2.

MEASUREMENT AND ERROR

An engineer has to make a lot of measurements, collect and analyze data, and make decisions
about the validity of his approaches and procedures. He must have a clear idea about the
results he is going to obtain. In this respect, he may develop models of his expectations and
compare the outcomes from the experiments to those from the model. He uses various
measuring instruments whose reliabilities have outmost importance in successes of his
decisions. Characteristics of measuring instruments that are used in selecting the proper ones
are reviewed in the first section. Section 2 deals with analyses of measurement data. Section 3
handles the analyses of uncertainties and establishment of engineering tolerances.

2.1. CHARACTERISTICS OF MEASURING INSTRUMENTS

2.1.1. Definition of Terms

The characteristics of measuring instruments are specified using terms shortly defined below.
The full description of some of these terms will be provided later with examples.

True value: standard or reference of known value or a theoretical value

Accuracy: closeness to the true value; closeness with which an instrument reading
approaches the true or accepted value of the variable (quantity) being measured. It is
considered to be an indicator of the total error in the measurement without looking into the
sources of errors.

Precision: a measure of the reproducibility of the measurements; given a fixed value of a

variable, precision is a measure of the degree to which successive measurements differ from
one another i.e., a measure of reproducibility or agreement with each other for multiple trials.

Sensitivity: the ability of the measuring instrument to respond to changes in the measured
quantity. It is expressed as the ratio of the change of output signal or response of the
instrument to a change of input or measured variable.

Resolution: the smallest change in measured value to which the instrument will respond, i.e.
the smallest incremental quantity that can be reliably measured.

Error: deviation from the true value of the measured variable.

 Measurement & Error / 2

Linearity: the percentage of departure from the linear value, i.e., maximum deviation of the
output curve from the best-fit straight line during a calibration cycle.

Tolerance: maximum deviation allowed from the conventional true value. It is not possible to
build a perfect system or make an exact measurement. All devices deviate from their ideal
(design) characteristics and all measurements include uncertainties (doubts). Hence, all
devices include tolerances in their specifications. If the instrument is used for high-precision
applications, the design tolerances must be small. However, if a low degree of accuracy is
acceptable, it is not economical to use expensive sensors and precise sensing components.

2.1.2. Static Calibration

The static calibration for a multi-input instrument is carried out by keeping all inputs except
one at some constant values. The single input under study is varied over some range of
constant values, causing the output(s) to vary over some range of constant values. The input-
output relation developed in this way is called “static calibration”. A calibration curve for a
dual-input single-output system is shown in figure 2.1. The static sensitivity (S) is the slope of

B = B2 Output A = A2
Input A
Measuring Outpu t
Input B Device
B = B1 A = A1
O/p

Input A Input B
Output (o/p) =
Sensitivity (S) x input (i/p)
Non-linear i/p-o/p relation Linear i/p-o/p relation

Figure 2.1. Static calibration curves for a multi-input single-output system.

the calibration curve and is defined as,

S = (output)/(input) ........................................................................................(2.1)

S is a constant for linear relation. Otherwise, S is a function of the input.

2.1.3. Accuracy and Precision

A measurement isn’t very meaningful without an error estimate! No measurement made is
ever exact. The accuracy (correctness) and precision (number of significant figures) of a
measurement are always limited by apparatus used, skill of the observer and the basic physics
in the experiment and the experimental technique used to access it. The goal of the
 Characteristics of Measuring Instruments / 3

experimenter is to obtain the best possible value of some quantity or to validate/falsify a

theory. What comprises a deviation from a theory? Every measurement must give the range of
possible value. In this section we will discuss the accuracy and precision with examples.

2.1.3.1. Accuracy
Accuracy is defined as the degree of conformity of a measured value to the true (conventional
true value – CTV) or accepted value of the variable being measured. It is a measure of the
total error in the measurement without looking into the sources of the errors. Mathematically
it is expressed as the maximum absolute deviation of the readings from the CTV. This is
called the absolute accuracy.

Absolute accuracy maximumdevation from CTV ......................................... (2.2)

absoluteaccuracy
Relative accuracy  ; ........................................................... (2.3)
CTV
Percentage(%)accuracy   relativeaccuracy100 ............................................ (2.4)

Example 2.1
A voltmeter is used for reading on a standard value of 50 volts, the following readings are
obtained: 47, 52, 51, 48

Conventional true value (CTV) = 50 volts,

Maximum (VMAX) = 52 volts and minimum (VMIN) = 47 volts.

CTV – VMIN = 50 – 47 = 3 volts; VMAX – CTV = 52 – 50 = 2 volts.

Absolute accuracy = max of {3, 2} = 3 volts.

Relative accuracy = 3/50 = 0.06 and % accuracy = 0.06x100 = 6%

2.1.3.2. Precision
Precision is composed of two characteristics as conformity and the number of significant
figures.

Conformity
The conformity is the ability of an instrument to produce the same reading, or it is the degree
of agreement between individual measurements. So, it is also called repeatability or
reproducibility.

