STATISTICS IN

ANALYTICAL
CHEMISTRY
Statistics in Analytical Chemistry

 Modern analytical chemistry is concerned with

 Detection, identification, and measurement of the
chemical composition of unknown substances through
use of existing and new instrumental techniques
 It is a quantitative science, meaning that the desired
result is almost always numeric
Statistics in Analytical Chemistry

 Quantitative data about a sample are obtained with

the help of an instrument from an observable signal
 Over time, due to noise and/or drift within the
instrument there is some variation in the signal
 This variation in signal causes error in measurement
 Calibration of response is necessary in order to
obtain meaningful quantitative data
 However, it is impossible to get identical data
Statistics in Analytical Chemistry

 Statistics, therefore, provides

 An estimate of the likely value of that error; or
 Help in establishing the uncertainty associated with the
measurement
Statistics in Analytical Chemistry

 The fundamental hypothesis in statistics is the Null

Hypothesis
 The null hypothesis states that random error is
sufficient to explain differences between two values
 Statistical tests are designed to test the null
hypothesis
Statistics in Analytical Chemistry

 Passing a statistical test means that the null

hypothesis is retained
 However, it is impossible to show that two values are
the same
 It is only possible to show they are different
 Hence, null hypothesis retained (not rejected)
indicates that
 There is insufficient evidence to show that there is a
difference between the samples
What does statistics involve?

 Defining properties of probability distributions for

infinite populations
 Application of these properties to treatment of finite
(real-world) data sets
 Probabilistic approaches to:
 Finite sampling
 Experimental design
 Data treatment
 Reporting data
Some useful statistics terms

 Mean
 Average of a set of values
 Median
 Mid-point of a set of values
 Population
 A collection of an infinite number of measurements
carrying the property being studied
 N  infinity
 Sample
 A finite set of measurements which represent the
population
Some useful statistics terms

 True value (true mean)

 Mean value for the population ()
 Observed Mean
 Mean value of the sample set ()
Accuracy and Precision

 Accuracy is a measure of rightness

 Precision is a measure of exactness

 Accurate means "capable of providing a correct

reading or measurement." A measurement is accurate
if it correctly reflects the size of the thing being
measured
 Precise means “repeatable, reliable, getting the same
measurement each time”
Precision

 Every experimentally measured value has an

associated uncertainty
 Precision is characterized as the distribution of
random fluctuations about the ‘true’ value
 Statistics assumes that the distribution is Gaussian
(a.k.a. ‘normal’)
 A Gaussian distributions’ width is defined by a single
parameter, the standard deviation, 
Precision

 Figure illustrates the dependence on 

 68.3 % of the Gaussian's area is contained between -
and 
 95.4 % between -2 and 2, and
 99.7 % between -3 and 3
Precision

  of a series of observations is used to determine the

certainty with which we can report a value
 It is impossible to reduce  to zero, even with an
infinite number of observations
 A multitude of factors affect precision:
 Instrument (detector sensitivity, noise, etc.)
 Experimental technique (pipetting, weighing, filling,
etc.)
 Sample inhomogeneity
Accuracy

 Accuracy is a measure of the difference between an

experimental value and the true value
 Any difference is due to systematic error(s)
 Accuracy can only be determined where the ‘true’
value of a sample is known, i.e., a reference
 Since their actual value is known, systematic errors
can be detected by comparing the experimental value
with the true value
Accuracy

 Accuracy is usually expressed as either an absolute

error

 or a percent relative error, Er

Absolute Error

 Absolute error is the amount of physical error in a

measurement
 E.g.
 Let’s say a meter stick is used to measure a given
distance
 The error is ± 1 mm
 This is the absolute error of the measurement
 Therefore, absolute error = ± 1 mm (0.001 m)
Relative Error

 It gives an indication of how good a measurement is

relative to the size of the thing being measured
 E.g.
 Height of a room = 3.215 m ± 1 mm (0.001 m)
 Height of a cylinder = 0.075 m ± 1 mm (0.001 m)
 Relative error shows accuracy of ceiling height is better
than that of cylinder
 Comparative accuracy of the measurements can be
determined by looking at their relative errors

