Engineering Physics – Discussion of Measurements, Error

and Uncertainties

Understanding Measurement and Graphing

The science of physics presents a logical, coherent and useful representation of nature. These
representations are sometimes called mathematical models because the language they are written
in is mathematics. The models allow us to convey ideas about the world in an accurate way and
in a way that can be understood by everyone.

Experimentation is the process by which we discover the exact nature of the mathematical
models that we use. In an experiment we define and determine the relationship between physical
characteristics of nature that have been observed. Measurement of those physical quantities
plays a crucial role in determining how nature behaves.

A measurement places a numerical value on some physical phenomenon or characteristic of a

phenomenon. A measurement consists of assigning a number along with appropriate units to
some property (or variable) of the system of our interest. This assignment is made after applying
a measuring instrument such as a scale, ruler, balance, or meter. We are always interested in
how accurate the measurement is and how precise is the instrument that we use to make the
measurement.

ACCURACY AND PRECISION

Suppose you wish to determine whether heating a metal rod changes the length of the rod.
You assume the rod has a fixed length at room temperature, and you first need to determine what
this length is. To measure the metal rod you place it next to a measuring scale, in this case a me-
ter stick. In determining the length of the rod, you will need to know something about the accu-
racy of your measurement and the precision of the meter stick.

The accuracy of the measurement indicates how well your measuring device, in this case,
the meter stick, agrees with a standard measuring device. If your measuring device is not prop-
erly calibrated with a standard, or if the device is out of calibration (e.g. the meter stick is
warped), then an error in measurement will occur whenever you use that measuring device. The
kind of error that occurs is called a systematic error. Systematic error consistently results in the
measurement being too large or too small. To reduce systematic errors due to calibration, the
measuring device should be compared with a standard and properly adjusted, if possible. The
experimenter can also make systematic errors by consistently reading the measuring device in-
correctly, for example. This kind of systematic error is reduced by independent measurements
by several experimenters.

The precision of a measuring instrument is determined by the fineness of the subdivisions on

the measuring scale of the instrument. For example, the smallest subdivisions on a meter stick
are millimeters. If you measure the length of the metal rod, you might observe that it is greater
than 0.523 meters (i.e., 523 mm), but less than 0.524 meters (524 mm). You can, however, esti-

Page 1 of 12
Last Update 8/20/2008
 Engineering Physics – Discussion of Measurements, Error
and Uncertainties
mate about where the end of the rod lies between these two marks. For example, it might lie
about halfway between the millimeter marks, in which case you could say that the rod was
0.5235 meters long (523 1/2 mm). Remember though, that you are uncertain about the last digit.
You are sure that the rod is longer than 0.523 and shorter than 0.524 meters. Your measurement
could be in error by as much a 0.5 mm. Assuming that you did everything else properly and
your measuring device was accurate (properly calibrated), then the maximum precision error of
the meter stick would be one-half of the smallest division, 0.5 mm or 0.0005 m. Sometimes this
precision error is expressed directly in the measurement as 0.5235 ± 0.0005 m.

SIGNIFICANT FIGURES
Of course each measurement that is made contains some precision error or uncertainty.
When computations are performed using these measurements, the experimenter must be aware of
how the uncertainty propagates through the calculation. Let’s look at an example of this. The
number 2.5004 is more precise than 2.5, 2.50, or 2.500. Each digit starting with 2.5 represents
an increasing precision to which the number is known. Since no measurement can be exactly
correct, the last digit is considered to be an estimate. So, when this last digit is used in any
mathematical operation, precision of a measurement cannot be gained, but it could be lost. To
avoid the propagation of uncertainty, significant figures are useful in keeping track of the preci-
sion of any measurements that are used in calculations.

Here are some guidelines that you can use to determine the number of significant figures in a
measurement and apply when you must perform mathematical operations using measurements
that vary in their precision.

1) All nonzero digits are significant; e.g., 4.583 has four significant figures.

2) All zeroes between two nonzero digits are significant, e.g., 101 contains three significant fig-
ures.

3) All zeroes to the left of a decimal and to the right of a nonzero digit are significant, e.g.,
100.0 contains four significant figures.

4) All zeroes to the right of the decimal point and to the right of a nonzero digit are significant,
2.010 has four significant figures.

5) The number of significant figures does not depend on the position of the decimal; e.g.,
0.00225 m or 0.225 cm both contain three significant digits.

6) In making measurements, record all significant figures. For example, when using a meter
stick a measurement of 75.0 cm can be made and should be recorded as such and not recorded as
75 cm.

Page 2 of 12
Last Update 8/20/2008
 Engineering Physics – Discussion of Measurements, Error
and Uncertainties
7) When adding or subtracting measurements, the measurements should be made to the same
number of decimal places regardless of the number of significant figures. That is, the sum or the
difference should have the same number of decimal places as the measurement with the fewer
decimal places.

8) When multiplying or dividing measurements, the measurements should be made to the same
number of significant figures. The number of significant figures in a product or quotient cannot
be larger than the quantity with the least significant figures. For example, 2.504 × 2.0 = 5.0 , not
5.008.
Caution: Avoid introducing additional error by not rounding off numbers before using them in a
calculation; round to the correct number of significant figures only when stating the result of the
calculation.