Mathematically it is expressed as “the absolute maximum deviation from the average of

the readings”, i.e.
 Measurement & Error / 4

Precision (Pr) = max {(VAV – VMIN ), (VMAX –VAV )}.............................................. (2.5)

2.1.3.3. Bias
The difference between CTV and average value (VAV) is called the bias. Ideally, the bias
should be zero. For a high quality digital voltmeter, the loading error is negligible yielding
bias very close to zero.

Bias = CTV - VAV .............................................................................................. (2.6)

In the previous example the average (VAV) = (47+48+51+52)/4 = 49.5 V

Pr = max {(49.5 – 47), (52 – 49.5)} = 2.5 volts. Thus, Bias = 50 – 49.5 = 0.5 volt.

A consistent bias can be due to the presence of a systematic error or instrument loading.
Hence, eliminating the causes removes the bias. However, if the bias is consistent and causes
cannot be identified and/or eliminated, the bias can be removed by re-calibrating the
instrument.

Example 2.2
A known voltage of 100 volts (CTV = 100 V) is read five times by a voltmeter and following
readings are obtained: 104, 103, 105, 103, 105

Average reading = (1/5)x(104+103+105+103+105) = 104 volts

Pr = max {(VAV – VMIN), (VMAX – VAV)} = max {(104 – 103), (105 – 104)} = 1 volt

Accuracy = max {(CTV – VMIN), (VMAX - CTV)} = max {(100 – 103), (105 – 100)} =5 V

Bias = average – CTV= 104 – 100 = 4 volts.

If we re-calibrated the instrument to remove the bias, then the average reading = CTV.
The new readings would be 100, 99, 101, 99, 101

Hence, after re-calibration, average = CTV = 100 volts, and accuracy = precision = 1 volt.

2.1.4. Accuracy versus Precision

The distinction between accuracy and precision can be illustrated by an example: two
voltmeters of the same make and model may be compared. Both meters have knife-edged
pointers and mirror-backed scales to avoid parallax, and they have carefully calibrated scales.
They may therefore be read to the same precision. If the value of the series resistance in one
meter changes considerably, its readings may be in error by a fairly large amount. Therefore
 Characteristics of Measuring Instruments / 5

the accuracy of the two meters may be quite different. To determine which meter is in error, a
comparison measurement with a standard meter should be made.

Accuracy refers to the degree of closeness or conformity to the true value at the quantity
under measurement. Precision refers to the degree of agreement within a group of
measurements or instruments. The target-shooting example shown in figure 2.2 illustrates the
difference. The high accuracy, poor precision situation occurs when the person hits all the
bullets on a target plate on the outer circle and misses the bull’s eye. In the second case, all

Poor accuracy High accuracy Average accuracy Poor accuracy

High precision High precision Poor precision Poor precision
Figure 2.2. An illustration of accuracy and precision.

bullets hit the bull’s eye and spaced closely enough leading to high accuracy and high
precision. The bullet hits are placed symmetrically with respect to the bull’s eye in the third
case but spaced apart yielding average accuracy but poor precision. In the last example, the
bullets hit in a random manner, hence poor accuracy and poor precision.

The scatter graph in figure 2.3 shows an alternative way of presenting the accuracy and
precision. Same quantity was measured three times by 5 different analyst or methods or
measuring instruments. Distribution of readings around the true value indicates the most
accurate, most precise and least accurate and least precise readings. The last reading is too far
away from the true value and from other readings that may indicate a systematic error.

accurate

true value

error?
Worst

Figure 2.3. An illustration of accuracy and precision by a scatter graph.

 Measurement & Error / 6

Precision is composed of two characteristics as stated: conformity and the number of

significant figures to which a measurement may be made. Consider, for example, that the
insulation resistance of a transformer has the true value 2,475,653 . It is measured by an
ohmmeter, which consistently and repeatedly indicates 2.5 M. But can the observer "read"
the true value from the scale? His estimates from the scale reading consistently yield a value
of 2.5 M. This is as close to the true value as he can read the scale by estimation. Although
there are no deviations from the observed value, the error produced by the limitation of the
scale reading is a precision error. The example illustrates that conformity is a necessary, but
not sufficient, condition for precision because of the lack of significant figures obtained.
Similarly, precision is a necessary, but not sufficient condition for accuracy.

Too often the beginning student is inclined to accept instrument readings at face value.
He is not aware that the accuracy of a reading is not necessarily guarantied by its precision. In
fact, good measurement technique demands continuous skepticism as to the accuracy of the
results.

In critical work, good practice dictates that the observer make an independent set of
measurements, using different instruments or different measurement techniques, not subject to
the same systematic errors. He must also make sure that the instruments function properly and
are calibrated against a known standard, and that no outside influence affects the accuracy of
his measurements.

2.1.5. Significant Figures

An indication of the precision of the measurement is obtained from the number of significant
figures in which the result is expressed. Significant figures convey actual information
regarding the magnitude and the measurement precision of a quantity. The more significant
figures the greater the precision of measurement.