Accuracy and Precision
Tolerance

 Tolerance is not a statistical parameter

 It is the permissible range of variation from the
expected standard
 E.g., If tolerance of a 10.00 mL volumetric pipette is
±0.02 mL
 The pipette is guaranteed to deliver between 9.98 mL
and 10.02 mL
 It does not mean that the pipette will deliver an
average of 10.00 mL
 A given pipette might routinely deliver 9.997 mL or
10.015 mL or 9.981 mL
Tolerance

 Unlike precision, tolerance does not have a Gaussian

distribution
 Analytical chemists can repeatedly deliver within
±0.02 mL with a 10.00 mL pipette
 It is a systematic error if you report the volume
delivered by a 10 mL pipette as (10.00 ± 0.02) mL,
which is the tolerance, when the pipette actually
delivers (10.011 ± 0.004) mL
Common Questions

 When I repeat a measurement, I get different

numbers; which do I use?
 Remember that the inherent variation associated with
any real measurement means you would always expect
to get somewhat different values for replicate
measurements
 Calculate the mean and standard deviation of the
values; this is the starting point for any statistical
evaluation of your data
Common Questions

 Did I get the ‘right’ answer?

 It’s almost impossible to answer this question in any
meaningful way for real samples!
 If you know what the true value is, you can assess
whether your result is significantly different using a t-
test
 If true value (or an accepted true value) is not known,
determine the range on either side of the measurement
within which the true value most probably lies
 This is called the confidence interval, and is a measure
of the uncertainty associated with any measurement
Common Questions

 One sample gives a value of 2.1, the other gives 2.2;

these are different values, right?
 This depends on the uncertainty associated with each
measurement
 A significance test has to be performed to determine if
the values can be considered same or different
 Test whether there is a significant difference in the
spread of replicate measurements
Common Questions

 One of the values is quite different from the others;

can I simply ignore it?
 This depends on
 The range of the values you obtained,
 How different the suspect value is from all the others,
and
 How close the remaining results are to one another
 Use either Dixon's Q test or Grubb's test on the data
Rejecting data

 It is good practice to check outliers in a data set to

see if they can statistically be rejected
Outliers

 A data object that deviates significantly from the

normal objects as if it were generated by a different
mechanism
 Outliers are different from the noise data
 Noise is random error or variance in a measured
variable
 Noise should be removed before outlier detection
 Outliers are interesting as it violates the mechanism
that generates the normal data
Outliers
Outliers

 Three kinds of outliers

 Global
 Contextual
 Collective
Outliers

 Global outlier (or point outlier) (Og)

 Object is Og if it significantly deviates from the rest of
the data set
 Simplest type of outlier
Outliers

 Contextual outlier (or conditional outlier) (Oc)

 Object is Oc if it deviates significantly based on a
selected context
 E.g.
 18Cin Mumbai: outlier?
 Depends on summer or winter
Outliers

 Contextual outlier (or conditional outlier) (Oc)

 Attributes of data objects should be divided into two
groups
 Behavioral attributes: characteristics of the object, used
in outlier evaluation, e.g., temperature
 Contextual attributes: defines the context, e.g., time &
location
Outliers

 Collective Outliers
 A subset of data objects collectively deviate
significantly from the whole data set, even if the
individual data objects may not be outliers
Outliers

 Collective Outliers
 Detection of collective outliers
 Consider not only behavior of individual objects, but also
that of groups of objects
 Need to have the background knowledge on the
relationship among data objects, such as a distance or
similarity measure on objects
Outliers

 Simple example
 Global outlier
 A fist-size meteorite impacting a house in your
neighbourhood
 It’s a truly rare event that meteorites hit buildings
 Contextual outlier
 Your neighbourhood getting buried in two feet of snow
 If snowfall happened in middle of summer and you
normally don’t get any snow outside of winter
Outliers

