ERRORS
Error is defined as the numerical difference between a measured value and the absolute or true
value of an analytical determination. The absolute or true value of a quantity is, however, never known.
All that we can use is only an accepted value.
value. The value for any quantity is 'accepted' whe the
uncertainty in this value is less than the uncertainty in some other quantity with which the given quantity
is to be compared. The values which we may like to accept' can be obtained by a variety of methods
which. however. have their own limitations. It is very rare that two different methods may vield identical
accepted values.
The error in a measured quantity may be represented either as absolute error or as relative error.
Absolute Error. The absolute error E in a measurement is expressed as
..(1)
where x, is the measured value and x, is the true (accepted)value for the given measurement.
Relative Error. The relative error E, in a measurement is expressed as
E, = ..(2)
where x, andas x,parts
haveperthethousand
same significance as mentioned above. Relative error is generally exDressed as
(ppt). Thus,
per cent or
E, =
xË Lx 100 %
... (3)
E, = x 1000 ppt
.. (4)
EVALUATION OF ANALYTICAL DATA 1087
Confidence Limits. For an infinite number of replicate measurements made on the same material, the
mean value, which could bc calculatcd in principle, would be the population mean, u. As already
mentioned, the value of u can never be known in practice. In contrast, the value of sample mcan x can
be known almost exactly from a finite number of measurements.
With the help of statistical theory, it is possible to set limits around experimentally derived mean
within which the sample mean x is likely to be in agreement with the population mean u with a given
degrec of probabiliy. These limits are referred to as the confidence limits and the range defined by these
limits is known as the confidence interval.
Mathennatically, the confidence limit is given bythe expression.
tS
Confidence limit= t ... (27)
where I is the sample or experimental mean, Nis the number of measurements (N< 25), S is the sample
standard deviation and is afactor whose magnitude depends upon the number of measurements as well
as the confidence level desired. The confidence levei signifies the degree of probability (99%, 95%,
90%, 50%o, etc.) acceptable in the results. The factor t is often called the Student's t. The values of t
coresponding to various levels of probability and varying degrees of freedom are available in Statistical
Tables. Aportion of one such table is reproduced here for convenience of reference (Table 1).
TABLE1
Student's t values for Various Levels of Probability and Varying
Degrees of Freedom
Number of Number of degrees Probability Levels :
observations of freedom
N N-1 50% 90% 95% 99%
1 1-000 6-314 12-706 27-320
3 2 0-816 2-920 4-303 9-925
4 3 0-765 2-353 3182 5-841
0-741 2-132 2-776 4-604
6 5 0-727 2-015 2-571 4-032
7 6 0-718 1·943 2-447 3-707
7 0711 1·895 2-365 3-500
9 0-706 1·860 2-306 3-355
10 0-703 1·833 2-262 3-250
11 10 0-700 1·812 2-228 3-169
21 20 0-687 1:725 2-086 2-845
0-674 1:645 1·960 2-576
Example 13. The following volumes of titrant were used in replication of atitration : 30-26, 30-24, 30-30,
30-18, 30-20 ml. Caleulate the confidence limit for the following probability levels : 50%, 90%, 98% and
99%.
Solution :
Serial Results of Deviaiton (Deviation)'
Number measurement, x; (t; - )
30:26 0-02 0-0004
2 30-24 0-00 0-0000
3 30:30 0-06 0-0036
4 3018 0-06 0-0036
30-20 0-04 0-0016
Mean, I -|5118/5 = 30-24 E(«;-) = 0-0092
1088 EVALUATION OF ANALYTICAL DATA
Standard deviation, S o.0092 0.(0479
4
Degrees of fiecdom =N-| -5-| 4
The Student's /values corresponding to various levels of probability reproduced from Table 1for degrees of freedom
- 4, are as follos :
Level of Probability Student's value
50 % 0-765
90 % 2:353
95 % 3-182
99 % 5-841
Confidence limit = I tIS/N (Eq. 27)
0-0765 >x0-0479
Confidence linnit for 50% probability level = 30-24 + = 30-24 ±0-016
The above result implies that the population mean
30-24 - 0-016.
would lie with 50% probability within 30:24 + 0-016 and
Similarly. the confidence limits for other probability levels come out to be as follows :
Probability level Confidence limit
2-353 x 0-0479
90 % 30-24 t = 30-24 + 0-050
5
3-182 x 0-0479
95 % 30-24 + = 30-24 ± 0-068
5
5-841 x 0-0479
99 % 30-24 t = 30-24 ± 0-125
Rejection of a Result. It is a common experience that when we perform a large number of replicate
analyses, one of the results may differ excessively from the average value, i.e., the particular result may
be too high or too low. So we have to take a decision whether we should retain or reject
that suspect
result. Several statistical tests have been proposed for this purpose but these have to be applied with
utmost care.
The simplest test which has been widely used by analysts is calied the test (Quotient test). In
test this
1. The results of various observations in a given set are arranged in the
numerical values. decreasing order of their
2. The difference between the suspected result x, and its nearest neighbour x,
or-) is divided by the spread wof the entire set (i.e., the difference (ignoring the sign +
between the minimum and maximum
values in the entire set ) to give a quantity exD: Thus,
Qexp = (x X) /w ... (28)
3. The ratioop is then compared with critical values of the rejection ratios obtained statistically at
appropriate contidence levels and tabulated as afunction of the number of observations. These ratios are
termed as critical
The critical values for rejection quotient at 90% confidence level are given in Table 2.