Figure 2.4 illustrates the importance of significant figures

66 68 70
with an example. If a resistor is specified as having a
resistance of 68 , its resistance should be closer to 68 
67.8 68.0 68.2
than to 67 or 69 . If the value of the resistor is described
as 68.0 , it means that its resistance is closer to 68.0 than
Figure 2.4. An illustration of
significant figures. it is to 67.9 or 68.1 . In 68 there are two significant
figures; in 68.0 there are three. The latter, with more
 Characteristics of Measuring Instruments / 7

significant figures, expresses a measurement of greater precision than the former.

It is customary to record a measurement with all the digits of which we are sure nearest to
the true value. For example in reading a voltmeter, the voltage may be read as 117.1 V. This
simply indicates that the voltage, read by the observer to best estimation, is closer to 117.1 V
than to 117.0 V or 117.2 V. Another way of expressing this result is that it indicates the range
of possible error. The voltage may be expressed as 117.1 0.05 V, indicating that the value of
the voltage lies between 117.05 V and 117.15 V.

When two or more measurements with different degrees of accuracy are added, the result
is only as accurate as the least accurate measurement. Consider the following example:

Example 2.3
Two resistors, R1 and R2, are connected in series. Individual resistance measurements using a
digital multimeter, yield R1 = 18.7 and R2 = 3.624 . Calculate the total resistance to the
appropriate number of significant figures.

SOLUTION
R1 = 18.7 (three significant figures)
R2 = 3.624 (four significant figures)
RT = R1 + R2 = 22.324 (five significant figures) = 22.3 

The doubtful figures are written in italic. Any digit in the result is doubtful if it’s
computation involves doubtful digits. In the addition of R1 and R2 the last three digits of the
sum are doubtful figures. There is no value whatsoever in retaining the last two digits (the 2
and the 4) because one of the resistors is accurate only to three significant figures or tenths of
an ohm. The result should therefore also be reduced to three significant figures or the nearest
tenth. i.e., 22.3 . Note that if extra digits accumulate in the answer, they should be discarded
or rounded off. In the usual practice, if the digit in the first place to be discarded (most
significant of digits to be discarded) is less than five, it and the following digits are dropped
from the answer as it was done in the example. If the digit in the first place to be discarded is
five or greater, the previous digit is increased by one. For three-digit precision, therefore,
22.324 should be rounded off to 22.3; and 22.354 to 22.4.

2.1.6. Types of Errors (Uncertainties)

No measurement can be made with perfect accuracy, but it is important to find out what the
accuracy actually is and how different errors have entered into the measurement. A study of
 Measurement & Error / 8

errors is a first step in finding ways to reduce them. Such a study also allows us to determine
the accuracy of the final test result. Errors may come from different sources and are usually
classified under three main headings as:

Gross errors: largely human errors, among them misreading of instruments, incorrect
adjustment and improper application of instruments, and computational mistakes.

Systematic (determinate) errors: shortcomings of the instruments, such as defective or worn

parts, and effects of the environment on the equipment or the user. They are sometimes called
bias due to error in one direction- high or low. They are generally originated from a known
cause such as result from mis-calibrated device, experimental technique that always gives a
measurement higher (or lower) than the true value, operator’s limitations and calibration of
glassware, sensor, or instrument. Their effects can be minimized by trying a different method
for the same measurement. They can be corrected when determined.

Systematic errors may be of a constant or proportional nature as illustrated in figure

2.5. The constant error influences the
intercept while the proportional error
influences the slope.

Random (indeterminate) errors: those

due to causes that cannot be directly
established because of random variations in
the parameter or the system of
measurement. Hence, we have no control
over them. Their random nature causes
Figure 2.5. Constant and proportional type errors
both high and low values to average out.
Multiple trials help to minimize their effects. We deal with them using statistics.

Figure 2.6 provides a schematic summary of errors and their possible means of
elimination. For example, errors caused by the loading effect of the voltmeter can be avoided
by using it intelligently. A low-resistance voltmeter should not be used to measure voltages at
the input of a voltage amplifier. In this particular measurement, a high-input impedance
voltmeter (such as a digital voltmeter - DVM) is required. Gross and systematic errors cannot
be treated mathematically. They can be avoided only by taking care in reading and recording
the measurement data. Good practice requires making more than one reading of the same
quantity, preferably by a different observer. Never place complete dependence on one reading
 Characteristics of Measuring Instruments / 9

but take at least three separate readings, preferably under conditions in which instruments are
switched off-on.

The error may be originated from the sampling of the source, preparation of the samples

Measurement errors

Human errors Systematic errors Random errors

(Gross errors)

Equipment errors Environmental errors

Examples:
Examples:
Misreading instruments
Unknown events
Erroneous calculations
that cause small
Improper choice of Examples: Examples: variations in
instrument Bearing friction Changes in measurements.
Incorrect adjustment, or Component nonlinearities temperature, Quite random and
forgetting to zero Calibration errors humidity, stray
unexplainable.
Neglect of loading effects Damaged equipment electric and
Loss during transmission magnetic fields.