 Simple example
 Collective outlier
 Every one of your neighbours is moving out of the
neighbourhood on the same day
 Although it’s definitely not rare that people move from
one residence to next, it is very unusual that an entire
neighbourhood relocates at same time
Rejecting Outliers

 Q-test for rejecting outliers

d = Gap from the nearest value = |x5 – x6|

w = Range/spread of the data = |x1 – x6|
Rejecting Outliers

 Q-test for rejecting outliers

 Qtab is looked up in a table and compared with Qcalc

 If Qcalc > Qtab, the outlier data point can be rejected at
the specified confidence level
Rejecting Outliers

 Consider the data set:

0.189, 0.167, 0.187, 0.183, 0.186, 0.182, 0.181,
0.184, 0.181, 0.177

Find if there is any outlier in the above data

Rejecting Outliers
Rejecting Outliers

 Grubb’s test for rejecting outliers

 denotes sample mean and ‘s’ denotes standard

deviation
 Ymin and Ymax denote minimum and maximum values,
respectively, that are suspect for outlier
 Gtab is looked up in a table and compared with Gcalc
 If Gcalc > Gtab, the outlier data point can be rejected at
the specified confidence level
Rejecting Outliers
Experimental Errors

 Successive measurements of same parameter, for the

same sample and method, will result in a set of
values which vary from the ‘true’ value by differing
amounts
 It means, our measurements are subject to error
 This is the principal reason why a result based on a
single measurement is meaningless in scientific terms
 Formally, the error is defined as the result of the
measurement minus the true value, (xi – )
 Consequently, errors have both sign and units
Experimental Errors

 Experimental scientists make a fundamental

distinction between three types of error
 These are known as
 Random (indeterminate) error,
 Systematic (determinate) error, and
 Gross error
Experimental Errors

 Random (indeterminate) error

 Affects precision – repeatability or reproducibility
 These are caused by many uncontrollable variables
 Caused by both humans and equipment
 Cannot be avoided
 Random distribution
 Mathematical laws of probability
 Normal distribution or Gaussian curve
 Cause replicate results to fall on either side of mean
Experimental Errors

 Random (indeterminate) error

 Represent the experimental uncertainty that occurs in
any measurement
 Small difference on successive measurements
 Can be minimized by good technique but not
eliminated
 If measurements repeated enough times, random errors
tend to cancel out when calculating the mean
Experimental Errors

 Random (indeterminate) error

Random errors follow a Gaussian or normal distribution

Experimental Errors

 Systematic (determinate) error

 Affects accuracy – produces bias
 An overall deviation of result from the true value even
when random errors are very small
 Cause all results to be affected in one sense (direction)
only
 All too high or all too low
 Cannot be detected by using replicate measurements
 Can be corrected, e.g. by using standard methods and
materials, calibrating the instrument etc.
 Caused by both humans and equipment
Experimental Errors

 E.g. Data demonstrating random and systematic errors

 Four students (A–D) each perform an analysis
 10.00 ml of 0.1 M NaOH is titrated with 0.1 M HCl
 Each student performs five replicate titrations, with the
results shown below

Student Results (ml) Mean

(ml)
A 10.08 10.11 10.09 10.10 10.12 10.10
B 9.88 10.14 10.02 9.80 10.21 10.01
C 10.19 9.79 9.69 10.05 9.78 9.90
D 10.04 9.98 10.02 9.97 10.04 10.01
Experimental Errors

 Bias and precision: dot-plots of the data

 =10.10
Precise, biased

 = 10.01
Imprecise, unbiased

 = 9.90
Imprecise, biased

 = 10.01
Precise, unbiased
Experimental Errors

 Gross error
 They are so serious that there is no alternative to
abandoning the experiment and making a completely
fresh start
 E.g.
 A complete instrument breakdown
 Accidentally dropping or discarding a crucial sample
 Discovering during the course of the experiment that a
supposedly pure reagent was in fact badly contaminated
Sources of Errors

