2.

Theory of Errors

2.1. Introduction

Once a measurement has been carried out, we obtain a result of the measurement, which we define as
“the value of a measurand obtained by measurement”. This is the uncorrected result, or “the result of a
measurement before correction for assumed systematic errors” which is necessary to obtain the
corrected result. The corrected result is “the result of a measurement obtained after having made
corrections to the uncorrected result in order to take account of assumed systematic errors”. Correction
of measurement results mostly depend on resolving errors.

One kind of error is an absolute error of measurement, defined as “the result of a measurand minus the
(conventional) true value of the measurand”. A systematic error is “the component of the error of
measurement which, in the course of a number of measurements of the same measurand, remains
constant or varies in a predictable way”. Understanding these different types of errors helps us make a
correction. This is “the value which, added algebraically to the uncorrected result of a measurement,
compensates for an assumed systematic error”.

We deduce from this the accuracy of measurement, or “the closeness of agreement between the result
of a measurement and the (conventional) true value of the measurand” and the experimental standard
deviation, defined as for a series of n measurements of the same measurand, the parameter s
characterizing the dispersion of the results given by the formula:

xi being the result of the ith measurement and x_bar, being the arithmetic mean of the n results
considered.

Description of results: Displaying results in an equation must take into account measurement uncertainty.
This is “an assessment characterizing the range of values in which is found the true value of a measured
variable”.

These definitions help us display the results: X = Xo [S.I. unit] + ∆ X [S.I.unit] (probability %) To be rigorous,
a measurement result must contain the most probable value. The uncertainty range will include this
probable value and the probability associated with both this uncertainty range and the unit being used.

2.2. Types of errors

Measurement uncertainty is the difference between the true value of the measurand and the
measurement carried out by the sensor. The only known measurands are the standards with values that
have been determined by convention. It is important to distinguish between systematic errors and
random uncertainties: they occur for different reasons and have very different consequences for
measurement.
2.2.1. Systematic errors

Systematic errors are always due to faulty sensor knowledge or utilization. This kind of error is detected
by comparing the mean values of the same measurand, given by two different sensors. The most frequent
causes of systematic errors are: – incorrect or non-existent calibration due to an aging or altered sensor;
– incorrect usage: some examples include failure to reach steady state, a defect or error of one of the
conditioner parts, or a modification of the measurand by the sensor itself; – inadequate data processing:
examples are error in linearization in the measurement chain or amplifier saturation in the measurement
chain. Obviously, detecting systematic errors leads to their elimination.

2.2.2. Random uncertainties

We can know the cause of random uncertainties without being able to anticipate the measurement value.
Their evaluation is always statistical. They are due to the presence of signals or interference; the amplitude
of these is random and they are called, rather vaguely, “noise”.

Examples:

– fluctuating supply sources in the measurement chain or in the conditioner (such as fluctuation of the
electromotive force in a bridge);

– electromagnetic signals produced in an environment and picked up by a sensor element, conditioner or

– thermal fluctuation, including thermal turbulence of current carriers;

– fluctuation in influence variables, etc.

There are many other causes of random uncertainties, such as reading errors, defects in sensor mobility
and hysteresis. Unlike systematic errors, random errors can never be completely avoided. We can,
however, reduce them by using protection methods, such as electrical regulations, temperature
stabilization, mechanical isolation and electromagnetic shields.

2.3. Analyzing random errors and uncertainties

Because, by definition, random errors cannot be anticipated, they become part of statistics and as such
are labeled “uncertainties”. Because of this vagueness, statistics can prove a very useful tool. The British
Prime Minister Benjamin Disraeli once said to Queen Victoria: “There are three kinds of lies: lies, damned
lies, and statistics.” This witticism demonstrates why governments, for example, like using statistics.
Statistics cannot anticipate a specific occurrence, such as a measurement result, but only give the
probability of one value among many occurring. Statistics represent a crowd and not an individual; a
forest, not a tree. That is why, Disraeli might have added, governments use statistics to direct and support
as many actions as possible, without necessarily having to address or even be aware of specific problems.
In the case of measurement, two statistical problems must be analyzed [DIE 92]:

–the evaluation of uncertainties;

– deciding how to analyze and treat these uncertainties.