute the spiked
sample to the same final volume, we measure its signal, Sspike. The following
two equations relate Ssamp and Sspike to the concentration of analyte, CA, in
the original sample.
Vo
S samp = k A C A V
f
5.7 The ratios Vo/Vf and Vstd/Vf account for
the dilution of the sample and the stan-

S spike = k A a C A V
Vo + C Vstd k dard, respectively.
f
std
Vf 5.8
As long as Vstd is small relative to Vo, the effect of the standard’s matrix on
the sample’s matrix is insignificant. Under these conditions the value of kA
is the same in equation 5.7 and equation 5.8. Solving both equations for
kA and equating gives
S samp S spike
Vo = Vo + C Vstd
CA V CA V 5.9
f f
std
Vf
which we can solve for the concentration of analyte, CA, in the original
sample.
156 Analytical Chemistry 2.1

add Vo of CA add Vstd of Cstd

dilute to Vf
Figure 5.5 Illustration showing the method of stan-
dard additions. The volumetric flask on the left con-
tains a portion of the sample, Vo, and the volumetric
flask on the right contains an identical portion of the
sample and a spike, Vstd, of a standard solution of the
analyte. Both flasks are diluted to the same final vol-
ume, Vf. The concentration of analyte in each flask is
shown at the bottom of the figure where CA is the ana- Vo Vo V
Concentration
lyte’s concentration in the original sample and Cstd is CA × CA × + C std × std
the concentration of analyte in the external standard. of Analyte Vf Vf Vf

Example 5.3
A third spectrophotometric method for the quantitative analysis of Pb2+ in
blood yields an Ssamp of 0.193 when a 1.00 mL sample of blood is diluted
to 5.00 mL. A second 1.00 mL sample of blood is spiked with 1.00 mL of
a 1560-ppb Pb2+ external standard and diluted to 5.00 mL, yielding an
Sspike of 0.419. What is the concentration of Pb2+ in the original sample
of blood?
Solution
We begin by making appropriate substitutions into equation 5.9 and solv-
ing for CA. Note that all volumes must be in the same units; thus, we first
covert Vstd from 1.00 mL to 1.00 × 10–3 mL.
0.193 = 0.419
1 . 00 mL
C A 5.00mL C A 5.00mL + 1560 ppb 1.005.#
1 . 00 mL 10 -3 mL
00 mL
0.193 0.419
0.200C A = 0.200C A + 0.3120 ppb

0.0386C A + 0.0602 ppb = 0.0838C A

0.0452C A = 0.0602 ppb

C A = 1.33 ppb
The concentration of Pb2+ in the original sample of blood is 1.33 ppb.
 Chapter 5 Standardizing Analytical Methods 157

It also is possible to add the standard addition directly to the sample,

measuring the signal both before and after the spike (Figure 5.6). In this
case the final volume after the standard addition is Vo + Vstd and equation
5.7, equation 5.8, and equation 5.9 become
S samp = k A C A

S spike = k A a C A V +Vo
V + C std V +Vstd k
Vstd 5.10
o std o

S samp S spike
CA = V Vstd 5.11
C A V +o V + C std V + Vstd
o std o

Example 5.4
A fourth spectrophotometric method for the quantitative analysis of Pb2+
in blood yields an Ssamp of 0.712 for a 5.00 mL sample of blood. After spik-
ing the blood sample with 5.00 mL of a 1560-ppb Pb2+ external standard,
an Sspike of 1.546 is measured. What is the concentration of Pb2+ in the
original sample of blood?
Solution
To determine the concentration of Pb2+ in the original sample of blood,
we make appropriate substitutions into equation 5.11 and solve for CA.
0.712 = 1.546 Vo + Vstd = 5.000 mL + 5.00×10
–3
mL
CA
C A 55..005
00 mL + 1560 ppb 5.00 # 10 -3 mL
mL 5.005 mL = 5.005 mL

0.712 = 1.546
CA 0.9990C A + 1.558 ppb

add Vstd of Cstd

Vo Vo

Concentration Vo Vstd
CA CA + C std
of Analyte Vo + Vs td Vo + Vs td

Figure 5.6 Illustration showing an alternative form of the method of standard

additions. In this case we add the spike of external standard directly to the sample
without any further adjust in the volume.
158 Analytical Chemistry 2.1

0.7113C A + 1.109 ppb = 1.546C A

C A = 1.33 ppb
2+
The concentration of Pb in the original sample of blood is 1.33 ppb.

Multiple Standard Additions

We can adapt a single-point standard addition into a multiple-point stan-
dard addition by preparing a series of samples that contain increasing
amounts of the external standard. Figure 5.7 shows two ways to plot a
standard addition calibration curve based on equation 5.8. In Figure 5.7a
we plot Sspike against the volume of the spikes, Vstd. If kA is constant, then
the calibration curve is a straight-line. It is easy to show that the x-intercept
is equivalent to –CAVo/Cstd.

Example 5.5
Beginning with equation 5.8 show that the equations in Figure 5.7a for
the slope, the y-intercept, and the x-intercept are correct.
Solution
We begin by rewriting equation 5.8 as

S spike = k A C A Vo k A C std
V f + V f # Vstd
which is in the form of the equation for a straight-line
y = y-intercept + slope × x
where y is Sspike and x is Vstd. The slope of the line, therefore, is kACstd/Vf
and the y-intercept is kACAVo/Vf. The x-intercept is the value of x when y
is zero, or

0 = kAC A Vo k A Vstd
V f + V f # x-intercept

k A C A Vo V f
x-intercept =- =- CCA Vo
k A C std V f std

Practice Exercise 5.2

Beginning with equation 5.8 show that the equations in Figure 5.7b for
the slope, the y-intercept, and the x-intercept are correct.
Click here to review your answer to this exercise.
Because we know the volume of the original sample, Vo, and the con-
centration of the external standard, Cstd, we can calculate the analyte’s con-
centrations from the x-intercept of a multiple-point standard additions.
 Chapter 5 Standardizing Analytical Methods 159

(a) 0.60
kACAVo
0.50 y-intercept =
Vf
0.40
Sspike 0.30 kACstd
slope =
Vf
0.20

0.10
0
-2.00 0 2.00 4.00 6.00
Figure 5.7 Shown at the top of the
Vstd (mL)
-C V figure is a set of six standard additions
x-intercept = A o
Cstd
for the determination of Mn2+. The
(b) 0.60
flask on the left is a 25.00 mL sample
kACAVo
0.50 y-intercept =
Vf
diluted to 50.00 mL with water. The
0.40
remaining flasks contain 25.00 mL of
sample and, from left-to-right, 1.00,
Sspike 0.30 slope = kA 2.00, 3.00, 4.00, and 5.00 mL spikes
0.20 of an external standard that is 100.6
mg/L Mn2+. Shown below are two
0.10
ways to plot the standard additions
0 calibration curve. The absorbance for
-4.00 -2.00 0 2.00 4.00 6.00 8.00 10.00 12.00
each standard addition, Sspike, is shown
-CAVo Vstd
x-intercept = Cstd × (mg/L) by the filled circles.
Vf Vf

Example 5.6
A fifth spectrophotometric method for the quantitative analysis of Pb2+
in blood uses a multiple-point standard addition based on equation 5.8.
The original blood sample has a volume of 1.00 mL and the standard used
for spiking the sample has a concentration of 1560 ppb Pb2+. All samples
were diluted to 5.00 mL before measuring the signal. A calibration curve
of Sspike versus Vstd has the following equation
S spike = 0.266 + 312 mL-1 # Vstd
What is the concentration of Pb2+ in the original sample of blood?
Solution
To find the x-intercept we set Sspike equal to zero.
0 = 0.266 + 312 mL-1 # Vstd
160 Analytical Chemistry 2.1

Solving for Vstd, we obtain a value of –8.526 × 10–4 mL for the x-intercept.
Substituting the x-intercept’s value into the equation from Figure 5.7a
- 8.526 # 10 -4 mL =- CCA Vo =- C A # 1.00 mL
std 1560 ppb
and solving for CA gives the concentration of Pb2+ in the blood sample as
1.33 ppb.

Practice Exercise 5.3

Figure 5.7 shows a standard additions calibration curve for the quantita-
tive analysis of Mn2+. Each solution contains 25.00 mL of the original
sample and either 0, 1.00, 2.00, 3.00, 4.00, or 5.00 mL of a 100.6 mg/L
external standard of Mn2+. All standard addition samples were diluted to
50.00 mL with water before reading the absorbance. The equation for the
calibration curve in Figure 5.7a is

Sstd = 0.0854 × Vstd + 0.1478

What is the concentration of Mn2+ in this sample? Compare your answer
to the data in Figure 5.7b, for which the calibration curve is

Sstd = 0.0425 × Cstd(Vstd/Vf) + 0.1478

Click here to review your answer to this exercise.
Since we construct a standard additions calibration curve in the sample,
we can not use the calibration equation for other samples. Each sample,
therefore, requires its own standard additions calibration curve. This is a
serious drawback if you have many samples. For example, suppose you need
to analyze 10 samples using a five-point calibration curve. For a normal
calibration curve you need to analyze only 15 solutions (five standards and
ten samples). If you use the method of standard additions, however, you
must analyze 50 solutions (each of the ten samples is analyzed five times,
once before spiking and after each of four spikes).

Using a Standard Addition to Identify Matrix Effects

We can use the method of standard additions to validate an external stan-
dardization when matrix matching is not feasible. First, we prepare a nor-
mal calibration curve of Sstd versus Cstd and determine the value of kA from
its slope. Next, we prepare a standard additions calibration curve using
equation 5.8, plotting the data as shown in Figure 5.7b. The slope of this
standard additions calibration curve provides an independent determina-
tion of kA. If there is no significant difference between the two values of
kA, then we can ignore the difference between the sample’s matrix and that
of the external standards. When the values of kA are significantly different,
 Chapter 5 Standardizing Analytical Methods 161

then using a normal calibration curve introduces a proportional determi-

nate error.

5C.4 Internal Standards

To use an external standardization or the method of standard additions, we
must be able to treat identically all samples and standards. When this is not
possible, the accuracy and precision of our standardization may suffer. For
example, if our analyte is in a volatile solvent, then its concentration will
increase if we lose solvent to evaporation. Suppose we have a sample and a
standard with identical concentrations of analyte and identical signals. If
both experience the same proportional loss of solvent, then their respective
concentrations of analyte and signals remain identical. In effect, we can ig-
nore evaporation if the samples and the standards experience an equivalent
loss of solvent. If an identical standard and sample lose different amounts
of solvent, however, then their respective concentrations and signals are
no longer equal. In this case a simple external standardization or standard
addition is not possible.
We can still complete a standardization if we reference the analyte’s
signal to a signal from another species that we add to all samples and stan-
dards. The species, which we call an internal standard, must be different
than the analyte.
Because the analyte and the internal standard receive the same treat-
ment, the ratio of their signals is unaffected by any lack of reproducibility in
the procedure. If a solution contains an analyte of concentration CA and an
internal standard of concentration CIS, then the signals due to the analyte,
SA, and the internal standard, SIS, are
SA = kACA

S IS = k IS C IS
where kA and kIS are the sensitivities for the analyte and the internal stan-
dard, respectively. Taking the ratio of the two signals gives the fundamental
equation for an internal standardization.
SA kACA CA
S IS = k IS C IS = K # C IS 5.12
Because K is a ratio of the analyte’s sensitivity and the internal standard’s
sensitivity, it is not necessary to determine independently values for either
kA or kIS.