 There are several sources of experimental errors that

affect accuracy and precision
 Sampling errors
 Instrument errors
 Method errors
 Personal or operational errors
Sources of Errors

 Determinate sampling error

 Occurs when our sampling strategy does not provide a
representative sample
 E.g. if you monitor the environmental quality of a lake
by sampling a single location near a point source of
pollution, such as an outlet for industrial effluent, then
your results will be misleading
Sources of Errors

 Determinate instrument error

 Arise from the way a measuring device is used
 E.g.pipettes, burettes, and volumetric flasks may have
volumes slightly different than those indicated by their
graduations
 These differences may result from
 Using the glassware at a temperature different than its
calibration temperature
 Contamination on the inner surfaces of the glass, or
 Distortions in the container walls due to heating
 These errors are usually corrected by proper calibration
of instruments and cleaning of glassware
Sources of Errors

 Determinate method error

 Arise from
 Non-ideal chemical or physical behavior of materials used
in analysis (Insolubility of precipitates, co-precipitates,
decomposition, and volatilization)
 Slow or incomplete reactions and
 Instability of some reactant species
 E.g. Use of indicator dye to signal end of titration
 Error due to small excess of reagent required to cause the
dye to change color
 This error can be minimized or eliminated by carefully
choosing the experimental method
Sources of Errors

 Determinate personal error

 Occur because most measurements require the analyst
to make a personal judgment
 E.g.
 Estimating level of a liquid between two graduation
marks on a burette. Some analysts may consistently read
a meniscus high
 Gauging color of a solution at the end point of a titration.
An analyst with color insensitivity may use excessive
reagent in a volumetric analysis
Errors, Uncertainty, and Residuals

 It is not uncommon for analytical chemists to use the

terms, “error” and “uncertainty” somewhat
interchangeably, although this can cause confusion
 The primary aim in analytical chemistry is to
determine
 How close a result is to the ‘true’ value (the accuracy),
and
 How well replicate values agree with one another (the
precision)
 Systematic and random errors cannot actually be
determined unless the true value, , is known
Error and Uncertainty

 E.g.,
 Consider a titration in which same 25.00 mL pipette is
used to dispense portions of sample for replicate
determinations
 If the volume of pure water delivered by the pipette at
a specified temperature is ±0.03 mL
 Itmeans, volume might be 24.97 mL (a systematic error
of -0.03 mL) or 25.03 mL (a systematic error of +0.03
mL)
 In theory error can be determined by calibrating it
through weighing replicate volumes dispensed by the
pipette, and converting the mass of water to volume
Error and Uncertainty

 This, however, raises other sources of error:

 Each weight will have its own associated error
 The operator will not use the pipette in exactly the
same way every time, introducing additional error
 To do the calculation, we need to measure the
temperature, which also has an associated error
 Evaporation losses, and changes in temperature and
humidity can also contribute to variation in the
measured volumes
Error and Uncertainty

 Clearly, it is unrealistic to try and account for all these

errors just to perform a titration
 We therefore use an estimate of the error in the volume
dispensed by the pipette, which we term as uncertainty
 Similarly, any measured value has an associated
measurement uncertainty, which is used as an estimate
of the range within which the error falls either side of
the actual value
 Since we cannot easily tell whether the result is above
or below the true value, such uncertainties are treated
in the same way as random errors
Residuals

 Residual is simply the difference between a single

observed value and the sample mean, (xi – )
 It has both sign and units
 Residuals can provide a useful comparison between
successive individual values within a set of
measurements, particularly when presented visually
in the form of a residual plot
Residuals
Residuals

 Such plots can reveal useful information about the

quality of the data set such as
 Whether there is a systematic drift in an instrument
under calibration, or
 If there might be cross-contamination between samples
of high and low concentration
Reporting Analytical Data

 Data are generated using appropriate procedures or

test plans
 Data generated is certified by analyst to be correct to
the best of the analyst's knowledge, with the
certification documented by a dated signature or
initials on the data
 If errors are found during the generation of the data,
the analyst corrects the errors. An explanation is
required if the reason for the correction is not
obvious
Reporting Analytical Data