TABLE 2
Critical Values for Rejection Quotient at 90% Confidence Level
Number of Observations,N Qeritlcal Number of Observations, N Qeritical
0-94
0-51
4 0-76 8 0-47
0-64
0-56
0-44
6 10
0-41
IfQexp is greater than Qeritical, the suspect resul1 is rejected.
1089
EVALUATION OF ANALYTICAL DATA
The application of the Q test is illustrated below.
to be 66-00, 65-55,
Eample 14. In the analysis of an iron ore, the percentayes of Fe,0, were found
on 07:85, 66-85, 69:90 and 65-00. The vaue 69.90 appears to be suspect. Ascertain whether this should be
inedor rejected. The eritical or 7 observations nt 90% confidence level is 0:49.
Solution: ) Aranging the various experimentul valucs in the decreasing order of magnitude, we have
65-00
0 Q0. 0SS, 66 8S, 6600, 65:90, 65-55,
.o) he ditfierenee betwcen the suspect result (r) and the nearest neighbour ,,
( ) 69-90%
17. 67:85"% 2-05%
w) Sred of the enire set, w 69.90%- 6500% *4-90%
() Oe 2-0S/4-90 0418
(1) Qeritical = 0:49 (given)
Since Qevp is less than Qoriticat, hence the suspect result cannot be rejected.
Example 15. The amount of mohr salt present in a given solution was determined volumetrically by
titration against a standard solution of potassium permanganate. The results obtained in seven observations
are as follows : 19-8, 20-2, 19-4, 19-0, 19-6, 21-0 and 23"4 gram per litre. Determine if the suspect result 23"4
can be rcjected or not. Qcritical for 7 observations at 90% confidence level is reported to be 0-52.
Solution : 1. Arranging the results in the decreasing order of their magnitude, we have :
23-4, 21-0, 20-2, 19-8, 19-6, 19-4, 19-0
2. The difference betwcen the suspect result and the nearest neighbour = 23-4 - 21-0 = 2:4
3 Spread of the entire set = 23-4 - 19-0 = 4-4
4 QexD 2-4/4-4 = 0:545
5. Qeritical =0$2 (given)
Since Qexp is greater than Qeritical hence the suspect result can be rejected.
Tests of Significance. As and when a new method is developed for analytical measurements, it is
subjected to certain tests in order to ascertain whether there is any significant difference between the new
method and the standard methods already accepted for these measurements. Two tests are commonly
employed for this purpose. These are: 1. The F-test and [Link] t-Test.
1. The [Link]. This test is based on measuremnents of standard deviation, S. The various steps
involved in this test are as follows:
1. Determination of standard deviation by employing the standard method. The standard deviation
is determined with the help of Eq. 8, viz.,
S = ...(29)
N-1
where N-lgives the number of degrees of freedom which is one less than the number Nof the results.
The value of Nis sufficiently large. The deviation obtained by the standard method is denoted as Sstd:
2. Determination of standard deviation by employing the newly developed method. The degrees of
freedom (or the number of results) used in this determination is small, generally less than 20. Let the
standard deviation obtained by the new method be denoted by S,
3. Determination of the F-value with thehelp of the expression
..(30)
4. Tabulation of critical values of F determined statistically at appropriate levels of contidence by
extendingthe degrees of frecdom (N values)to infnity both for the determination of S,d and S,..
5. Comparison of the value of Fasobtained trom Eq. 30with the critical value ofE correspondine to
N as infinity. /fthe value of F is less than the critical value, there is no significant diflerence in the
method.
stundurd
precision of the new method and the
4 7
methud.
2, The est, n this twet, different samps of hs
maerialtder te, having iifsrsnt mHHM,
of the telatit
The citital values fae al4, ohed stathelly
a apppis svels of suriftense and tablatd
signiftant lllereIe between lhe tww melhodi
EVALUATION OF ANALYTICAL DATA 1091
Example 18. The amount of oxalie acid present in a given solution was determined by two methods, one
standard and the other new, when the following results were obtained :
Amount of Oxalie Acid (g/litre) determined by :
Sample Number Standard Method New Method
865 9:35
2 11-70 11·06
3 7"38 8-90
4 13·95 12-58
17-26 18-99
12-80 11-72
Show that there is no significant difference between the two methods.
Solution: According to the data given above, the values of various expressionsare as follows :
Result by Result by
Sample No. d; = new method) standard d;-d (a,-d)?
method
1 + 0:70 + 0-56 0-314
- 0-64 - 0-78 0-608
+ 1:52 + 1-38 1-904
4 -1:37 - 1:51 2-280
+ 1:73 + 1·59 2-528
6 -- 1·08 - 1:22 1·488
Sum = 0-86 E(d; - d)² =9.122
d = 0-86/6 = 0-14
9-122 = 1·35
S,
d = V
N-1
1= d
-VN= 0-146 =0.254
1·35
The critical value of 1for 5degrees of freedom at appropriate confidence level is reported to be 0-48. Since the
experimental value is less then the critical value of t, it shows that there is no significant difference between the two
methods.
