The test statistic can now be calculated as
p= Fite U6 0123 _ yg
SIWN ~ 0032/
From Table al-5, we find that the critical value of ¢ for 3
degrees of freedom and the 95% confidence level is 3.18,
Because t = ~3.18, we conclude that there is a significant
difference at the 95% confidence level and, thus, bias in the
method, Note that if we were to carry out this test atthe 993%
confidence level, ,,, = 5.84 (Table a1-5). Because ~5.84 <
4.375, we would accept the null hypothesis at the 99% con-
dence level and conclude there is no dilference between the
experimental and the accepted values,
‘Note that sn this example the outcome depends on the confi-
dence evel that is used, The choice of confidence level depends
‘on our willingness to accept an error in the outcome.” The
significance level (0.05, 0.01, 0.001, etc) is the probability of
‘making an error by rejecting the null hypothesis.
Hypothesis testing is widely used in science and engineer
ing Comparison of one or two samples is carried out as
described here, The principles can, however, be extended to
‘comparisons among more than two population means. Multiple
‘comparisons fall under the general category of analysis of vari
‘ance (ANOVA). These methods use a single test to determine
‘whether thee is a ditference among the population means rath
cer than pairwise comparisons as is done with the f test. After
ANOVA indicates a potential ditference, multiple comparison
procedures can be used to identify which specific population
‘means differ from the others. Experimental design methods take
advantage of ANOVA in planning and performing experiments.
a1D METHOD OF LEAST SQUARES
Most analytical methods are based on a calibration curve in
which @ measured quantity y is plotted as a function of the
known concentration x of a series of standards. Figure al-6
shows a typical calibration curve, which was computed for the
‘chromatographic determination of isooctane in hydrocarbon
samples. The ordinate (the dependent variable) is the area un.
der the chromatographic peak for isooctane, and the abscissa
(che independent variable) is the mole percent of isooctane. AS
is typical, the plot approximates a straight line. Note, however,
"hor discussion a ror fn hypothe tenting, ee. 1, Devore, Probably
and tats for Engineering and he Seiences, hed, Chap, 8 Boston: Brooks
Cle 2016
see. A, Skoog DM. West FJ. Hele and SR. Crouch, Purdament of
“Anayial Chemistry, the, Chap 7 Belmont, CA: Brook Cale, 201: 5.8
(Crouch and ff Hele Applications of Mirza" Excel n Analytical Chemis
"ora dssion of ANOVA methods, sD. A, Skoog .M. Wes FJ Hale
and §, 8 Grouch, Fundamental of Analytical Chemistry, th eda Ses, 7
D Method of Least Squares 903,
\
20
Peak are, abiary
S
05 10 15 20
+ Concenation af soctan, mol %
[FIGURE 21-6 Calibration curve for determining iseactane in
hydrocarbon mixtures
that not all the data fall exactly on the line because of the
random errors in the measurement process. Thus, we must try
o find the “best” straight line through the points. Regression
analysis provides the means for objectively obtaining such a
line and also for specifying the uncertainties associated with its
subsequent use, We consider here only the basic method of least
squares for two-dimensional data.
a1D-1 Assumptions of the
Least-Squares Method
Two assumptions are made in using the method of least
squares The fist is that there is actualy a linear relationship
between the measited response y and the standard analyte
concentration x. The mathematical relationship that describes
this sumption is ealed the ragresion model, which may be
represented as
yrmetb
where b is the y intercept (the value of y when x is 0) and m
is the slope of the Tine, We also assume that any deviation of
the individual points from the straight line arises from error
in the measurement. That is, we assume there is no ertor in x
values of the points (concentrations). Both ofthese assumptions
are appropriate for many analytical methods, but bear in mind906 Appendix 1 Evaluation of Analytical Data
that whenever there i significant uncertainty in the data, ba
sic linear least squares analysis may not give the best straight
line. In such a case, a moze complex correlation analysis may be
necessary, In addition, simple least squares analysis may not be
appropriate when the uncertainties in the y values vary signifi
‘cantly with x. In this case, it may be necessary to apply diferent
Weighting factors to the points and perform a weighted least
squares analysis.
a1D-2 Finding the Least-Squares Line
Asillustrated in Figureal-6,the vertical deviation ofeach point
from the straight in is called a residual, Te line generated by
the least-squares method is the one that minimizes the sum
ofthe squares ofthe residuals forall the points. In addition to
roving the bed ft between the experimental pointe and the
Straight line, the method gives the standard deviations for m
and
“he leastquares method finds the sum ofthe squares ofthe
resale SS, and minimizes these according othe minim
zation technigue of calculus! Theva of SS, found from
Db. - (0+ mad?