 Significant Figures
 Significant figures reflect the accuracy and precision of
a given result
 A result should always be rounded to the number of
figures that are consistent with the confidence that can
be placed on it
 Thus, the number of significant figures is the number
of digits remaining after the data is rounded
 Reported values should contain only significant figures
Reporting Analytical Data

 Significant Figures
 A value is made up of significant figures when it
contains all digits known to be true and one last digit
in doubt
 E.g.,if a value is reported as 18.86 mg/l, the 18.8 must be
firm while the 0.06 is somewhat uncertain
 Final zeros after a decimal point are always meant to
be significant figures
 E.g. 9.800 g would be considered as having 4 significant
figures
Reporting Analytical Data

 Significant Figures
 Zeros before a decimal point with non-zero digits
preceding them are assumed to be significant
 With no preceding non-zero digit, a zero before the
decimal point is not considered significant
 With no non-zero digits preceding a decimal point, the
zeros after the decimal point but preceding other non-
zero digits are not considered to be significant. These
zeros only indicate the position of the decimal point
Reporting Analytical Data

 Significant Figures
 Final zeros in a whole number may or may not be
significant (e.g., in 1,000 g, zeros may not be
significant and only indicate magnitude of the number)
 A good way to determine if zeros interspersed in a
number are significant or not is by determining
whether they can be dropped by expressing the number
in exponential form
 E.g., no zeros can be dropped when expressing 100.08 g
in exponential form; therefore, the zeros are significant
 However, 0.0008 g can be expressed in exponential form
as 8 x 10−4 g; therefore, the zeros are not significant
Reporting Analytical Data

 Significant Figures
 Once the number of significant figures obtainable from
a given type of analysis is established, data resulting
from such analyses are reduced according to set rules
for rounding
Reporting Analytical Data

 Rounding Off Numbers

 If the figure following those to be retained is less than
5, the figure is dropped, and the retained figures are
kept unchanged (e.g., 11.443 is rounded off to 11.44)
 If the figure following those to be retained is greater
than 5, the figure is dropped, and the last retained
figure is raised by 1 (e.g., 11.446 is rounded off to
11.45)
Reporting Analytical Data

 Rounding Off Numbers

 If the figure following those to be retained is 5, and if
there are no figures other than zeros beyond the five,
the figure 5 is dropped, and
 The last-place figure retained is increased by one if it is
an odd number (e.g. 11.435 is rounded off to 11.44) or
 The last placed figure is kept unchanged if it is an even
number (E.g. 11.425 is rounded off to 11.42)
Reporting Analytical Data

 Ways of Expressing Accuracy

 Absolute Errors: difference between true and measured
values
 Mean Errors: difference between true and measured
mean values
Reporting Analytical Data

 Ways of Expressing Accuracy

 Relative Error: Absolute or Mean Errors expressed as a
percentage of the true value
 = % Relative Error
 Relative Accuracy: measured or mean value expressed
as a percentage of true value
 = % Relative Accuracy
Reporting Analytical Data

 The main measures in quantitative statistics are:

 Measures of central tendency
 Mean
 Median
 Mode
 Measures of dispersion
 Variance
 Standard deviation
 Relative standard deviation
 Standard error of mean

 These measures form the basis of any statistical analysis

Population vs. Sample Mean & Standard Deviation

 If we make only a limited number of measurements

(called replicates), some will be closer to the ‘true’
value than others
 This is because there can be variations in
 The amount of chemical being measured (e.g. as a
result of evaporation or reaction)
 The actual measurement itself (e.g. due to random
electrical noise in an instrument, or fluctuations in
ambient temperature, pressure, or humidity)
Population vs. Sample Mean & Standard Deviation

 This variability contributes to dispersion in the

measured values
 Greater the variability greater likelihood of all the
measured values differing significantly from ‘true’ value
 To adequately take the variability into account and
determine the actual dispersion, all possible values will
have to be obtained
 Infinite number of replicate measurements required
 n
 This would allow us to determine the population mean
and standard deviation, μ and 
Population vs. Sample Mean & Standard Deviation

 This is hardly practical, for a number of reasons!