How to estimate:
How to estimate: How to estimate: Take many readings
Not possible to estimate
1. Compare with more Careful monitoring of and apply statistical
their value
accurate standards changes in the analysis to unexplained
mathematically
2. Determine if error is variables. Calculating variations
constant or a expected changes.
proportional error

Methods of elimination or
reduction: Methods of reduction Methods of reduction or Methods of
1. Careful attention to or elimination: elimination: reduction:
detail when making 1. Careful calibration 2. Hermetically seal 1. Careful design
measurements and of instruments. equipment and of measurement
calculations. 2. Inspection of components under test. apparatus to
2. Awareness of equipment to ensure 3. Maintain constant reduce unwanted
instrument limitations. proper operation. temperature and interference.
3. Use two or more 3. Applying humidity by air 2. Use of statistical
observers to take critical correction factors conditioning. evaluation to
data. after finding 4. Shield components determine best
4. Taking at least three instrument errors. and equipment against true estimate of
readings or reduce 4. Use more than one stray magnetic fields. measurement
possible occurrences of method of measuring 5. Use of equipment that readings.
gross errors. a parameter. is not greatly effected
5. Be properly motivated by the environmental
to the importance of changes.
correct results.

Figure 2.6. A schematic summary of measurement errors.

 Measurement & Error / 10

and measurement and analysis of the measurand. Care must be taken so that the sample is
representative of the whole population (homogeneous vs. heterogeneous). No unwanted
additions or deletions are allowed during the preparatory phase. Finally, calibration of the
measuring instrument using standard measurands or standard solutions is done as frequent as
defined by the equipment manufacturer. One way to assess total error is to treat a reference
standard as a sample. The reference standard would be carried through the entire process to
see how close the results are to the reference value.

2.2. ANALYSIS OF MEASUREMENT DATA

A statistical analysis of measurement data is common practice because it allows an analytical
determination of the uncertainty of the final test result. The outcome of a certain measurement
method may be predicted on the basis of sample data without having detailed information on
all the disturbing factors. To make statistical methods and interpretations meaningful, a large
number of measurements are usually required. Also, systematic errors should be small
compared with residual or random errors, because statistical treatment of data cannot remove
a fixed bias contained in all the measurements.

2.2.1. Arithmetic Mean

The most probable value of a measured variable is the arithmetic mean of the number of
readings taken. The best approximation will be made when the number of readings of the
same quantity is very large. Theoretically, an infinite number of readings would give the best
result although in practice only a finite number of measurements can be made. The arithmetic
mean is given by:

 
x x x x x
 1 2 3
n
........................................................................ (2.7)
x
n n

where x = arithmetic mean, x1 . . . xn = readings taken, and n = number of readings.

Example 2.4
A set of independent current measurements was taken by six observers and recorded as 12.8
mA, 12.2 mA, 12.5 mA, 13.1 mA, 12.9 mA, and 12.4 mA. Calculate the arithmetic mean.

12.8 12.2 12.5 13.112.9 12.4

x  12.65mA
6
 Analyses of Measurement Data / 11

2.2.2. Deviation from the Mean Table 2.1. Deviations around mean
In addition to knowing the mean value of a series of d1 = 12.8 - 12.65 = 0.15 mA
measurements, it is often informative to have some idea d2 = 12.2 - 12.65 = -0.45 mA
of their range about the mean. Deviation is the departure d3 = 12.5 - 12.65 = -0.15 mA
of a given reading from the arithmetic mean of the group d4 = 13.1 - 12.65 = 0.45 mA
of readings. If the deviation of the first reading x1 is d5 = 12.9 - 12.65 = 0.25 mA
called d1, and that of the second reading, x2 is called d2 d6 = 12.4 - 12.65 = -0.25 mA
and so on, then the deviations from the mean can be
expressed as

d1 x1 x ; d2 x2 x ; ; d n xn x ........................................................................ (2.8)

The deviation from the mean may have a positive or a negative value and that the algebraic
sum of all the deviations must be zero. The computation of deviations for the previous
example is given in Table 2.1.

2.2.2.1. Average Deviation

The average deviation is an indication of the precision at the instruments used in making the
measurements. Highly precise instruments will yield a low average deviation between
readings. By definition average deviation is the sum of the absolute values of the deviations
divided by the number of readings. The absolute value of the deviation is the value without
respect to sign. Average deviation may be expressed as

D  d1  d2  d3  dn 
 d
.................................................................. (2.9)
n n

Example 2.5
The average deviation for the data given in the above example:

0.15 0.45 0.15 0.45 0.25 0.25

D  0.283mA
6

2.2.2.2. Standard Deviation

The range is an important measurement. It indicates figures at the top and bottom around the
average value. The findings farthest away from the average may be removed from the data set
without affecting generality. However, it does not give much indication of the spread of
observations about the mean. This is where the standard deviation comes in.