Single Internal Standard

In a single-point internal standardization, we prepare a single standard that
contains the analyte and the internal standard, and use it to determine the
value of K in equation 5.12.
162 Analytical Chemistry 2.1

K =aC C A kstd # a S IS kstd

IS SA
5.13
Having standardized the method, the analyte’s concentration is given by
C A = CKIS # a SS A k
IS samp

Example 5.7
A sixth spectrophotometric method for the quantitative analysis of Pb2+
in blood uses Cu2+ as an internal standard. A standard that is 1.75 ppb
Pb2+ and 2.25 ppb Cu2+ yields a ratio of (SA/SIS)std of 2.37. A sample of
blood spiked with the same concentration of Cu2+ gives a signal ratio,
(SA/SIS)samp, of 1.80. What is the concentration of Pb2+ in the sample of
blood?
Solution
Equation 5.13 allows us to calculate the value of K using the data for the
standard

K =aC k a S A k = 2.25 ppb Cu # 2.37 = 3.05 ppb Cu

2+ 2+
IS
C A std # S IS std 1.75 ppb Pb 2+
ppb Pb 2+

The concentration of Pb2+, therefore, is

C A = CKIS # a SS A k =
2.25 ppb Cu 2+
# 1.80 = 1.33 ppb Pb 2+
IS samp ppb Cu 2+
3.05
ppb Pb 2+

Multiple Internal Standards

A single-point internal standardization has the same limitations as a single-
point normal calibration. To construct an internal standard calibration
Although the usual practice is to prepare
the standards so that each contains an
curve we prepare a series of standards, each of which contains the same
identical amount of the internal standard, concentration of internal standard and a different concentrations of analyte.
this is not a requirement. Under these conditions a calibration curve of (SA/SIS)std versus CA is linear
with a slope of K/CIS.

Example 5.8
A seventh spectrophotometric method for the quantitative analysis of Pb2+
in blood gives a linear internal standards calibration curve for which
a SS A k = (2.11 ppb -1) # C A - 0.006
IS std

What is the ppb Pb2+ in a sample of blood if (SA/SIS)samp is 2.80?

Solution
To determine the concentration of Pb2+ in the sample of blood we replace
(SA/SIS)std in the calibration equation with (SA/SIS)samp and solve for CA.
 Chapter 5 Standardizing Analytical Methods 163

a SS A k + 0.006
= 2.80 + 0.006
IS samp
CA = = 1.33 Pb 2+
2.11 ppb -1 2.11 ppb -1
The concentration of Pb2+ in the sample of blood is 1.33 ppb.
You might wonder if it is possible to in-
clude an internal standard in the method
In some circumstances it is not possible to prepare the standards so of standard additions to correct for both
that each contains the same concentration of internal standard. This is the matrix effects and uncontrolled variations
case, for example, when we prepare samples by mass instead of volume. We between samples; well, the answer is yes
as described in the paper “Standard Dilu-
can still prepare a calibration curve, however, by plotting (SA/SIS)std versus tion Analysis,” the full reference for which
CA/CIS, giving a linear calibration curve with a slope of K. is Jones, W. B.; Donati, G. L.; Calloway,
C. P.; Jones, B. T. Anal. Chem. 2015, 87,
2321-2327.
5D Linear Regression and Calibration Curves
In a single-point external standardization we determine the value of kA
by measuring the signal for a single standard that contains a known con-
centration of analyte. Using this value of kA and our sample’s signal, we
then calculate the concentration of analyte in our sample (see Example
5.1). With only a single determination of kA, a quantitative analysis using
a single-point external standardization is straightforward.
A multiple-point standardization presents a more difficult problem.
Consider the data in Table 5.1 for a multiple-point external standardiza-
tion. What is our best estimate of the relationship between Sstd and Cstd? It
is tempting to treat this data as five separate single-point standardizations,
determining kA for each standard, and reporting the mean value for the
five trials. Despite it simplicity, this is not an appropriate way to treat a
multiple-point standardization.
So why is it inappropriate to calculate an average value for kA using
the data in Table 5.1? In a single-point standardization we assume that the
reagent blank (the first row in Table 5.1) corrects for all constant sources
of determinate error. If this is not the case, then the value of kA from a
single-point standardization has a constant determinate error. Table 5.2
demonstrates how an uncorrected constant error affects our determination

Table 5.1 Data for a Hypothetical Multiple-Point External

Standardization
Cstd (arbitrary units) Sstd (arbitrary units) kA = Sstd/ Cstd
0.000 0.00 —
0.100 12.36 123.6
0.200 24.83 124.2
0.300 35.91 119.7
0.400 48.79 122.0
0.500 60.42 122.8
mean value for kA = 122.5
164 Analytical Chemistry 2.1

Table 5.2 Effect of a Constant Determinate Error on the Value of kA From a Single-
Point Standardization
Sstd kA = Sstd/ Cstd (Sstd)e kA = (Sstd)e/ Cstd
Cstd (without constant error) (actual) (with constant error) (apparent)
1.00 1.00 1.00 1.50 1.50
2.00 2.00 1.00 2.50 1.25
3.00 3.00 1.00 3.50 1.17
4.00 4.00 1.00 4.50 1.13
5.00 5.00 1.00 5.50 1.10
mean kA (true) = 1.00 mean kA (apparent) = 1.23

of kA. The first three columns show the concentration of analyte in a set of
standards, Cstd, the signal without any source of constant error, Sstd, and
the actual value of kA for five standards. As we expect, the value of kA is the
same for each standard. In the fourth column we add a constant determi-
nate error of +0.50 to the signals, (Sstd)e. The last column contains the cor-
responding apparent values of kA. Note that we obtain a different value of
kA for each standard and that each apparent kA is greater than the true value.
How do we find the best estimate for the relationship between the sig-
nal and the concentration of analyte in a multiple-point standardization?
Figure 5.8 shows the data in Table 5.1 plotted as a normal calibration curve.
Although the data certainly appear to fall along a straight line, the actual
calibration curve is not intuitively obvious. The process of determining the
best equation for the calibration curve is called linear regression.

5D.1 Linear Regression of Straight Line Calibration Curves

When a calibration curve is a straight-line, we represent it using the follow-
ing mathematical equation
y = b0 + b1 x 5.14
where y is the analyte’s signal, Sstd, and x is the analyte’s concentration, Cstd.
The constants b0 and b1 are, respectively, the calibration curve’s expected
y-intercept and its expected slope. Because of uncertainty in our measure-
ments, the best we can do is to estimate values for b0 and b1, which we
represent as b0 and b1. The goal of a linear regression analysis is to de-
termine the best estimates for b0 and b1. How we do this depends on the
uncertainty in our measurements.

5D.2 Unweighted Linear Regression with Errors in y

The most common method for completing the linear regression for equa-
tion 5.14 makes three assumptions:
 Chapter 5 Standardizing Analytical Methods 165

60

50

40

Sstd 30

20

10

0
0.0 0.1 0.2 0.3 0.4 0.5
Cstd
Figure 5.8 Normal calibration curve data for the hypothetical multiple-point
external standardization in Table 5.1.

(1) that the difference between our experimental data and the calculated
regression line is the result of indeterminate errors that affect y,
(2) that indeterminate errors that affect y are normally distributed, and
(3) that the indeterminate errors in y are independent of the value of x.
Because we assume that the indeterminate errors are the same for all stan-
dards, each standard contributes equally in our estimate of the slope and
the y-intercept. For this reason the result is considered an unweighted
linear regression.
The second assumption generally is true because of the central limit the-
orem, which we considered in Chapter 4. The validity of the two remaining
assumptions is less obvious and you should evaluate them before you accept
the results of a linear regression. In particular the first assumption always is
suspect because there certainly is some indeterminate error in the measure-
ment of x. When we prepare a calibration curve, however, it is not unusual
to find that the uncertainty in the signal, Sstd, is significantly larger than the
uncertainty in the analyte’s concentration, Cstd. In such circumstances the
first assumption is usually reasonable.

How a Linear Regression Works

To understand the logic of a linear regression consider the example shown
in Figure 5.9, which shows three data points and two possible straight-lines
that might reasonably explain the data. How do we decide how well these
straight-lines fit the data, and how do we determine the best straight-line?
Let’s focus on the solid line in Figure 5.9. The equation for this line is
Vy = b 0 + b 1 x 5.15
166 Analytical Chemistry 2.1

Figure 5.9 Illustration showing three data points and two

possible straight-lines that might explain the data. The goal
of a linear regression is to find the mathematical model, in
this case a straight-line, that best explains the data.

nounce T
where b0 and b1 are estimates for the y-intercept and the slope, and Vy is the
If you are reading this aloud, you pro-
y as y-hat.
predicted value of y for any value of x. Because we assume that all uncer-
tainty is the result of indeterminate errors in y, the difference between y and
Vy for each value of x is the residual error, r, in our mathematical model.
ri = (y i - Vy i)
The reason for squaring the individual
residual errors is to prevent a positive re-
Figure 5.10 shows the residual errors for the three data points. The smaller
sidual error from canceling out a negative the total residual error, R, which we define as
R = / (y i - Vy i) 2
residual error. You have seen this before in n
the equations for the sample and popula- 5.16
tion standard deviations. You also can see i=1

from this equation why a linear regression the better the fit between the straight-line and the data. In a linear regres-
is sometimes called the method of least
squares. sion analysis, we seek values of b0 and b1 that give the smallest total residual
error.

ŷ3 ŷ = b0 + b1 x
y2
r3 = ( y3 − yˆ3 )
r2 = ( y 2 − yˆ2 )
y3
ŷ1 ŷ 2

r1 = ( y1 − yˆ1 )
y1
Figure 5.10 Illustration showing the evaluation of a linear regression in which we assume that all un-
certainty is the result of indeterminate errors in y. The points in blue, yi, are the original data and the
points in red, Vy i , are the predicted values from the regression equation, Vy = b 0 + b 1 x .The smaller
the total residual error (equation 5.16), the better the fit of the straight-line to the data.
 Chapter 5 Standardizing Analytical Methods 167

Finding the Slope and y-Intercept

Although we will not formally develop the mathematical equations for a
linear regression analysis, you can find the derivations in many standard
statistical texts.6 The resulting equation for the slope, b1, is
n n n
n / xi yi - / xi / yi
b1 = i=1 i=1 i=1
5.17
n / x - c/ xi m
n n 2
2
i
i=1 i=1

and the equation for the y-intercept, b0, is

n n
/ y -b /x i
5.18
1 i

b0 =n
i=1 i=1

Although equation 5.17 and equation 5.18 appear formidable, it is neces-

sary only to evaluate the following four summations
n n n n
/x i /y i /x y i i /x 2
i
i=1 i=1 i=1 i=1

Many calculators, spreadsheets, and other statistical software packages are

capable of performing a linear regression analysis based on this model. To See Section 5F in this chapter for details
on completing a linear regression analysis
save time and to avoid tedious calculations, learn how to use one of these using Excel and R.
tools. For illustrative purposes the necessary calculations are shown in detail
in the following example.