 Therefore limited number of replicate measurements
are performed
 On the same sample, using the same instrument and
method, under the same conditions
 This allows us to calculate the sample mean and
standard deviation, and s
 The sample mean, standard deviation, and variance
(s2) provide estimates of the population values
Mean

 Used for describing an entire set of observations with

a single value representing the center of the data
 Many statistical analyses use the mean as a standard
reference point
 The mean is the sum of all observations divided by
the number of observations
 It has the same units as each individual measurement
value
 Denoted by μ or
Calculating the Mean

 The sample mean is the average value for a finite set

of replicate measurements on a sample
 It provides an estimate of the population mean for the
sample using the specific measurement method
 The sample mean, denoted , is calculated using the
formula:
Median

 Used for describing an entire set of observations with

a single value representing the center of the data
 Half of the observations are above the median, half
are below it
 Determined by ranking the data and finding the
middle observation
Mode

 It is the value that occurs most frequently in a set of

observations
 It may be used with mean and median to give an
overall characterization of your data distribution
 Identifying mode can help you understand your
distribution
 A distribution with more than one mode may indicate
that you actually have sampled from a mixed
population
Variance (2)

 Represents the spread (the dispersion) of the repeated

measurements on either side of the mean
 As the notation implies, the units of the variance are
the square of the units of the mean value
 The greater the variance, the greater the probability
that any given measurement will have a value
noticeably different from the mean
Calculating the Variance

 It is the average of the squared differences between

each measurement and the sample mean

 When n is sufficiently large so that n  (n – 1), the

sample mean and variance approximate the
population values and we can use the equation:
Standard Deviation ()

 Also provides a measure of the spread of repeated

measurements on either side of the mean
 An advantage of the standard deviation over the
variance is that its units are the same as those of the
measurement
Standard Deviation ()

 Standard deviation – the most important statistic

 Standard Deviation () of an infinite set of
experimental data is theoretically given by
 =

 xi = individual measurement
 = mean of infinite number of measurements (true
value)
 n = number of measurements
Standard Deviation ()

 Standard deviation
 Estimated Standard Deviation, s (n < 30)
 s=
 For finite sets the precision is represented by ‘s’
Calculating the Standard Deviation

 It is the square root of average of the squared

differences between each measurement and the
sample mean

 When n is sufficiently large so that n  (n – 1), the

sample mean and standard deviation approximate the
population values and we can use the equation
Reporting the Results

 It is more convenient to use the standard deviation,

which is simply the square root of the variance,

 The final value for the sodium content of the soup

would be written as:
C = 102.1 ± 4.7 mg (mean ± s, n = 5)
Relative Standard Deviation

 It tells whether the standard deviation is a small or

large quantity when compared to the mean for the
data set
 It is calculated by the following formula:

 100%
Relative Standard Deviation

 Relative Standard Deviation and Coefficient of

Variation
 In some cases, the coefficient of variation and the RSD
are the same thing
 However, RSD cannot be negative while the
Coefficient of Variation can be positive or negative
 This is because the two formulas differ in a minor way
 Coefficient
of Variation divides by the mean
100%
 RSD divides by the absolute value of the mean

100%
Standard Error of mean

 It is the standard deviation of the means of samples

from population mean
 Standard deviation of the mean smean

SEM (Smean) =
Test of Significance

 Analytical method should be free from bias

 E.g. the value it gives for the amount of analyte should
be the true value
 However, even if no systematic errors, the measured
amount will not be exactly equal to the standard due
to random errors
 Significance test helps to decide whether the
difference between the measured and standard
amounts is significant or can it be accounted for by
random errors
Test of Significance

 Types of significance tests

 One-sided and two-sided tests
 Comparison of two experimental means
 Unpaired t-test
 Paired t–test
 F-test
 Analysis of variance
 Chi-squared test