In statistical analysis of random errors, the root-mean-square deviation or standard

deviation is a very valuable aid. By definition, the standard deviation of a finite number of
 Measurement & Error / 12

data is the square root of the sum of all the individual deviations squared, divided by the
number of readings minus one. Expressed mathematically:

 1 2 3 n
 ............................................................ (2.10)
d d d d
2 2
1
2 2
 i

Another expression for essentially the same quantity is the variance or mean square
deviation, which is the same as the standard deviation except that the square root is not
extracted. Therefore
variance (V) = mean square deviation = 2 ........................................................ (2.11)
The variance is a convenient quantity to use in many computations because variances are
additive. The standard deviation however, has the advantage of being of the same units as the
variable making it easy to compare magnitudes. Most scientific results are now stated in terms
of standard deviation.

2.2.3. Probability of Errors

2.2.3.1. Normal Distribution of Errors
A practical point to note is that, whether the calculation is done on the whole “population” of
data or on a sample drawn from it, the population itself should at least approximately fall into
a so called “normal (or Gaussian)” distribution.
For example, 50 readings of voltage were taken at small time intervals and recorded to

20

16
Number of Observed Readings

12

0
99.6 99.8 100.0 100.2 100.4
Volts
 Analyses of Measurement Data / 13

the nearest 0.1 V. The nominal value of the measured graphically in the form of a block
diagram or histogram in which the number of observations is plotted against each observed
voltage reading. The histogram and the table data are given in figure 2.7. The figure shows
that the largest number of readings (19) occurs at the central value of 100.0 V while the other
readings are placed more or less symmetrically on either side of the central value. If more
readings were taken at smaller increments, say 200 readings at 0.05-V intervals, the
distribution of observations would remain approximately symmetrical about the central value
and the shape of the histogram would be about the same as before. With more and more data
taken at smaller and smaller increments, the contour of the histogram would finally become a
smooth curve as indicated by the dashed line in the figure. This bell-shaped curve is known as
a Gaussian curve. The sharper and narrower the curve, the more definitely an observer may
state that the most probable value of the true reading is the central value or mean reading.

For unbiased experiments all observations include small disturbing effects, called random
errors. Random errors undergo a Normal (Gaussian) law of distribution shown in figure 2.8.

1 x2
Pr obability of error  exp( )
2 2
Probability of Error


 Area Under the Probability Curve

 Deviation  Fraction of total area


0.6745 0.5000


1.0 0.6828
 2.0 0.9546
3.0 0.9972 -4 -3 -2 -1 0 1 2 3 4
Error (standard deviation - sigma)

They can be positive or negative and there is equal probability of positive and negative
random errors. The error distribution curve indicates that:

Small errors are more probable than large errors.

Large errors are very improbable.
There is an equal probability of plus and minus errors so that the probability of a given
error will be symmetrical about the zero value.
 Measurement & Error / 14

Table 2.2. Deviations in readings

Deviation The error distribution curve in figure 2.8 is based on
Reading, x
d d2 the Normal (Gaussian) law and shows a symmetrical
101. -0.1 0.01
101.7 0.4 0.16 distribution of errors. This normal curve may be regarded
101.3 0.0 0.00 as the limiting form of the histogram in which the most
101.0 -0.3 0.09
101.5 0.2 0.04 probable value of the true voltage is the mean value of
101.3 0.0 0.00 100.0V. Table 2.2 lists the readings, deviations and
101.2 -0.1 0.01
101.4 0.1 0.01 deviation squares of readings from the mean value. The
101.3 0.0 0.00 reason why the standard deviation is such a useful
101.1 -0.2 0.04
measure of the scatter of the observations is illustrated in
x=1013.0 d=1.4 d2=0.36
the figure. If the observations follow a “normal”
distribution, a range covered by one standard deviation above the mean and one standard
deviation below it (i.e. x 1 SD) includes about 68% of the observations. A range of 2

standard deviations above and below ( x 2 SD) covers about 95% of the observations. A

range of 3 standard deviations above and below ( x 3 SD) covers about 99.72% of the
observations.

2.2.3.2. Range of a Variable

If we know the mean and standard deviation of a set of observations, we can obtain some
useful information by simple arithmetic. By putting 1, 2, or 3 standard deviations above and
below the mean we can estimate the ranges that would be expected to include about 68%,
95% and 99.7% of observations. Ranges for SD and 2 SD are indicated by vertical lines.
The table in the inset (next to the figure) indicates the fraction of the total area included within
a given standard deviation range.

Acceptable range of possible values is called the confidence interval. Suppose we

measure the resistance of a resistor as (2.65 ± 0.04) k. The value indicated by the color code
is 2.7 k. Do the two values agree? Rule of thumb: if the measurements are within 2 SD, they
agree with each other. Hence, 2 SD around the mean value is called the range of the
variable.