Example 5.9
Using the data from Table 5.1, determine the relationship between Sstd and
Cstd using an unweighted linear regression.
Solution
We begin by setting up a table to help us organize the calculation.
xi yi xiyi xi2
0.000 0.00 0.000 0.000
Equations 5.17 and 5.18 are written in
0.100 12.36 1.236 0.010 terms of the general variables x and y. As
0.200 24.83 4.966 0.040 you work through this example, remem-
ber that x corresponds to Cstd, and that y
0.300 35.91 10.773 0.090 corresponds to Sstd.
0.400 48.79 19.516 0.160
0.500 60.42 30.210 0.250
Adding the values in each column gives
n n n n
/x i = 1.500 /y i = 182.31 /x y i i = 66.701 /x 2
i = 0.550
i=1 i=1 i=1 i=1
Substituting these values into equation 5.17 and equation 5.18, we find
that the slope and the y-intercept are
6 See, for example, Draper, N. R.; Smith, H. Applied Regression Analysis, 3rd ed.; Wiley: New
York, 1998.
168 Analytical Chemistry 2.1

(6 # 66.701) - (1.500 # 182.31)

b1 = = 120.706 . 120.71
(6 # 0.550) - (1.500) 2
182.31 - (120.706 # 1.500)
b1 = = 0.209 . 0.21
6
The relationship between the signal and the analyte, therefore, is
Sstd = 120.71 × Cstd + 0.21
For now we keep two decimal places to match the number of decimal
places in the signal. The resulting calibration curve is shown in Figure 5.11.

Uncertainty in the Regression Analysis

As shown in Figure 5.11, because indeterminate errors in the signal, the
regression line may not pass through the exact center of each data point.
The cumulative deviation of our data from the regression line—that is, the
total residual error—is proportional to the uncertainty in the regression.
We call this uncertainty the standard deviation about the regression,
Did you notice the similarity between the
standard deviation about the regression
sr, which is equal to

/ (y - Vy )
(equation 5.19) and the standard devia- n
2
tion for a sample (equation 4.1)? i i
5.19
n
sr = n-2
i=1

where yi is the i experimental value, and Vy i is the corresponding value pre-

/ (X i - X)
i=1 th
s=
n-1
dicted by the regression line in equation 5.15. Note that the denominator
of equation 5.19 indicates that our regression analysis has n–2 degrees of
freedom—we lose two degree of freedom because we use two parameters,
the slope and the y-intercept, to calculate Vy i .

60

50

40

Sstd 30

20

10

0
0.0 0.1 0.2 0.3 0.4 0.5
Cstd
Figure 5.11 Calibration curve for the data in Table 5.1 and Example 5.9.
 Chapter 5 Standardizing Analytical Methods 169

A more useful representation of the uncertainty in our regression analy-

sis is to consider the effect of indeterminate errors on the slope, b1, and the
y-intercept, b0, which we express as standard deviations.

sb = ns r2 = s 2r
n / x i2 - c / x i m / ^x - x h
n n n
1 2
2 5.20
i
i=1 i=1 i=1

n n
s 2r / x 2i s 2r / x 2i
sb = i=1
= i=1
5.21
n / x - c/ xi m n / ^ x i - x h2
0 n n 2 n
2
i
i=1 i=1 i=1

We use these standard deviations to establish confidence intervals for the

expected slope, b1, and the expected y-intercept, b0
b 1 = b 1 ! ts b 1 5.22

b 0 = b 0 ! ts b 0 5.23 You might contrast equation 5.22 and

equation 5.23 with equation 4.12
where we select t for a significance level of a and for n–2 degrees of free- ts
n = X!
dom. Note that equation 5.22 and equation 5.23 do not contain a factor of
^ n h because the confidence interval is based on a single regression line.
-1
n
for the confidence interval around a sam-
ple’s mean value.
Example 5.10
Calculate the 95% confidence intervals for the slope and y-intercept from
Example 5.9.
Solution
We begin by calculating the standard deviation about the regression. To do
this we must calculate the predicted signals, Vy i , using the slope and y‑in-
As you work through this example, re-

tercept from Example 5.9, and the squares of the residual error, _ y i - Vy i i .
2
member that x corresponds to Cstd, and
that y corresponds to Sstd.
Using the last standard as an example, we find that the predicted signal is
Vy 6 = b 0 + b 1 x 6 = 0.209 + (120.706 # 0.500) = 60.562
and that the square of the residual error is
_ y i - Vy i i = (60.42 - 60.562) 2 = 0.2016 . 0.202
2

The following table displays the results for all six solutions.
Vy i _ y i - Vy i i
2
xi yi
0.000 0.00 0.209 0.0437
0.100 12.36 12.280 0.0064
0.200 24.83 24.350 0.2304
0.300 35.91 36.421 0.2611
0.400 48.79 48.491 0.0894
0.500 60.42 60.562 0.0202
170 Analytical Chemistry 2.1

Adding together the data in the last column gives the numerator of equa-
tion 5.19 as 0.6512; thus, the standard deviation about the regression is
0.6512 = 0.4035
sr =
6-2
Next we calculate the standard deviations for the slope and the y-intercept
using equation 5.20 and equation 5.21. The values for the summation
terms are from in Example 5.9.
ns r2 6 # (0.4035) 2
sb = = = 0.965
n / x i2 - c / x i m
n n
1 2
(6 # 0.550) - (1.500) 2
i=1 i=1

n
s 2r / x 2i
(0.4035) 2 # 0.550
sb = i=1
= = 0.292
n / x i2 - c / x i m
n n
0 2
(6 # 0.550) - (1.500) 2
i=1 i=1

You can find values for t in Appendix 4. Finally, the 95% confidence intervals (a = 0.05, 4 degrees of freedom) for
the slope and y-intercept are
b 1 = b 1 ! ts b = 120.706 ! (2.78 # 0.965) = 120.7 ! 2.7
1

b 0 = b 0 ! ts b = 0.209 ! (2.78 # 0.292) = 0.2 ! 0.8

0

The standard deviation about the regression, sr, suggests that the signal, Sstd,
is precise to one decimal place. For this reason we report the slope and the
y-intercept to a single decimal place.

Minimizing Uncertainty in Calibration Curves

To minimize the uncertainty in a calibration curve’s slope and y-intercept,
we evenly space our standards over a wide range of analyte concentrations.
A close examination of equation 5.20 and equation 5.21 help us appreci-
ate why this is true. The denominators of both equations include the term
/ ^x i - x h2 . The larger the value of this term—which we accomplish by
increasing the range of x around its mean value—the smaller the standard
deviations in the slope and the y-intercept. Furthermore, to minimize the
uncertainty in the y‑intercept, it helps to decrease the value of the term
/ x i in equation 5.21, which we accomplish by including standards for
lower concentrations of the analyte.

Obtaining the Analyte’s Concentration From a Regression Equation

Once we have our regression equation, it is easy to determine the concen-
tration of analyte in a sample. When we use a normal calibration curve,
for example, we measure the signal for our sample, Ssamp, and calculate the
analyte’s concentration, CA, using the regression equation.
 Chapter 5 Standardizing Analytical Methods 171

S samp - b 0
CA = 5.24
b1
What is less obvious is how to report a confidence interval for CA that
expresses the uncertainty in our analysis. To calculate a confidence interval
we need to know the standard deviation in the analyte’s concentration, s C , A Equation 5.25 is written in terms of a cali-
which is given by the following equation bration experiment. A more general form
of the equation, written in terms of x and
^ S samp - S std h2 y, is given here.
sC = sr 1 1
m+n+
(b 1) 2 / ^C std - C std h2
b1 n
5.25
^Y - y h
A
2
i
sr 1 1
^ x i - x h2
i=1 sx = + + n
b1 m n 2 /
(b 1)
where m is the number of replicate we use to establish the sample’s average i=1
signal, S samp , n is the number of calibration standards, S std is the average
signal for the calibration standards, and C std and C std are the individual and
i A close examination of equation 5.25
the mean concentrations for the calibration standards.7 Knowing the value should convince you that the uncertainty
in CA is smallest when the sample’s av-
of   s C , the confidence interval for the analyte’s concentration is
A
erage signal, S samp , is equal to the aver-
n C = C A ! ts C
A A
age signal for the standards, S std . When
practical, you should plan your calibration
where nCA is the expected value of CA in the absence of determinate errors, curve so that Ssamp falls in the middle of
and with the value of t is based on the desired level of confidence and n–2 the calibration curve.
degrees of freedom.

Example 5.11
Three replicate analyses for a sample that contains an unknown concentra-
tion of analyte, yield values for Ssamp of 29.32, 29.16 and 29.51 (arbitrary
units). Using the results from Example 5.9 and Example 5.10, determine
the analyte’s concentration, CA, and its 95% confidence interval.
Solution
The average signal, S samp , is 29.33, which, using equation 5.24 and the
slope and the y-intercept from Example 5.9, gives the analyte’s concentra-
tion as
S samp - b 0
CA = = 29.33 - 0.209 = 0.241
b1 120.706
To calculate the standard deviation for the analyte’s concentration we must
determine the values for S std and for / ^C std - C std h2 . The former is just
i

the average signal for the calibration standards, which, using the data in
Table 5.1, is 30.385. Calculating / ^C std - C std h2 looks formidable, but
i

we can simplify its calculation by recognizing that this sum-of-squares is

the numerator in a standard deviation equation; thus,
n
/ (C std i - C std ) 2 = (s C ) 2 # (n - 1)
std
i=1

7 (a) Miller, J. N. Analyst 1991, 116, 3–14; (b) Sharaf, M. A.; Illman, D. L.; Kowalski, B. R. Che-
mometrics, Wiley-Interscience: New York, 1986, pp. 126-127; (c) Analytical Methods Commit-
tee “Uncertainties in concentrations estimated from calibration experiments,” AMC Technical
Brief, March 2006.
172 Analytical Chemistry 2.1

60

50

40

Sstd 30

Figure 5.12 Example of a normal calibration curve with 20

a superimposed confidence interval for the analyte’s con-
centration. The points in blue are the original data from 10
Table 5.1. The black line is the normal calibration curve
as determined in Example 5.9. The red lines show the
0
95% confidence interval for CA assuming a single deter-
mination of Ssamp. 0.0 0.1 0.2 0.3 0.4 0.5
Cstd

where   s C is the standard deviation for the concentration of analyte in

std

the calibration standards. Using the data in Table 5.1 we find that s C is std

0.1871 and

/ ^C - C std h2 = (0.1872) 2 # ^6 - 1h = 0.175

n

You can find values for t in Appendix 4. std i

i=1

Substituting known values into equation 5.25 gives

1 + 1 + (29.33 - 30.385) = 0.0024
2

s C = 0.4035 3
A
120.706 6 (120.706) 2 # 0.175
Finally, the 95% confidence interval for 4 degrees of freedom is
n C = C A ! ts C = 0.241 ! (2.78 # 0.0024) = 0.241 ! 0.007
A A

Figure 5.12 shows the calibration curve with curves showing the 95%
confidence interval for CA.