9) In any number that ends in zero and is a count of something, such as 30 rocks, the final zero
is significant. If an integer number ends in zero, the final zero may or may not be significant,
depending on whether it is a measurement or a count.

Illustration: Length Measurement

Let’s see what some of these common practices in measurement mean. Suppose you have the
measuring devices and object as shown in Figure 1.

Figure 1 Meter sticks with different precision

Using the upper meter stick, the length of the arrow would be written as 7.15 ± 0.05 cm. Using
the lower meter stick, the length of the arrow would be written as 7.0 ± 0.5 cm. Notice that these
measurements agree with each other when the estimated uncertainty is considered. The upper
meter stick is more precise than the lower one. However, we cannot say if one device is more
accurate than the other until we compare them to a standard meter stick.

Page 3 of 12
Last Update 8/20/2008
 Engineering Physics – Discussion of Measurements, Error
and Uncertainties

The object being measured in Figure 1 has a well-defined length. This is not always the case,
however. For example, the measurement uncertainty of a cotton ball could be as large as several
millimeters, even if vernier calipers with a precision of 0.1 mm were used. In this case the
measurement uncertainty is inherent in the object being measured, not in the measuring device.

Now, back to our rod, whose length might depend on temperature. Suppose you measure the
length of the rod, and you enter the measurements you and four other students make in a table
like the one shown below.

Table 1 Some Length Measurements

Measurement Length (cm)

1 5.37
2 5.39
3 5.33
4 5.31
5 5.35

The second decimal place in the measurements above is an estimate. Each student taking a read-
ing estimated the final digit. Since there is some uncertainty in this digit and each measurement
is independent, we can take an average of the measurements to obtain the best estimate of the
length of the rod. To compute the average, we simply add the measurements together and divide
by the number of measurements. The average of the five measurements is 5.35 cm. We need to
assign an uncertainty to this average that reflects the spread in the measurements and the preci-
sion of the meter stick that was used. The uncertainty in the average (or any other quantity that
you calculate using measured values) cannot itself be specified exactly. Various methods exist
for estimating the uncertainty. Some calculators generate an “uncertainty” when they average a
set of numbers. That uncertainty is called a “standard deviation,” and it is usually less than the
actual spread of the measurements. The standard deviation of the five measured lengths in the
table is 0.03 cm, and the average could be quoted as 5.35 ± 0.03 cm. Standard deviations are
good estimates of uncertainty when the number of measurements is large.

In most experiments that you will perform, measurements of mass and length may be taken
with more precision than we usually need. Hence it is often unnecessary to average several
measurements. Some measurements, however, are not as precise as we would like, and we usu-
ally take several trial measurements that are averaged to obtain the best estimate.

Tolerance, Percent Error, and Percent Difference

Page 4 of 12
Last Update 8/20/2008
 Engineering Physics – Discussion of Measurements, Error
and Uncertainties
The tolerance of a component is a measure of the acceptable range of performance the equip-
ment can exhibit during an experiment. When given as a percentage, the tolerance of a compo-
nent is the target performance multiplied by the percent accepted deviation from that target.
When performing experiments, it typical to have a theoretical prediction for the expected out-
come of the experiment; it is the value of the physical quantity that is predicted from the mathe-
matical model that describes the physical phenomenon under investigation. A measure of the ac-
curacy of the results is given by the percent error:

(1)

Suppose you measured how long it took your car to pass by two mile-markers at the side of the
road and computed your speed by using those measurements. You could compare that experi-
mental measurement with the reading on the speedometer in your car. In this case, where no
theoretical prediction is available, percent difference is used to estimate the error in experimen-
tal measurements:

E1 − E 2
percent difference = *100 (2)
(E1 + E 2 )
2
where E1 and E2 are the two experimental values that are being compared.

Remember, when taking the difference in the numerator, Guideline #7 above regarding the num-
€
ber of decimal places to keep. For example, you would not compare 4.56 to 4.6, since one has
two places after the decimal and the other has only one. You must first round 4.56 to two sig-
nificant figures: 4.6. In this example you see that there is zero percent difference or percent er-
ror between 4.56 and 4.6.

The Percent Difference and Percent Error give the experimenter some idea of how good the ex-
periment actually is. For example, the measured value g = 10.2 m/s2 ± 5% agrees with the ac-
cepted value g = 9.8 m/s 2

It is important to remember that you should only compare values with the same number of
significant figures.

Often, we use fractional or relative uncertainty to quantitatively express the precision of a meas-
urement. For example, a measurement of L = 13.5 cm ± 0.05 cm would have a fractional uncer-
tainty of 0.05/13.5 = 0.0037 = 3.7%, also called percent uncertainty.

PROPAGATION OF ERRORS
Measurement and calculation are slightly different methods of determining answers to scientific
questions. Often it is necessary to use them together, but the mathematical treatment of measured
and calculated values is significantly different. Treating each with the proper statistical signifi-

Page 5 of 12
Last Update 8/20/2008
 Engineering Physics – Discussion of Measurements, Error
and Uncertainties
cance will give results a proper context and allow reasonable interpretations of experimental
findings.