2.2.3.3. Probable Error

The table also shows that half of the cases are included in the deviation limits of 0.6745.
The quantity r is called the probable error and is defined as
probable error r = 0.6745 .......................................................... (2.12)
 Analyses of Measurement Data / 15

This value is probable in the sense that there is an even chance that any one observation
will have a random error no greater than r. Probable error has been used in experimental
work to some extent in the past, but standard deviation is more convenient in statistical work
and is given preference.

Example 2.6
Ten measurements of the resistance of a resistor gave 101.2 , 101.7 , 101.3 , 101.0 ,
101.5 , 101.3 , 101.2 , 101.4 , 101.3 , and 101.1 . Assume that only random errors
are present. Calculate the arithmetic mean, the standard deviation of the readings, and the
probable error.

SOLUTION: With a large number of readings a simple tabulation of data is very convenient
and avoids confusion and mistakes.

Arithmetic mean, x   1013.0

x
= 101.3 
n 10

Standard deviation,   = 0.2 

9

Probable error = 0.6745 = 0.6745 x 0.2 = 0.1349 


2.2.4. Some MS Excel Functions
The electronic spreadsheet program Microsoft Excel offers many built-in statistical functions
that can be used in data analysis. They can be easily accessed from the “insert function”
menu. The salient ones are:

 =SUM(A2:A5) Find the sum of values in the range of cells A2 to A5.
 .=AVERAGE(A2:A5) Find the average of the numbers in the range of cells A2 to A5.
 =AVEDEV(A2:A5) Find the average deviation of the numbers in the range of cells
A2 to A5.
 =STDEV(A2:A5) Find the sample standard deviation (unbiased) of the numbers in
the range of cells A2 to A5.
 =STDEVP(A2:A5) Find the sample standard deviation (biased) of the numbers in
the range of cells A2 to A5.

2.2.5. Determining Random Errors

Random errors are due to random variations in the parameter or the system of measurement as
mentioned before. We deal with them using statistics and multiple trials generally help to
 Measurement & Error / 16

minimize their effects. One of their primary causes can be pinpointed to instrument limit of
error and least count. The least count is the smallest division that is marked on the instrument.
The instrument limit of error is the precision to which a measuring device can be read, and
is always equal to or smaller than the least count. The estimation of the uncertainty is
important. For example, assume a volt meter may give us 3 significant digits, but we observe
that the last two digits oscillate during the measurement. What is the error? Average deviation
or standard deviation based on repeated measurements of the same quantity are used in
determining the uncertainty.

2.2.5.1. Uncertainties in Reading Digital Displays

A digital meter involves counting
(a) (b) from a clock signal during the
Gate gate interval as depicted in figure

Clock 2.9. As the gate and clock signals

are not synchronized and
Figure 2.9. Unsynchronized gate and clock signals.
combined in an AND gate, case
(b) results 4 pulses while case (a) supplies only 3 pulses. Hence, a digital read-out has an
uncertainty of 1 digit.

2.2.5.2. Uncertainties in Reading Analog Displays

The uncertainty in analog displays depends upon the organization of display screen and
capabilities of the reader. In analog multi meters it is accepted as ½ scale divisions (the least
count). In oscilloscope displays, it depends upon the thickness of the trace and it is around ½
mm. For both analog and digital displays, it is recommended to take the measurement as close
to full scale as possible to minimize the effect of the reading error. The following example
illustrates the uncertainties in analog meter readings.

Example 2.7
An analog voltmeter is used to measure a voltage. It has 100
5.8 6.0 6.2
divisions on the scale. The voltage read is 6 volts and the
meter has two ranges as 0 – 10 volts and 0 – 100 volts. Find
4 6 8 the uncertainty in the measured value in both ranges.
Uncertainty = ½ VFSD / # of divisions, where VFSD is the
Figure for example 2.7. voltage measured at full-scale deflection of the meter.
On 10 V range, uncertainty = ½ 10 / 100 = 0.05 V
 Analyses of Measurement Data / 17

yielding V = 6 0.05 volt.

On 100 V range, uncertainty = ½ 100 / 100 = 0.5 V yielding V = 6 0.5 volt.
Relative uncertainty: on 10 V range, 0.05/6 = 1/120 = 0.0083;
on 100 V range, 0.5/6 = 1/12 = 0.083
Percentage uncertainty: on 10 V range, (0.05/6)x100 = 0.83%, and

on 100 V range, (0.5/6)x100 = 8.3%

Exercise(adapted from http://www.hep.vanderbilt.edu/~julia/VUteach/PHY225a)

For each of the three rulers in figure 2.10, determine and record
 The least count of the scale (smallest division) – scales are all in cm
 Length of the gray rods
 Uncertainties in your readings
 Compare your result with those of the student next to you

2.3. UNCERTAINTY ANALYSIS

Any system that relies on a measurement system will involve some amount of uncertainty
(doubt). The uncertainty may be caused by individual inaccuracy of sensors, limitations of the
display devices, random variations in measurands, or environmental conditions. The accuracy
of the total system depends on the interaction of components and their individual accuracies.
This is true for measuring instruments as well as production systems that depend on many
subsystems and components. Each component will contribute to the overall error, and errors
and inaccuracies in each of these components can have a large cumulative effect.