In a standard addition we determine the analyte’s concentration by

extrapolating the calibration curve to the x-intercept. In this case the value
of CA is

C A = x-intercept = - b 0
b1
and the standard deviation in CA is
^ S std h2
sC = sr 1+
(b 1) 2 / ^C std - C std h2
A
b1 n n

i
i=1

where n is the number of standard additions (including the sample with no

added standard), and S std is the average signal for the n standards. Because
we determine the analyte’s concentration by extrapolation, rather than by
 Chapter 5 Standardizing Analytical Methods 173

Practice Exercise 5.4

Figure 5.3 shows a normal calibration curve for the quantitative analysis
of Cu2+. The data for the calibration curve are shown here.
[Cu2+] (M) Absorbance
0 0
1.55×10–3 0.050
3.16×10–3 0.093
4.74×10–3 0.143
6.34×10–3 0.188
7.92×10–3 0.236

Complete a linear regression analysis for this calibration data, reporting

the calibration equation and the 95% confidence interval for the slope
and the y-intercept. If three replicate samples give an Ssamp of 0.114, what
is the concentration of analyte in the sample and its 95% confidence
interval?
Click here to review your answer to this exercise.

interpolation, s C for the method of standard additions generally is larger

A

than for a normal calibration curve.

Evaluating a Linear Regression Model

You should never accept the result of a linear regression analysis without
evaluating the validity of the model. Perhaps the simplest way to evaluate
a regression analysis is to examine the residual errors. As we saw earlier, the
residual error for a single calibration standard, ri, is
r i = (y i - y i )
If the regression model is valid, then the residual errors should be distrib-
uted randomly about an average residual error of zero, with no apparent
trend toward either smaller or larger residual errors (Figure 5.13a). Trends
such as those in Figure 5.13b and Figure 5.13c provide evidence that at least
one of the model’s assumptions is incorrect. For example, a trend toward
larger residual errors at higher concentrations, Figure 5.13b, suggests that
the indeterminate errors affecting the signal are not independent of the
analyte’s concentration. In Figure 5.13c, the residual errors are not random,
which suggests we cannot model the data using a straight-line relationship.
Regression methods for the latter two cases are discussed in the following
sections.
174 Analytical Chemistry 2.1

(a) (b) (c)

residual error
residual error
residual error

0.0 0.1 0.2 0.3 0.4 0.5 0.0 0.1 0.2 0.3 0.4 0.5 0.0 0.1 0.2 0.3 0.4 0.5
Cstd Cstd Cstd
Figure 5.13 Plots of the residual error in the signal, Sstd, as a function of the concentration of analyte, Cstd, for an
unweighted straight-line regression model. The red line shows a residual error of zero. The distribution of the residual
errors in (a) indicates that the unweighted linear regression model is appropriate. The increase in the residual errors in
(b) for higher concentrations of analyte, suggests that a weighted straight-line regression is more appropriate. For (c),
the curved pattern to the residuals suggests that a straight-line model is inappropriate; linear regression using a quadratic
model might produce a better fit.
Practice Exercise 5.5
Using your results from Practice Exercise 5.4, construct a residual plot
and explain its significance.
Click here to review your answer to this exercise.

5D.3 Weighted Linear Regression with Errors in y

Our treatment of linear regression to this point assumes that indeterminate
errors affecting y are independent of the value of x. If this assumption is
false, as is the case for the data in Figure 5.13b, then we must include the
variance for each value of y into our determination of the y-intercept, bo,
and the slope, b1; thus
n n
/w y -b /w x
i i 1 i i
5.26
b0 = i=1
n
i=1

n n n
n / wi xi yi - / wi xi / wi yi
b1 = i=1 i=1 i=1
5.27
n / wi x - c/ wi xi m
n n 2
2
i
i=1 i=1

where wi is a weighting factor that accounts for the variance in yi

n ^s y h-2
wi =
i

/ ^s h-2
n
5.28
yi
i=1

and s y is the standard deviation for yi. In a weighted linear regression,

i

each xy-pair’s contribution to the regression line is inversely proportional

to the precision of yi; that is, the more precise the value of y, the greater its
contribution to the regression.
 Chapter 5 Standardizing Analytical Methods 175

Example 5.12
Shown here are data for an external standardization in which sstd is the
standard deviation for three replicate determination of the signal.
Cstd (arbitrary units) Sstd (arbitrary units) sstd
0.000 0.00 0.02
0.100 12.36 0.02 This is the same data used in Example 5.9
0.200 24.83 0.07 with additional information about the
standard deviations in the signal.
0.300 35.91 0.13
0.400 48.79 0.22
0.500 60.42 0.33
As you work through this example, re-
member that x corresponds to Cstd, and
Determine the calibration curve’s equation using a weighted linear regres- that y corresponds to Sstd.
sion.
Solution
We begin by setting up a table to aid in calculating the weighting factors.

xi yi sy i ^s y h-2
i wi
0.000 0.00 0.02 2500.00 2.8339
As a check on your calculations, the sum
0.100 12.36 0.02 2500.00 2.8339 of the individual weights must equal the
0.200 24.83 0.07 204.08 0.2313 number of calibration standards, n. The
sum of the entries in the last column is
0.300 35.91 0.13 59.17 0.0671 6.0000, so all is well.
0.400 48.79 0.22 20.66 0.0234
0.500 60.42 0.33 9.18 0.0104
Adding together the values in the forth column gives

/ ^s h-2
n

yi
i=1

which we use to calculate the individual weights in the last column. After
we calculate the individual weights, we use a second table to aid in calculat-
ing the four summation terms in equation 5.26 and equation 5.27.
xi yi wi wi xi wi yi wi xi2 wi xi yi
0.000 0.00 2.8339 0.0000 0.0000 0.0000 0.0000
0.100 12.36 2.8339 0.2834 35.0270 0.0283 3.5027
0.200 24.83 0.2313 0.0463 5.7432 0.0093 1.1486
0.300 35.91 0.0671 0.0201 2.4096 0.0060 0.7229
0.400 48.79 0.0234 0.0094 1.1417 0.0037 0.4567
0.500 60.42 0.0104 0.0052 0.6284 0.0026 0.3142
Adding the values in the last four columns gives
176 Analytical Chemistry 2.1

n n
/w x i i = 0.3644 /w y i i = 44.9499
i=1 i=1

n n
/w x i
2
i = 0.0499 /w x yi i i = 6.1451
i=1 i=1

Substituting these values into the equation 5.26 and equation 5.27 gives
the estimated slope and estimated y-intercept as
(6 # 6.1451) - (0.3644 # 44.9499)
b1 = = 122.985
(6 # 0.0499) - (0.3644) 2
44.9499 - (122.985 # 0.3644)
b0 = = 0.0224
6
The calibration equation is
S std = 122.98 # C std + 0.02
Figure 5.14 shows the calibration curve for the weighted regression and the
calibration curve for the unweighted regression in Example 5.9. Although
the two calibration curves are very similar, there are slight differences in the
slope and in the y-intercept. Most notably, the y-intercept for the weighted
linear regression is closer to the expected value of zero. Because the stan-
dard deviation for the signal, Sstd, is smaller for smaller concentrations of
analyte, Cstd, a weighted linear regression gives more emphasis to these
standards, allowing for a better estimate of the y-intercept.

60 weighted linear regression

unweighted linear regression
50

40

Sstd 30

20

10

0
0.0 0.1 0.2 0.3 0.4 0.5
Cstd

Figure 5.14 A comparison of the unweighted and the weighted normal calibra-
tion curves. See Example 5.9 for details of the unweighted linear regression and
Example 5.12 for details of the weighted linear regression.
 Chapter 5 Standardizing Analytical Methods 177

Equations for calculating confidence intervals for the slope, the y-in-
tercept, and the concentration of analyte when using a weighted linear
regression are not as easy to define as for an unweighted linear regression.8
The confidence interval for the analyte’s concentration, however, is at its
optimum value when the analyte’s signal is near the weighted centroid, y c ,
of the calibration curve.
n
1 /w x
yc = n i i
i=1

5D.4 Weighted Linear Regression with Errors in Both x and y

If we remove our assumption that indeterminate errors affecting a calibra-
tion curve are present only in the signal (y), then we also must factor into
the regression model the indeterminate errors that affect the analyte’s con-
centration in the calibration standards (x). The solution for the resulting
regression line is computationally more involved than that for either the
unweighted or weighted regression lines.9 Although we will not consider
the details in this textbook, you should be aware that neglecting the pres-
ence of indeterminate errors in x can bias the results of a linear regression.

5D.5 Curvilinear and Multivariate Regression

A straight-line regression model, despite its apparent complexity, is the
simplest functional relationship between two variables. What do we do if See Figure 5.2 for an example of a calibra-
tion curve that deviates from a straight-
our calibration curve is curvilinear—that is, if it is a curved-line instead of line for higher concentrations of analyte.
a straight-line? One approach is to try transforming the data into a straight-
line. Logarithms, exponentials, reciprocals, square roots, and trigonometric It is worth noting that the term “linear”
functions have been used in this way. A plot of log(y) versus x is a typical does not mean a straight-line. A linear
function may contain more than one ad-
example. Such transformations are not without complications, of which ditive term, but each such term has one
the most obvious is that data with a uniform variance in y will not maintain and only one adjustable multiplicative
that uniform variance after it is transformed. parameter. The function
Another approach to developing a linear regression model is to fit a y = ax + bx2
polynomial equation to the data, such as y = a + bx + cx2. You can use is an example of a linear function because
2
the terms x and x each include a single
linear regression to calculate the parameters a, b, and c, although the equa- multiplicative parameter, a and b, respec-
tions are different than those for the linear regression of a straight-line.10 tively. The function
If you cannot fit your data using a single polynomial equation, it may be y = xb
possible to fit separate polynomial equations to short segments of the cali- is nonlinear because b is not a multiplica-
bration curve. The result is a single continuous calibration curve known as tive parameter; it is, instead, a power. This
is why you can use linear regression to fit a
a spline function. polynomial equation to your data.
Sometimes it is possible to transform a
nonlinear function into a linear function.
8 Bonate, P. J. Anal. Chem. 1993, 65, 1367–1372. For example, taking the log of both sides
9 See, for example, Analytical Methods Committee, “Fitting a linear functional relationship to of the nonlinear function above gives a
data with error on both variable,” AMC Technical Brief, March, 2002), as well as this chapter’s linear function.
Additional Resources. log(y) = blog(x)
10 For details about curvilinear regression, see (a) Sharaf, M. A.; Illman, D. L.; Kowalski, B. R.
Chemometrics, Wiley-Interscience: New York, 1986; (b) Deming, S. N.; Morgan, S. L. Experi-
mental Design: A Chemometric Approach, Elsevier: Amsterdam, 1987.
178 Analytical Chemistry 2.1

The regression models in this chapter apply only to functions that con-
tain a single independent variable, such as a signal that depends upon the
analyte’s concentration. In the presence of an interferent, however, the signal
may depend on the concentrations of both the analyte and the interferent
S = k A C A + k I C I + S reag
Check out this chapter’s Additional Re- where kI is the interferent’s sensitivity and CI is the interferent’s concentra-
sources at the end of the textbook for
more information about linear regression
tion. Multivariate calibration curves are prepared using standards that con-
with errors in both variables, curvilinear tain known amounts of both the analyte and the interferent, and modeled
regression, and multivariate regression. using multivariate regression.11

5E Compensating for the Reagent Blank (Sreag)

Thus far in our discussion of strategies for standardizing analytical methods,
we have assumed that a suitable reagent blank is available to correct for sig-
nals arising from sources other than the analyte. We did not, however ask
an important question: “What constitutes an appropriate reagent blank?”
Surprisingly, the answer is not immediately obvious.
In one study, approximately 200 analytical chemists were asked to
evaluate a data set consisting of a normal calibration curve, a separate ana-
lyte-free blank, and three samples with different sizes, but drawn from the
same source.12 The first two columns in Table 5.3 shows a series of external
standards and their corresponding signals. The normal calibration curve
for the data is
Sstd = 0.0750 × Wstd + 0.1250
where the y-intercept of 0.1250 is the calibration blank. A separate reagent
blank gives the signal for an analyte-free sample.