‘where N ie the number of points used. The calculation of the
slope and intercept is simplified by defining three quantities S,,
Syp and S,, 28 follows
Gs
Sa = Di - FP = Yd - (al-31)
Sy = YO. - FF = Sy (al-32)
$= Dl w=) = Ya - SE aay
where x, and y, are individual pairs of data for x and y, Nis the
number of pairs, and ¥ and Yar the average values for xand y,
thats, = Syx/N andy ~ Sy tN.
Note that ,, and S,, are the sums ofthe squares of the de
vations from the mesh for individual values of x and y. ‘The
‘expressions shown on the fa right in Equations a-31 through
41-33 are more convenient when a calculator without a built-in
regression function is being used
"or an Bice approach to weighted linear
FU Hole, Applications f Micros elm Analy
rons CA. Cengage Learning, 2017 pp. 31-397
reason, see SR. Crouch and
I Chemistry. 3rd, Bel
"the procedure lavolvesdierentting SS. with sespect test and then B
sand rtng the derivatives equa 00. Tie ls two equions, called oral
“suationsin the two unknowns and These are then ole the eae
Six useful quantities can be derived from S,,.$,, and S,, as
se Syp and Sy
follows:
41. The slope of the tine, m:
e139
2
135)
3
@1369)
44. The standard deviation of the slop, s,
Ve @rs7)
5, The standard deviation of the intercept
4 (al-38)
Vi-Sa De
The standard deviation for results obtained from
the calibration curve, 5:
=
5
139)
Equation 1-39 allows us to calculate the standard deviation
from the mean J. ofa set of M replicate analyses of unknowns
when a calibration curve that contains N points is used; recall
that Fis the mean value of y for the N calibration points, This
equation is only approximate and assumes that the slope and
Intercept are independent parameters, which isnot strictly true,
“The standard deviation about regression s, (Equation al-36)
is the standard deviation for y when the deviations are mea
sured not from the mean of y (as isthe usual case), but from the
straight line that results from the least-squares prediction, The
value of sis related to 85, by
S.-co + made
N=2 VN=2
In this equation, the number of degrees of freedom is N- 2
because one degree of freedom is lst in calculating m and one
in determining b. The standard deviation about regression is
often called the standard error of the estimate, It roughly
responds to the size ofa typical deviation from the estimated
regression line. Examples al-I1 and al-12 illustrate how these
quantities are calculated and used.fee
Complete a least-squares analysis of the experimental data
provided in the fist two columns of Table al-7 and plotted in
Figureal-6.
Solution
Columns 3,4, and 5 of the table contain computed values for
2S, yfvand xy, with their sums appearing asthe lst entry in
‘each column. Note thatthe number of digits carried in the
‘computed values should be the maximum allowed by the cal-
culatoror computer, tat is, rounding should not be performed
tun the calculation is complete
‘We now substitute into Equations al-31, a1-32, and
21-33 and obtain
yy -
(228i
= 363775 - = 5.07748
5.365 12.51
= 1ssiss2 ~ SE *
39669
Substitution of these quantities into Equations al-34 and
al-35 yields
2.39669
14537
= 20925 = 209
‘TABLE al-7 Calibration Data for the Chromato-
graphic Determination of Isooctane in a Hydrocarbon
Mixture
co
ot
res
a
rn
012390 1.1881
3.1684
6.7600
9.1809
16.0801
36.3775
0.38368
142934
2.80800
4.18140
7.01750
15.8192
0.64481
1.16640
1.90440
3.06250
690201
> anD Method of Least Squares 905
“Thus, the equation for the least-squares ine is
y= 209x + 026
Substitution into Equation al-36 yields the standard deviation
about regression
(= BT
5-2
and substitution into Equation al-37 gives the standard
deviation of the slope
Sa
Ne
= 01442 = 014
2 _ feiway
Sa V104537
Finally, we find the standard deviation of the intercept from
Equation al-38:
‘The calibration curve found in Example al-11 was used for
the chromatographic determination of isooctane in a hydro-
carbon mixture. A peak area of 2.65 was obtained. Calculate
the mole percent of isooctane in the mixture and the standard