2.3.1. Mathematical Analysis of the Uncertainty

If an experiment has number of component sources, each being measured individually using
independent instruments, a procedure to compute the total accuracy is necessary. Let
 Measurement & Error / 18

R = f(x1, x2, x3, … , xn )...................................................................................... (2.13)

where x1, x2, x3, … , xn are independent variables. Each variable is defined as

xi xi xi ....................................................................................................... (2.14)

i = 1, 2, … , n; x is known as the nominal value; xiis known as the uncertainty in the

variable x1; then

R R R ........................................................................................................ (2.15)

whereR f (x1, x2 ,...., xn )

The uncertainty R = wR can be computed using Taylor’s series expansion and statistical
f
analysis. All partial derivatives of R are taken. The partial derivative shows the
xi
sensitivity of R to variable xi. Since the measurements have been taken, the xi values are
known and can be substituted into the expressions for the partial derivatives and partial
derivatives are evaluated at known values of x1, x2, . , xn.

2.3.1.1. Limiting Error

Two methods are commonly used for determining the uncertainty. The first one is called the
method of equal effects and it yields the limiting (guarantied) error (maximum uncertainty
possible).
n
f
R R  [ ]o xi ................................................................................... (2.16)
i 1 xi

f
where ( )o is the partial derivative of the function with respect to xI calculated at the
xi
nominal value. The absolute value is used because some of the partial derivatives may be
negative and would have a canceling effect. If one of the partial derivative is high compared
to the others, then a small uncertainty in the corresponding variable has large effect on the
total error. Hence, the equation also illustrates which of the variable exerts strongest influence
on the accuracy of the overall results.

Example 2.8
The voltage generated by a circuit is equally dependent on the value of three resistors and is
given by the following equation: V0 = I(R1R2/R3)
 Uncertainty Analysis / 19

If the tolerance of each resistor is 1 per cent, what is the maximum error of the generated
voltage?
SOLUTION:
V0 R2 R RR V
Let us find the sensitivities first. I V0 V0 I 1 V0 V0 I 1 2  0
 ;  ;
R1 R3 R1 R2 R3 R2 R3
2
R3 R3
All tolerances are given as 1%, therefore: R1 = 0.01R1 ; R2 = 0.01R2 ; R3 = 0.01R3
V0 V V
V0  R1  0 R2  0 R3 That yields V0 = 0.03V0
R1 R2 R3
The total variation of the resultant voltage is 0.3 per cent, which is the algebraic sum of
the three tolerances. This is true in the first approximation. The maximum error is slightly
different from the sum of the individual tolerances. On the other hand, it is highly unlikely
that all three components of this example would have the maximum error and in such a
fashion to produce the maximum or minimum voltage. Therefore, the statistical method
outlined below is preferred.

2.3.1.2. Expected Value of Uncertainty

The second method is called the square root of sum of squares. It is based on the observations
stated before for the random errors. It yields the expected value of the uncertainty and
computed as
f
(R)2  2 
n
[( )2 ] (x )2 ................................................................................ (2.17)
R o i
i 1 xi

This will be used throughout the course unless the question asks the limiting error, or
maximum possible uncertainty.

Example 2.9
P = VI, if V = 100 2 volt (measured) and I = 10 0.2 Amp (measured), determine the
maximum allowable uncertainty, and the expected uncertainty in power.

SOLUTION: Pm wPm  P V 1 0x2 watts is the

V 100x0 limiting
  .2 40

P

I
I
value of the uncertainty.
 P P
However, the expected uncertainty P w  ( V )2 ( I )2
V
P

wP  
 Measurement & Error / 20

 10 I

8 28.3 watts.
 Uncertainty Analysis / 21

The nominal value of power = 100x10 = 1000 watts

Percentage uncertainty = (28.3/1000)x100 = 2.83%, and P = 1000 28.3 watt.

Example 2.10
The resistance of a certain size of copper wire is given by R Ro[1 (T 20)] . The
resistance at 20C is Ro = 60.3%, temperature coefficient = 0.004/C 1%,
temperature T = 30C 1C. Compute the uncertainty in the resistance R.

SOLUTION: the nominal value of R, R 6[1 (0.004)(30 20)] 6.24



R/R0 = 1 + (T – 20) = 1 + 0.004(30 – 20) = 1.04

R/= R0 [T – 20] = 6(30 – 20) = 60; R/T = R0 = 6x0.004 = 0.024

Uncertainty in the nominal value of R0 = percentage uncertainty of R0 x nominal R0

Ro = (0.3/100) x 6 = 0.018;  = (1/100)(0.004) = 4x10-5/C; T = 1C

The uncertainty in the resistance R is given by:

 R  = 0.0305(0.0305/6.24)x100 =0.49%

If the maximum error in the resistance is asked, it can be found as:

Rm 1.04x0.018 60x4x105 0.024x1 0.045

2.3.1.3. Special Case
R Y Y Y ................................................................................................................... (2.18)
l n k

1 2 3

then

R Y Y Y
( )2 l 2 ( 1
)2 n2 ( 2
)2 k 2 ( 3
)2 ........................................................................ (2.19)
Ro Y1o Y2o Y3o

2.3.2. Series and Parallel Analysis

Example 2.11
Two resistors R1 and R2 are connected first in series, then in parallel. Let R1 = 10 0.5
and R2 = 10 0.5. Find the maximum and expected values for the uncertainty in the
combination.