11 Beebe, K. R.; Kowalski, B. R. Anal. Chem. 1987, 59, 1007A–1017A.

12 Cardone, M. J. Anal. Chem. 1986, 58, 433–438.

Table 5.3 Data Used to Study the Blank in an Analytical Method

Wstd Sstd Sample Number Wsamp Ssamp
1.6667 0.2500 1 62.4746 0.8000
5.0000 0.5000 2 82.7915 1.0000
8.3333 0.7500 3 103.1085 1.2000
11.6667 0.8413
18.1600 1.4870 reagent blank 0.1000
19.9333 1.6200
Calibration equation: Sstd = 0.0750 × Wstd + 0.1250
Wstd: weight of analyte used to prepare the external standard; diluted to volume, V.
Wsamp: weight of sample used to prepare sample; diluted to volume, V.
 Chapter 5 Standardizing Analytical Methods 179

Table 5.4 Equations and Resulting Concentrations of Analyte for Different Approaches
to Correcting for the Blank
Concentration of Analyte in...
Approach for Correcting The Signal Equation Sample 1 Sample 2 Sample 3
W A = S samp
CA = W
ignore calibration and reagent blank samp k A W samp 0.1707 0.1610 0.1552

W A = S samp - CB
CA = W
use calibration blank only samp k A W samp 0.1441 0.1409 0.1390

W A = S samp - RB
CA = W
use reagent blank only samp k A W samp 0.1494 0.1449 0.1422

W A = S samp - CB - RB
use both calibration and reagent blank C A = W samp k A W samp 0.1227 0.1248 0.1261

W A = S samp - TYB
CA = W
use total Youden blank samp k A W samp 0.1313 0.1313 0.1313

CA = concentration of analyte; WA = weight of analyte; Wsamp = weight of sample; kA = slope of calibration curve (0.075; see Table
5.3); CB = calibration blank (0.125; see Table 5.3); RB = reagent blank (0.100; see Table 5.3); TYB = total Youden blank (0.185; see
text)

In working up this data, the analytical chemists used at least four dif-
ferent approaches to correct the signals: (a) ignoring both the calibration
blank, CB, and the reagent blank, RB, which clearly is incorrect; (b) using
the calibration blank only; (c) using the reagent blank only; and (d) using
both the calibration blank and the reagent blank. The first four rows of
Table 5.4 shows the equations for calculating the analyte’s concentration
using each approach, along with the reported concentrations for the analyte
in each sample.
That all four methods give a different result for the analyte’s concentra-
tion underscores the importance of choosing a proper blank, but does not
tell us which blank is correct. Because all four methods fail to predict the
same concentration of analyte for each sample, none of these blank correc-
tions properly accounts for an underlying constant source of determinate
error.
To correct for a constant method error, a blank must account for sig-
nals from any reagents and solvents used in the analysis and any bias that
results from interactions between the analyte and the sample’s matrix. Both
the calibration blank and the reagent blank compensate for signals from Because we are considering a matrix effect
reagents and solvents. Any difference in their values is due to indeterminate of sorts, you might think that the method
of standard additions is one way to over-
errors in preparing and analyzing the standards. come this problem. Although the method
Unfortunately, neither a calibration blank nor a reagent blank can cor- of standard additions can compensate for
rect for a bias that results from an interaction between the analyte and the proportional determinate errors, it cannot
correct for a constant determinate error;
sample’s matrix. To be effective, the blank must include both the sample’s see Ellison, S. L. R.; Thompson, M. T.
matrix and the analyte and, consequently, it must be determined using the “Standard additions: myth and reality,”
sample itself. One approach is to measure the signal for samples of differ- Analyst, 2008, 133, 992–997.
180 Analytical Chemistry 2.1

ent size, and to determine the regression line for a plot of Ssamp versus the
amount of sample. The resulting y-intercept gives the signal in the absence
of sample, and is known as the total Youden blank.13 This is the true
blank correction. The regression line for the three samples in Table 5.3 is
Ssamp = 0.009844 × Wsamp + 0.185
giving a true blank correction of 0.185. As shown by the last row of Table
5.4, using this value to correct Ssamp gives identical values for the concentra-
tion of analyte in all three samples.
The use of the total Youden blank is not common in analytical work,
with most chemists relying on a calibration blank when using a calibra-
tion curve and a reagent blank when using a single-point standardization.
As long we can ignore any constant bias due to interactions between the
analyte and the sample’s matrix, which is often the case, the accuracy of an
analytical method will not suffer. It is a good idea, however, to check for
constant sources of error before relying on either a calibration blank or a
reagent blank.

5F Using Excel and R for a Regression Analysis

Although the calculations in this chapter are relatively straightforward—
consisting, as they do, mostly of summations—it is tedious to work through
problems using nothing more than a calculator. Both Excel and R include
functions for completing a linear regression analysis and for visually evalu-
ating the resulting model.

5F.1 Excel
Let’s use Excel to fit the following straight-line model to the data in Ex-
ample 5.9.
y = b0 + b1 x
Enter the data into a spreadsheet, as shown in Figure 5.15. Depending
upon your needs, there are many ways that you can use Excel to complete
A B
a linear regression analysis. We will consider three approaches here.
1 Cstd Sstd
2 0.000 0.00 Use Excel’s Built-In Functions
3 0.100 12.36 If all you need are values for the slope, b1, and the y-intercept, b0, you can
4 0.200 24.83 use the following functions:
5 0.300 35.91 = intercept(known_y’s, known_x’s)
6 0.400 48.79
7 0.500 60.42 = slope(known_y’s, known_x’s)
Figure 5.15 Portion of a spread-
sheet containing data from Exam-
ple 5.9 (Cstd = Cstd; Sstd = Sstd).
13 Cardone, M. J. Anal. Chem. 1986, 58, 438–445.
 Chapter 5 Standardizing Analytical Methods 181

where known_y’s is the range of cells that contain the signals (y), and
known_x’s is the range of cells that contain the concentrations (x). For ex-
ample, if you click on an empty cell and enter
= slope(B2:B7, A2:A7)
Excel returns exact calculation for the slope (120.705 714 3).

Use Excel’s Data Analysis Tools

To obtain the slope and the y-intercept, along with additional statistical Excel’s Data Analysis Toolpak is available
for Windows. Older versions of Excel
details, you can use the data analysis tools in the Data Analysis ToolPak. for Mac included the toolpak; however,
The ToolPak is not a standard part of Excel’s instillation. To see if you have beginning with Excel for Mac 2011, the
access to the Analysis ToolPak on your computer, select Tools from the toolpak no longer is available.
menu bar and look for the Data Analysis... option. If you do not see Data Once you install the Analysis ToolPak, it
Analysis..., select Add-ins... from the Tools menu. Check the box for the will continue to load each time you launch
Excel.
Analysis ToolPak and click on OK to install them.
Select Data Analysis... from the Tools menu, which opens the Data
Analysis window. Scroll through the window, select Regression from the
available options, and press OK. Place the cursor in the box for Input Y
range and then click and drag over cells B1:B7. Place the cursor in the box
for Input X range and click and drag over cells A1:A7. Because cells A1 and Including labels is a good idea. Excel’s
B1 contain labels, check the box for Labels. Select the radio button for summary output uses the x-axis label to
identify the slope.
Output range and click on any empty cell; this is where Excel will place the
results. Clicking OK generates the information shown in Figure 5.16.
There are three parts to Excel’s summary of a regression analysis. At the
top of Figure 5.16 is a table of Regression Statistics. The standard error is the
standard deviation about the regression, sr. Also of interest is the value for
Multiple R, which is the model’s correlation coefficient, r, a term with which
you may already be familiar. The correlation coefficient is a measure of the
extent to which the regression model explains the variation in y. Values of r

SUMMARY OUTPUT

Regression Statistics
Multiple R 0.99987244
R Square 0.9997449
Adjusted R Square 0.99968113
Standard Error 0.40329713
Observations 6

ANOVA
df SS MS F Significance F
Regression 1 2549.727156 2549.72716 15676.296 2.4405E-08
Residual 4 0.650594286 0.16264857
Total 5 2550.37775

Coefficients Standard Error t Stat P-value Lower 95% Upper 95% Lower 95.0% Upper 95.0%
Intercept 0.20857143 0.29188503 0.71456706 0.51436267 -0.60183133 1.01897419 -0.60183133 1.01897419
Cstd 120.705714 0.964064525 125.205016 2.4405E-08 118.029042 123.382387 118.029042 123.382387

Figure 5.16 Output from Excel’s Regression command in the Analysis ToolPak. See the text for a discussion of how to
interpret the information in these tables.
182 Analytical Chemistry 2.1

10 range from –1 to +1. The closer the correlation coefficient is to ±1, the bet-
r = 0.993
8 ter the model is at explaining the data. A correlation coefficient of 0 means
there is no relationship between x and y. In developing the calculations for
6
y
linear regression, we did not consider the correlation coefficient. There is
4 a reason for this. For most straight-line calibration curves the correlation
2
coefficient is very close to +1, typically 0.99 or better. There is a tendency,
however, to put too much faith in the correlation coefficient’s significance,
0
and to assume that an r greater than 0.99 means the linear regression model
0 2 4 6 8 10
x is appropriate. Figure 5.17 provides a useful counterexample. Although
Figure 5.17 Example of fitting a the regression line has a correlation coefficient of 0.993, the data clearly is
straight-line (in red) to curvilinear curvilinear. The take-home lesson here is simple: do not fall in love with
data (in blue). the correlation coefficient!
The second table in Figure 5.16 is entitled ANOVA, which stands for
analysis of variance. We will take a closer look at ANOVA in Chapter 14.
For now, it is sufficient to understand that this part of Excel’s summary
provides information on whether the linear regression model explains a
significant portion of the variation in the values of y. The value for F is the
result of an F-test of the following null and alternative hypotheses.
See Section 4F.2 and Section 4F.3 for a H0: the regression model does not explain the variation in y
review of the F-test.
HA: the regression model does explain the variation in y
The value in the column for Significance F is the probability for retaining
the null hypothesis. In this example, the probability is 2.5×10–6%, which
is strong evidence for accepting the regression model. As is the case with
the correlation coefficient, a small value for the probability is a likely out-
come for any calibration curve, even when the model is inappropriate. The
probability for retaining the null hypothesis for the data in Figure 5.17, for
example, is 9.0×10–7%.
The third table in Figure 5.16 provides a summary of the model itself.
The values for the model’s coefficients—the slope, b1, and the y-intercept,
b0—are identified as intercept and with your label for the x-axis data, which
in this example is Cstd. The standard deviations for the coefficients, sb0 and
sb1, are in the column labeled Standard error. The column t Stat and the
column P-value are for the following t-tests.
slope H0: b1 = 0, HA: b1 ≠ 0
See Section 4F.1 for a review of the t-test.
y-intercept H0: b0 = 0, HA: b0 ≠ 0
The results of these t-tests provide convincing evidence that the slope is not
zero, but there is no evidence that the y-intercept differs significantly from
zero. Also shown are the 95% confidence intervals for the slope and the
y-intercept (lower 95% and upper 95%).
 Chapter 5 Standardizing Analytical Methods 183

A B C D E F
1 x y xy x^2 n= 6
2 0.000 0.00 =A2*B2 =A2^2 slope = =(F1*C8 - A8*B8)/(F1*D8-A8^2)
3 0.100 12.36 =A3*B3 =A3^2 y-int = =(B8-F2*A8)/F1
4 0.200 24.83 =A4*B4 =A4^2
5 0.300 35.91 =A5*B5 =A5^2
6 0.400 48.79 =A6*B6 =A6^2
7 0.500 60.42 =A7*B7 =A7^2
8
9 =sum(A2:A7) =sum(B2:B7) =sum(C2:C7) =sum(D2:D7) <--sums
Figure 5.18 Spreadsheet showing the formulas for calculating the slope and the y-intercept for the data in Example 5.9.
The shaded cells contain formulas that you must enter. Enter the formulas in cells C3 to C7, and cells D3 to D7. Next,
enter the formulas for cells A9 to D9. Finally, enter the formulas in cells F2 and F3. When you enter a formula, Excel
replaces it with the resulting calculation. The values in these cells should agree with the results in Example 5.9. You can
simplify the entering of formulas by copying and pasting. For example, enter the formula in cell C2. Select Edit: Copy,
click and drag your cursor over cells C3 to C7, and select Edit: Paste. Excel automatically updates the cell referencing.

Program the Formulas Yourself

A third approach to completing a regression analysis is to program a spread-
sheet using Excel’s built-in formula for a summation
=sum(first cell:last cell)
and its ability to parse mathematical equations. The resulting spreadsheet
is shown in Figure 5.18.

Using Excel to Visualize the Regression Model

You can use Excel to examine your data and the regression line. Begin by
plotting the data. Organize your data in two columns, placing the x values
in the left-most column. Click and drag over the data and select Charts
from the ribbon. Select Scatter, choosing the option without lines that
connect the points. To add a regression line to the chart, click on the chart’s
data and select Chart: Add Trendline... from the main men. Pick the
straight-line model and click OK to add the line to your chart. By default,
Excel displays the regression line from your first point to your last point.
Figure 5.19 shows the result for the data in Figure 5.15.
Excel also will create a plot of the regression model’s residual errors. To
create the plot, build the regression model using the Analysis ToolPak, as
described earlier. Clicking on the option for Residual plots creates the plot
shown in Figure 5.20.

Limitations to Using Excel for a Regression Analysis

Excel’s biggest limitation for a regression analysis is that it does not provide
a function to calculate the uncertainty when predicting values of x. In terms
of this chapter, Excel can not calculate the uncertainty for the analyte’s
184 Analytical Chemistry 2.1

70

60

50

40

y-axis
y-axis
30

20

10

0
Figure 5.19 Example of an Excel scatterplot showing
0 0.1 0.2 0.3 0.4 0.5 0.6
the data and a regression line. x-axis

Practice Exercise 5.6 concentration, CA, given the signal for a sample, Ssamp. Another limitation
is that Excel does not have a built-in function for a weighted linear regres-
Use Excel to complete the sion. You can, however, program a spreadsheet to handle these calculations.
regression analysis in Practice
Exercise 5.4. 5F.2 R
Click here to review your an- Let’s use R to fit the following straight-line model to the data in Example
swer to this exercise. 5.9.
y = b0 + b1 x

Entering Data and Creating the Regression Model

To begin, create objects that contain the concentration of the standards and
their corresponding signals.
> conc = c(0, 0.1, 0.2, 0.3, 0.4, 0.5)
> signal = c(0, 12.36, 24.83, 35.91, 48.79, 60.42)

0.6

0.4

0.2
Residuals

0
0 0.1 0.2 0.3 0.4 0.5 0.6
Residuals
-0.2

-0.4

Figure 5.20 Example of Excel’s plot of a re- -0.6

gression model’s residual errors. Cstd

 Chapter 5 Standardizing Analytical Methods 185

The command for a straight-line linear regression model is

As you might guess, lm is short for linear
lm(y ~ x) model.

where y and x are the objects the objects our data. To access the results of
the regression analysis, we assign them to an object using the following
command
You can choose any name for the object
> model = lm(signal ~ conc) that contains the results of the regression
analysis.
where model is the name we assign to the object.

Evaluating the Linear Regression Model

To evaluate the results of a linear regression we need to examine the data
and the regression line, and to review a statistical summary of the model. To
examine our data and the regression line, we use the plot command, which
takes the following general form
plot(x, y, optional arguments to control style)
where x and y are the objects that contain our data, and the abline com-
mand The name abline comes from the follow-
ing common form for writing the equa-
abline(object, optional arguments to control style) tion of a straight-line.
y = a + bx
where object is the object that contains the results of the linear regression. where a is the y-intercept and b is the
Entering the commands slope.
> plot(conc, signal, pch = 19, col = “blue”, cex = 2)
> abline(model, col = “red”)
creates the plot shown in Figure 5.21.
To review a statistical summary of the regression model, we use the
summary command.
> summary(model)
60
50
40

Figure 5.21 Example of a regression plot in R showing the data (in

signal

blue)and the regression line (in red). You can customize your plot
30

by adjusting the plot command’s optional arguments. For example,

the argument pch controls the symbol used for plotting points, the
20

argument col allows you to select a color for the points or the line,
10

and the argument cex sets the size for the points. You can use the
command
0

0.0 0.1 0.2 0.3 0.4 0.5 help(plot)

conc to learn more about the options for plotting data in R.
186 Analytical Chemistry 2.1

> model=lm(signal~conc)
> summary(model)

Call:
lm(formula = signal ~ conc)

Residuals:
1 2 3 4 5 6
-0.20857 0.08086 0.48029 -0.51029 0.29914 -0.14143

Coefficients:
Estimate Std. Error t value Pr(>|t|)
(Intercept) 0.2086 0.2919 0.715 0.514
conc 120.7057 0.9641 125.205 2.44e-08 ***
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
Figure 5.22 The summary of R’s regression analysis. See the
Residual standard error: 0.4033 on 4 degrees of freedom
text for a discussion of how to interpret the information in the Multiple R-Squared: 0.9997, Adjusted R-squared: 0.9997
output’s three sections. F-statistic: 1.568e+04 on 1 and 4 DF, p-value: 2.441e-08

The reason for including the argument The resulting output, shown in Figure 5.22, contains three sections.
which = 1 is not immediately obvious. The first section of R’s summary of the regression model lists the re-
When you use R’s plot command on an sidual errors. To examine a plot of the residual errors, use the command
object created by the lm command, the
default is to create four charts summa- > plot(model, which = 1)
rizing the model’s suitability. The first
of these charts is the residual plot; thus, which produces the result shown in Figure 5.23. Note that R plots the re-
which = 1 limits the output to this plot. siduals against the predicted (fitted) values of y instead of against the known
values of x. The choice of how to plot the residuals is not critical, as you can
see by comparing Figure 5.23 to Figure 5.20. The line in Figure 5.23 is a
smoothed fit of the residuals.
The second section of Figure 5.22 provides the model’s coefficients—
the slope, b1, and the y-intercept, b0—along with their respective standard
deviations (Std. Error). The column t value and the column Pr(>|t|) are for
the following t-tests.

Residuals vs Fitted
0.6

3
0.4

5
0.2
Residuals
0.0
-0.2
-0.4

4
-0.6

Figure 5.23 Example showing R’s plot of a regression model’s

residual error. 0 10 20 30 40 50 60

Fitted values
lm(signal ~ conc)
 Chapter 5 Standardizing Analytical Methods 187

slope H0: b1 = 0, HA: b1 ≠ 0

See Section 4F.1 for a review of the t-test.
y-intercept H0: b0 = 0, HA: b0 ≠ 0
The results of these t-tests provide convincing evidence that the slope is not
zero, but no evidence that the y-intercept differs significantly from zero.
The last section of the regression summary provides the standard devia-
tion about the regression (residual standard error), the square of the cor- See Section 4F.2 and Section 4F.3 for a
relation coefficient (multiple R-squared), and the result of an F-test on the review of the F-test.
model’s ability to explain the variation in the y values. For a discussion of
the correlation coefficient and the F-test of a regression model, as well as
their limitations, refer to the section on using Excel’s data analysis tools.

Predicting the Uncertainty in CA Given Ssamp

Unlike Excel, R includes a command for predicting the uncertainty in an
analyte’s concentration, CA, given the signal for a sample, Ssamp. This com-
mand is not part of R’s standard installation. To use the command you need You need to install a package once, but
to install the “chemCal” package by entering the following command (note: you need to load the package each time
you will need an internet connection to download the package). you plan to use it. There are ways to con-
figure R so that it automatically loads
> [Link](“chemCal”) certain packages; see An Introduction to R
for more information (click here to view a
After installing the package, you need to load the functions into R using the PDF version of this document).
following command. (note: you will need to do this step each time you begin
a new R session as the package does not automatically load when you start R).
> library(“chemCal”)
The command for predicting the uncertainty in CA is [Link],
which takes the following form for an unweighted linear regression
[Link](object, newdata, alpha = value)
where object is the object that contains the regression model’s results, new-
data is an object that contains values for Ssamp, and value is the numerical
value for the significance level. Let’s use this command to complete Ex-
ample 5.11. First, we create an object that contains the values of Ssamp
> sample = c(29.32, 29.16, 29.51)
and then we complete the computation using the following command
> [Link](model, sample, alpha = 0.05)
producing the result shown in Figure 5.24. The analyte’s concentration, CA,
is given by the value $Prediction, and its standard deviation, sCA, is shown
as $`Standard Error`. The value for $Confidence is the confidence interval,
±tsCA, for the analyte’s concentration, and $`Confidence Limits` provides
the lower limit and upper limit for the confidence interval for CA.
188 Analytical Chemistry 2.1

> [Link](model, sample, alpha = 0.05)

$Prediction
[1] 0.2412597

$`Standard Error`
[1] 0.002363588

$Confidence
[1] 0.006562373

$`Confidence Limits`
[1] 0.2346974 0.2478221
Figure 5.24 Output from R’s command for predicting the ana-
lyte’s concentration, CA, from the sample’s signal, Ssamp.