Series analysis
Rs = R1 + R2 ; Rs/R1 = Rs/R2 = 1; Rs R1 R2 = 10 + 10 = 20
 Measurement & Error / 22

Rs R 1 1
The limiting error (maximum uncertainty) = R  R  s R   1
R1 R2
sm 1 2
2 2
Rs Rs
The uncertainty: (R )2 ( )2 (R )2 ( )2 (R )2 (1)2 ( 1 )2 (1)2 ( 1 )2 1 1 1
R1
s 1 2
R2
2 2 4 4 2

R1 R2

Two resistors in series.

Figure for example 2.11. Series and parallel connected resistors

yielding Rs 0.7. The relative uncertainty = 0.7/20 = 0.035, and the percentage
uncertainty = 3.5%. Therefore, Rs = 200.7= 203.5%

Parallel analysis
R1R2 10 x10
Rp  Rp Rp ; Rp  5
R1 R2 10 10
R (R R )R  R R R 2

p 1 2) 2 1 2
 2

R1 (R1 R 2 ) 2 (R1 R 2 ) 2

R (R R )R  R R R
2
R (R R )R R R 2

p 1 2) 2 1 2 R1
   
2 p 1 2) 1 1 2

R1 (R1 R 2 ) 2 (R1 R 2 ) 2 R2 (R1 R2 ) (R1 R2 ) 2
2

2 1 R
Hence, Rp  10 2  100   p

R1 (10 10) 400 4 R2
R p R p 1 1 1 1 1 1 1
(R )2 ( ) 2 (R ) 2 ( ) 2 (R )2 ( )2 ( )2 ( )2 ( )2 ( )( )(11) 
R1 R2
p 1 2
4 2 4 2 16 4 32

1
Therefore the uncertainty in Rp is : Rp  0.175

The nominal value of Rp = 5, the percentage uncertainty = (0.175/5)x100=3.5%

Then Rp = 5 0.175 = 53.5%
1
Limiting error in Rp = 1 ( 1  ) 0.25
4 2 2

2.3.3. Summary of how to propagate the errors

 Addition and subtraction ( x+y; x-y): add absolute errors;
 Uncertainty Analysis / 23

 Multiplication by an exact number (a*x): multiply absolute error by the number;

 Multiplication and division: add relative errors;
 Further explanations can be obtained from MathWorld -
http://mathworld.wolfram.com/ErrorPropagation.html.

Population Sample
2.3.4. Sample and Population Statistics
In many instances, we take samples from a population
and infer the population statistics as illustrated in figure
2.10. Suppose we want to know the average weight of
adults. It is not feasible to weigh every single adult and
then take the average of all the weights. All adults are Figure 2.10. Population and sample

called the population. Instead, we decide to take a small fraction of the adults, say 1 out of
every 1000, and average these weights. This small fraction is called our sample population.
Now we have an average weight for our sample population. We want to know if our sample
population average weight is a good estimation of the population average weight. In addition,
measurement is a costly process. Hence, we also want to know the minimum sample size that
yields uncertainties within the tolerance range.

Figure 2.11 illustrates the distribution for the population and the sample. For the

Estimated mean x-s Population standard

standard deviation s-x deviation 


Frequency



  + x
Mean

Figure 2.11. Normal distribution curves for population and sample.

normal distribution, 68% of the data lies within ±1 standard deviation. By measuring samples
and averaging, we obtain the estimated mean x , which has a smaller standard deviation sx. 
is the tail probability that xs does not differ from by more than .
 Measurement & Error / 24

The population standard deviation is

population   .................................................................. (2.20)

And the sample standard deviation is

( x – x) 2

sample s s  

i ......................................................... (2.21)
n –1

The sample standard deviation allows for more variation in the sample compared to the
population, since sample is only part of population. Dividing by (n-1) increases the estimate
of the population variation. This attempts to eliminate the possibility of bias.

The estimated sample standard deviation is a measure of the spread of data about the
mean. The standard deviation of the mean x is

s
 x  .................................................................................................................... (2.23)
n 1

Equation 2.23 illustrates an important fact. The standard deviation doesn’t change much,

but the error on the mean improves dramatically! It goes as  , where n is the number of

measurements. As a rule of thumb, the range R of the random variable x can be roughly taken
as R ≈ 4. If Δ is the error that can be tolerated in the measurement, then the number of

samples required to achieve the desired: n 

2
. Then x 
2
 Problems / 25