Using R for a Weighted Linear Regression

R’s command for an unweighted linear regression also allows for a weighted
linear regression if we include an additional argument, weights, whose value
You may have noticed that this way of
is an object that contains the weights.
defining weights is different than that lm(y ~ x, weights  = object)
shown in equation 5.28. In deriving equa-
tions for a weighted linear regression, you Let’s use this command to complete Example 5.12. First, we need to create
can choose to normalize the sum of the
weights to equal the number of points, or
an object that contains the weights, which in R are the reciprocals of the
you can choose not to—the algorithm in standard deviations in y, (syi)–2. Using the data from Example 5.12, we enter
R does not normalize the weights.
> syi=c(0.02, 0.02, 0.07, 0.13, 0.22, 0.33)
> w=1/syi^2
to create the object that contains the weights. The commands
> modelw = lm(signal ~ conc, weights = w)
> summary(modelw)
generate the output shown in Figure 5.25. Any difference between the
results shown here and the results shown in Example 5.12 are the result of
round-off errors in our earlier calculations.
Practice Exercise 5.7
Use Excel to complete the regression analysis in Practice Exercise 5.4.
Click here to review your answer to this exercise.
 Chapter 5 Standardizing Analytical Methods 189

> modelw=lm(signal~conc, weights = w)

> summary(modelw)

Call:
lm(formula = signal ~ conc, weights = w)

Residuals:
1 2 3 4 5 6
-2.223 2.571 3.676 -7.129 -1.413 -2.864

Coefficients:
Estimate Std. Error t value Pr(>|t|)
(Intercept) 0.04446 0.08542 0.52 0.63
conc 122.64111 0.93590 131.04 2.03e-08 ***
---
Figure 5.25 The summary of R’s regression analysis for
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
a weighted linear regression. The types of information
Residual standard error: 4.639 on 4 degrees of freedom shown here is identical to that for the unweighted linear
Multiple R-Squared: 0.9998, Adjusted R-squared: 0.9997 regression in Figure 5.22.
F-statistic: 1.717e+04 on 1 and 4 DF, p-value: 2.034e-08

5G Key Terms
calibration curve external standard internal standard
method of standard
linear regression matrix matching
additions
multiple-point
normal calibration curve primary standard
standardization
reagent grade residual error secondary standard
single-point standard deviation about
serial dilution
standardization the regression
unweighted linear
total Youden blank weighted linear regression
regression

5H Chapter Summary
In a quantitative analysis we measure a signal, Stotal, and calculate the
amount of analyte, nA or CA, using one of the following equations.
S total = k A n A + S reag
S total = k A C A + S reag
To obtain an accurate result we must eliminate determinate errors that af-
fect the signal, Stotal, the method’s sensitivity, kA, and the signal due to the
reagents, Sreag.
To ensure that we accurately measure Stotal, we calibrate our equipment
and instruments. To calibrate a balance, for example, we use a standard
weight of known mass. The manufacturer of an instrument usually suggests
appropriate calibration standards and calibration methods.
To standardize an analytical method we determine its sensitivity. There
are several standardization strategies available to us, including external
standards, the method of standard addition, and internal standards. The
190 Analytical Chemistry 2.1

most common strategy is a multiple-point external standardization and

a normal calibration curve. We use the method of standard additions, in
which we add known amounts of analyte to the sample, when the sample’s
matrix complicates the analysis. When it is difficult to reproducibly handle
samples and standards, we may choose to add an internal standard.
Single-point standardizations are common, but are subject to greater
uncertainty. Whenever possible, a multiple-point standardization is pre-
ferred, with results displayed as a calibration curve. A linear regression
analysis provides an equation for the standardization.
A reagent blank corrects for any contribution to the signal from the
reagents used in the analysis. The most common reagent blank is one in
which an analyte-free sample is taken through the analysis. When a simple
reagent blank does not compensate for all constant sources of determinate
error, other types of blanks, such as the total Youden blank, are used.

5I Problems
1. Suppose you use a serial dilution to prepare 100 mL each of a series of
standards with concentrations of 1.00×10–5, 1.00×10–4, 1.00×10–3,
and 1.00×10–2 M from a 0.100 M stock solution. Calculate the uncer-
tainty for each solution using a propagation of uncertainty, and com-
pare to the uncertainty if you prepare each solution as a single dilution
of the stock solution. You will find tolerances for different types of
volumetric glassware and digital pipets in Table 4.2 and Table 4.3. As-
sume that the uncertainty in the stock solution’s molarity is ±0.0002.

2. Three replicate determinations of Stotal for a standard solution that is

10.0 ppm in analyte give values of 0.163, 0.157, and 0.161 (arbitrary
units). The signal for the reagent blank is 0.002. Calculate the concen-
tration of analyte in a sample with a signal of 0.118.

3. A 10.00-g sample that contains an analyte is transferred to a 250-mL

volumetric flask and diluted to volume. When a 10.00 mL aliquot of
the resulting solution is diluted to 25.00 mL it gives a signal of 0.235
(arbitrary units). A second 10.00-mL portion of the solution is spiked
with 10.00 mL of a 1.00-ppm standard solution of the analyte and di-
luted to 25.00 mL. The signal for the spiked sample is 0.502. Calculate
the weight percent of analyte in the original sample.

4. A 50.00 mL sample that contains an analyte gives a signal of 11.5 (arbi-

trary units). A second 50 mL aliquot of the sample, which is spiked with
1.00 mL of a 10.0-ppm standard solution of the analyte, gives a signal
of 23.1. What is the analyte’s concentration in the original sample?
 Chapter 5 Standardizing Analytical Methods 191

5. A standard additions calibration curve based on equation 5.10 places

Sspike×(Vo + Vstd) on the y-axis and Cstd × Vstd on the x-axis. Derive
equations for the slope and the y-intercept and explain how you can
determine the amount of analyte in a sample from the calibration curve.
In addition, clearly explain why you cannot plot Sspike on the y-axis and
C std # " Vstd / (Vo + Vstd ) , on the x-axis.

6. A standard sample contains 10.0 mg/L of analyte and 15.0 mg/L of in-
ternal standard. Analysis of the sample gives signals for the analyte and
the internal standard of 0.155 and 0.233 (arbitrary units), respectively.
Sufficient internal standard is added to a sample to make its concentra-
tion 15.0 mg/L. Analysis of the sample yields signals for the analyte
and the internal standard of 0.274 and 0.198, respectively. Report the
analyte’s concentration in the sample.

7. For each of the pair of calibration curves shown in Figure 5.26, select
the calibration curve that uses the more appropriate set of standards.
Briefly explain the reasons for your selections. The scales for the x-axis
and the y-axis are the same for each pair.
(a)
Signal

Signal

CA CA

(b)
Signal

Signal

CA CA
(c)
Signal

Signal

Figure 5.26 Calibration curves to accom-

CA CA
pany Problem 7.
192 Analytical Chemistry 2.1

8. The following data are for a series of external standards of Cd2+ buffered
to a pH of 4.6.14
[Cd2+] (nM) 15.4 30.4 44.9 59.0 72.7 86.0
Sspike (nA) 4.8 11.4 18.2 26.6 32.3 37.7
(a) Use a linear regression analysis to determine the equation for the
calibration curve and report confidence intervals for the slope and
the y-intercept.
(b) Construct a plot of the residuals and comment on their significance.
At a pH of 3.7 the following data were recorded for the same set of
external standards.
[Cd2+] (nM) 15.4 30.4 44.9 59.0 72.7 86.0
Sspike (nA) 15.0 42.7 58.5 77.0 101 118
(c) How much more or less sensitive is this method at the lower pH?
(d) A single sample is buffered to a pH of 3.7 and analyzed for cadmium,
yielding a signal of 66.3 nA. Report the concentration of Cd2+ in
the sample and its 95% confidence interval.

9. To determine the concentration of analyte in a sample, a standard ad-

dition is performed. A 5.00-mL portion of sample is analyzed and then
successive 0.10-mL spikes of a 600.0 ppb standard of the analyte are
added, analyzing after each spike. The following table shows the results
of this analysis.
Vspike (mL) 0.00 0.10 0.20 0.30
Stotal (arbitrary units) 0.119 0.231 0.339 0.442

Construct an appropriate standard additions calibration curve and use

a linear regression analysis to determine the concentration of analyte in
the original sample and its 95% confidence interval.

10. Troost and Olavsesn investigated the application of an internal stan-

dardization to the quantitative analysis of polynuclear aromatic hy-
drocarbons.15 The following results were obtained for the analysis of
phenanthrene using isotopically labeled phenanthrene as an internal
standard. Each solution was analyzed twice.
CA/CIS 0.50 1.25 2.00 3.00 4.00
0.514 0.993 1.486 2.044 2.342
SA/SIS
0.522 1.024 1.471 2.080 2.550

14 Wojciechowski, M.; Balcerzak, J. Anal. Chim. Acta 1991, 249, 433–445.

15 Troost, J. R.; Olavesen, E. Y. Anal. Chem. 1996, 68, 708–711.
 Chapter 5 Standardizing Analytical Methods 193

(a) Determine the equation for the calibration curve using a linear
regression, and report confidence intervals for the slope and the y-
intercept. Average the replicate signals for each standard before you
complete the linear regression analysis.
(b) Based on your results explain why the authors concluded that the
internal standardization was inappropriate.

11. In Chapter 4 we used a paired t-test to compare two analytical methods Although this is a common approach for
comparing two analytical methods, it
that were used to analyze independently a series of samples of vari- does violate one of the requirements for
able composition. An alternative approach is to plot the results for one an unweighted linear regression—that in-
method versus the results for the other method. If the two methods determinate errors affect y only. Because
indeterminate errors affect both analytical
yield identical results, then the plot should have an expected slope, b1, methods, the result of an unweighted lin-
of 1.00 and an expected y-intercept, b0, of 0.0. We can use a t-test to ear regression is biased. More specifically,
compare the slope and the y-intercept from a linear regression to the ex- the regression underestimates the slope,
b1, and overestimates the y-intercept, b0.
pected values. The appropriate test statistic for the y-intercept is found We can minimize the effect of this bias by
by rearranging equation 5.23. placing the more precise analytical meth-
od on the x-axis, by using more samples
b0 - b0 b0 to increase the degrees of freedom, and
t exp = sb = s by using samples that uniformly cover the
0 b 0
range of concentrations.

Rearranging equation 5.22 gives the test statistic for the slope. For more information, see Miller, J. C.;
Miller, J. N. Statistics for Analytical Chem-
istry, 3rd ed. Ellis Horwood PTR Pren-
b1 - b1 1 - b1 tice-Hall: New York, 1993. Alternative
t exp = sb = sb
1 1 approaches are found in Hartman, C.;
Smeyers-Verbeke, J.; Penninckx, W.; Mas-
Reevaluate the data in problem 25 from Chapter 4 using the same sart, D. L. Anal. Chim. Acta 1997, 338,
19–40, and Zwanziger, H. W.; Sârbu, C.
significance level as in the original problem. Anal. Chem. 1998, 70, 1277–1280.

12. Consider the following three data sets, each of which gives values of y
for the same values of x.

Data Set 1 Data Set 2 Data Set 3 These three data sets are taken from Ans-
combe, F. J. “Graphs in Statistical Analy-
x y1 y2 y3 sis,” Amer. Statis. 1973, 27, 17-21.
10.00 8.04 9.14 7.46
8.00 6.95 8.14 6.77
13.00 7.58 8.74 12.74
9.00 8.81 8.77 7.11
11.00 8.33 9.26 7.81
14.00 9.96 8.10 8.84
6.00 7.24 6.13 6.08
4.00 4.26 3.10 5.39
12.00 10.84 9.13 8.15
7.00 4.82 7.26 6.42
5.00 5.68 4.74 5.73
194 Analytical Chemistry 2.1

(a) An unweighted linear regression analysis for the three data sets gives
nearly identical results. To three significant figures, each data set
has a slope of 0.500 and a y-intercept of 3.00. The standard devia-
tions in the slope and the y-intercept are 0.118 and 1.125 for each
data set. All three standard deviations about the regression are 1.24.
Based on these results for a linear regression analysis, comment on
the similarity of the data sets.
(b) Complete a linear regression analysis for each data set and verify
that the results from part (a) are correct. Construct a residual plot
for each data set. Do these plots change your conclusion from part
(a)? Explain.
(c) Plot each data set along with the regression line and comment on
your results.
(d) Data set 3 appears to contain an outlier. Remove the apparent out-
lier and reanalyze the data using a linear regression. Comment on
your result.
(e) Briefly comment on the importance of visually examining your
data.

13. Fanke and co-workers evaluated a standard additions method for a

voltammetric determination of Tl.16 A summary of their results is tabu-
lated in the following table.
ppm Tl
Instrument Response (mA)
added
0.000 2.53 2.50 2.70 2.63 2.70 2.80 2.52
0.387 8.42 7.96 8.54 8.18 7.70 8.34 7.98
1.851 29.65 28.70 29.05 28.30 29.20 29.95 28.95
5.734 84.8 85.6 86.0 85.2 84.2 86.4 87.8

Use a weighted linear regression to determine the standardization rela-

tionship for this data.

5J Solutions to Practice Exercises

Practice Exercise 5.1
Substituting the sample’s absorbance into the calibration equation and
solving for CA give
Ssamp = 0.114 = 29.59 M–1 × CA + 0.015
CA = 3.35 × 10-3 M
For the one-point standardization, we first solve for kA
16 Franke, J. P.; de Zeeuw, R. A.; Hakkert, R. Anal. Chem. 1978, 50, 1374–1380.
 Chapter 5 Standardizing Analytical Methods 195

kA = C S std = 0.0931 = 29.46 M -1

std 3.16 # 10 -3 M
and then use this value of kA to solve for CA.
S samp
CA = = 0.114 -1 = 3.87 # 10 -3 M
kA 29.46 M
When using multiple standards, the indeterminate errors that affect the
signal for one standard are partially compensated for by the indeterminate
errors that affect the other standards. The standard selected for the one-
point standardization has a signal that is smaller than that predicted by
the regression equation, which underestimates kA and overestimates CA.
Click here to return to the chapter.

Practice Exercise 5.2

We begin with equation 5.8
S spike = k A a C A V
Vo + C Vstd k
std
Vf
f

rewriting it as
0 = kAC A Vo
& Vstd 0
V f + k A # C std V f
which is in the form of the linear equation
y = y-intercept + slope × x
where y is Sspike and x is Cstd × Vstd/Vf. The slope of the line, therefore,
is kA, and the y-intercept is kACAVo/Vf. The x-intercept is the value of x
when y is zero, or
0 = kAC V f + k A # " x-intercept ,
A Vo

k A C A Vo V f
x-intercept =- =- CVA Vo
kA f

Click here to return to the chapter.

Practice Exercise 5.3

Using the calibration equation from Figure 5.7a, we find that the x-in-
tercept is
x - intercept =- 0.1478 -1 =- 1.731 mL
0.0854 mL
If we plug this result into the equation for the x-intercept and solve for
CA, we find that the concentration of Mn2+ is
(x-intercept) C std (- 1.731 mL) # 100.6 mg/L
C A =- =- = 6.96 mg/L
Vo 25.00 mL
For Figure 7b, the x-intercept is
196 Analytical Chemistry 2.1

x-int ercept =- 0.1478 =- 3.478 mg/mL

0.0425 mL/mg
and the concentration of Mn2+ is
(x-intercept) V f (- 3.478 mg/mL) # 50.00 mL
C A =- =- = 6.96 mg/L
Vo 25.00 mL
Click here to return to the chapter.

Practice Exercise 5.4

We begin by setting up a table to help us organize the calculation.
xi yi xiyi xi2
0.000 0.00 0.000 0.000
1.55×10–3 0.050 7.750×10–5 2.403×10–6
3.16×10–3 0.093 2.939×10–4 9.986×10–6
4.74×10–3 0.143 6.778×10–4 2.247×10–5
6.34×10–3 0.188 1.192×10–3 4.020×10–5
7.92×10–3 0.236 1.869×10–3 6.273×10–5
Adding the values in each column gives
n n
/x i = 2.371 # 10 -2 /y i = 0.710
i=1 i=1

n n
/x y i i = 4.110 # 10 -3 /x 2
i = 1.378 # 10 -4
i=1 i=1

When we substitute these values into equation 5.17 and equation 5.18,
we find that the slope and the y-intercept are
6 # (4.110 # 10 -3) - (2.371 # 10 -2) # (0.710)
b1 = = 29.57
6 # (1.378 # 10 -4) - (2.371 # 10 -2) 2
0.710 - 29.57 # (2.371 # 10 -2)
b0 = = 0.0015
6
and that the regression equation is
Sstd = 29.57 × Cstd + 0.0015
To calculate the 95% confidence intervals, we first need to determine
the standard deviation about the regression. The following table helps us
organize the calculation.
xi yi Vy i (y i - Vy i) 2
0.000 0.00 0.0015 2.250×10–6
1.55×10–3 0.050 0.0473 7.110×10–6
3.16×10–3 0.093 0.0949 3.768×10–6
 Chapter 5 Standardizing Analytical Methods 197

4.74×10–3 0.143 0.1417 1.791×10–6

6.34×10–3 0.188 0.1890 9.483×10–7
7.92×10–3 0.236 0.2357 9.339×10–8
Adding together the data in the last column gives the numerator of equa-
tion 5.19 as 1.596×10–5. The standard deviation about the regression,
therefore, is
1.596 # 10 -5 = 1.997 # 10 -3
sr =
6-2
Next, we need to calculate the standard deviations for the slope and the
y-intercept using equation 5.20 and equation 5.21.
6 # (1.997 # 10 -3) 2
sb = = 0.3007
1
6 # (1.378 # 10 -4) - (2.371 # 10 -2) 2
(1.997 # 10 -3) 2 # (1.378 # 10 -4)
sb = = 1.441 # 10 -3
0
6 # (1.378 # 10 -4) - (2.371 # 10 -2) 2
and use them to calculate the 95% confidence intervals for the slope
and the y-intercept
b 1 = b 1 ! ts b = 29.57 ! (2.78 # 0.3007) = 29.57 M -1 ! 0.84 M -1
1

b 0 = b 0 ! ts b = 0.0015 ! (2.78 # 1.441 # 10 -3) = 0.0015 ! 0.0040

0

With an average Ssamp of 0.114, the concentration of analyte, CA, is

S samp - b 0
CA = = 0.114 - 0.0015 = 3.80 # 10 -3 M -1
b1 29.57 M -1
The standard deviation in CA is
(0.114 - 0.1183) 2
s C = 1.997 # 10 -3
29.57
1+1+
3
A
6 (29.57) 2 # (4.408 # 10 -5)
= 4.778 # 10 -5
and the 95% confidence interval is
n = C A ! ts C = 3.80 # 10 -3 ! " 2.78 # (4.778 # 10 -5) ,
A

n = 3.880 # 10 -3 M ! 0.13 # 10 -3 M
Click here to return to the chapter.

Practice Exercise 5.5

To create a residual plot, we need to calculate the residual error for each
standard. The following table contains the relevant information.
198 Analytical Chemistry 2.1

xi yi Vy i y i - Vy i
0.000 0.00 0.0015 –0.0015
0.010 1.55×10 –3 0.050 0.0473 0.0027
3.16×10–3 0.093 0.0949 –0.0019
4.74×10–3 0.143 0.1417 0.0013
residual error

0.000
6.34×10–3 0.188 0.1890 –0.0010
7.92×10–3 0.236 0.2357 0.0003

-0.010
Figure 5.27 shows a plot of the resulting residual errors. The residual er-
0.000 0.002 0.004 0.006 0.008 rors appear random, although they do alternate in sign, and that do not
CA show any significant dependence on the analyte’s concentration. Taken
Figure 5.27 Plot of the residual errors for together, these observations suggest that our regression model is appro-
the data in Practice Exercise 5.5. priate.
Click here to return to the chapter

Practice Exercise 5.6

Begin by entering the data into an Excel spreadsheet, following the format
shown in Figure 5.15. Because Excel’s Data Analysis tools provide most of
the information we need, we will use it here. The resulting output, which
is shown in Figure 5.28, provides the slope and the y-intercept, along
with their respective 95% confidence intervals. Excel does not provide a
function for calculating the uncertainty in the analyte’s concentration, CA,
given the signal for a sample, Ssamp. You must complete these calculations
by hand. With an Ssamp of 0.114, we find that CA is
S samp - b 0
CA = = 0.114 - 0.0014 = 3.80 # 10 -3 M
b1 29.59 M -1
The standard deviation in CA is

SUMMARY OUTPUT

Regression Statistics
Multiple R 0.99979366
R Square 0.99958737
Adjusted R Square
0.99948421
Standard Error 0.00199602
Observations 6

ANOVA
df SS MS F Significance F
Regression 1 0.0386054 0.0386054 9689.9103 6.3858E-08
Residual 4 1.5936E-05 3.9841E-06
Total 5 0.03862133

Coefficients Standard Error t Stat P-value Lower 95% Upper 95% Lower 95.0% Upper 95.0%
Intercept 0.00139272 0.00144059 0.96677158 0.38840479 -0.00260699 0.00539242 -0.00260699 0.00539242
Cstd 29.5927329 0.30062507 98.437342 6.3858E-08 28.7580639 30.4274019 28.7580639 30.4274019

Figure 5.28 Excel’s summary of the regression results for Practice Exercise 5.6.
 Chapter 5 Standardizing Analytical Methods 199

(0.114 - 0.1183) 2
s C = 1.996 # 10 -3
29.59
1+1+
3
A
6 (29.59) 2 # (4.408 # 10 -5)
= 4.772 # 10 -5
and the 95% confidence interval is
n = C A ! ts C = 3.80 # 10 -3 ! " 2.78 # (4.772 # 10 -5) ,
A

n = 3.80 # 10 -3 M ! 0.13 # 10 -3 M
Click here to return to the chapter

Practice Exercise 5.7

Figure 5.29 shows the R session for this problem, including loading the
chemCal package, creating objects to hold the values for Cstd, Sstd, and
Ssamp. Note that for Ssamp, we do not have the actual values for the three
replicate measurements. In place of the actual measurements, we just en-
ter the average signal three times. This is okay because the calculation
depends on the average signal and the number of replicates, and not on
the individual measurements.
Click here to return to the chapter
> library("chemCal")
> conc=c(0, 1.55e-3, 3.16e-3, 4.74e-3, 6.34e-3, 7.92e-3)
> signal=c(0, 0.050, 0.093, 0.143, 0.188, 0.236)
> model=lm(signal~conc)
> summary(model)

Call:
lm(formula = signal ~ conc)

Residuals:
1 2 3 4 5 6
-0.0013927 0.0027385 -0.0019058 0.0013377 -0.0010106 0.0002328

Coefficients:
Estimate Std. Error t value Pr(>|t|)
(Intercept) 0.001393 0.001441 0.967 0.388
conc 29.592733 0.300625 98.437 6.39e-08 ***
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

Residual standard error: 0.001996 on 4 degrees of freedom

Multiple R-Squared: 0.9996, Adjusted R-squared: 0.9995
F-statistic: 9690 on 1 and 4 DF, p-value: 6.386e-08

> samp=c(0.114, 0.114, 0.114)

> [Link](model,samp,alpha=0.05)
$Prediction
[1] 0.003805234

$`Standard Error`
[1] 4.771723e-05

$Confidence
[1] 0.0001324843
Figure 5.29 R session for completing
$`Confidence Limits`
[1] 0.003672750 0.003937719 Practice Exercise 5.7.
200 Analytical Chemistry 2.1