Uncertainty in Measurement
We have seen that any measurement that we make has some uncertainty associated with it.
Recall that if we measure an object, like a desktop, and find that it is 4.1055 meters long, we
should also include a statement about how accurate that measurement is. The rule adopted in
this lab course is that the uncertainty in a measurement is one half of the smallest division on the
scale. (This is true whether the scale is on a meter stick or a volt meter, etc., even if the device
has a digital readout.) If the smallest division on our measuring device is 1 mm, our uncertainty
would be 0.5 mm, or 0.0005 m, and we should say that the desk is 4.1055 ± 0.0005 m in length.

III. Uncertainty in Calculation

The limitations of measuring devices are inevitably reached in the course of experimentation.

The uncertainty in a measurement is defined as half of the smallest increment the device can
register. This is the most accurate reading we can consistently make at the limit of the device.
Calculating the uncertainty in a calculation requires more mathematics, but all uncertainty cal-
culations follow the same formula. To understand this process, we need to first define Standard
Deviation and Standard Error.

The mean of a set of N measurements is defined as:

(3)

This is just the average value of the set. However, the mean is not the only useful statistic one
can calculate from a set of data. A measurement of the precision of the data is called the vari-
ance, and is defined as:

(4)

The standard deviation is just the square root of the variance. In some situations it is necessary
to calculate a slightly different quantity, the bias-corrected variance:
(5)

The standard error is the square root of the bias-corrected variance.

Example 1:

Page 6 of 12
Last Update 8/20/2008
 Engineering Physics – Discussion of Measurements, Error
and Uncertainties
The heights of five male students in one lab section are 5 feet, 8 inches; 6 feet, 2 inches; 5 feet,
10 inches; 5 feet, 9 inches and 6 feet even. The average height is 5 feet and 10.6 inches.

Table 2 Height Measurements

Reading Height (in

inches)
Student 1 68 −2.6
Student 2 74 +3.4
Student 3 70 0.6
Student 4 69 −1.6
Student 5 72 +1.4

The sum of the squares of the last column in the table above is 23.2 in2. the variance is 4.64 in2
and the standard deviation is 2.15 in, while the bias corrected variance is 5.8 in2 and the standard
error is 2.41 in.

We can think of the standard deviation as the possible error in the next measurement of some
quantity. In the above example, if we measured the height of a sixth lab student, we can estimate
that there is roughly a 68% likelihood that the sixth student will have a height within one stan-
dard deviation of the average; in other words, between 68.2 inches and 73.0 inches. But, what-
ever the height of the sixth student, the average will be affected less by the sixth measurement
than the second measurement. So the average becomes more and more well known with each
succeeding measurement.

Propagation of Uncertainty
For a function F of multiple variables x, each with uncertainty σX, the square of the uncertainty
in F is given as

(6)

This applies to any calculation involving uncertain measurements. The uncertainty in the mean
can be calculated in this way:

(7)

(8)

and

Page 7 of 12
Last Update 8/20/2008
 Engineering Physics – Discussion of Measurements, Error
and Uncertainties

(9)

Now let us look at some examples.

Example 2: f = x1 + x2

2
 ∂f  2  ∂f 2
σ =  σ x1  + 
f σx 
 ∂x1   ∂x 2 2 
(10)
2 2
= (σ x1 ) + (σ x 2 )

Example 3 (Product rule):

€

The partial of f with respect to A is: ;

Similarly and . So

2  mf  2  nf  2  pf 2
(σ )
f =  σ A  +  σ B  +  σC 
 A  B  C 

 m σ A  2  nσ B  2  pσ C  2
= f 2  +  + 
 A   B   C 

Known as €
the product rule, this can also be written as

(12)

Note that the constant, k does not appear in the result.

Page 8 of 12
Last Update 8/20/2008
 Engineering Physics – Discussion of Measurements, Error
and Uncertainties
If there are more than 2 variables in the product, we just add more terms inside the square root
sign of the equation.. Note that a and/or b can be negative, so this works for division as well as
for multiplication.

You should be aware that these equations are just approximations, and they apply only when
the variations in the measured values of x and y are independent and random, not systematic. If
you continue your study of physics or statistics you will learn more accurate equations for de-
termining the uncertainty of your results.

Important note: Any two measurements of a quantity that agree within their uncertainties are
treated as indistinguishable. For example, 23.52 ± 0.05 and 23.49 ± 0.05 are indistinguishable
from one another, since each number lies within the uncertainty of the other.

GRAPHICAL INTERPRETATION OF MEASUREMENTS

Graphing is not just an exercise to give you something else to do in lab. It is an important
tool in physics, as it allows us to obtain certain answers from our data. Throughout your physics
lab course you will be asked to rearrange equations so that they can be graphed in such a way
that the graph yields useful information.

When two or more characteristics are measured and we wish to examine the mathematical
relationship of these physical characteristics, graphing is a convenient and useful tool. Rather
than guessing at the functional relationship between the variables, we plot the dependent variable
versus the independent variable. Then we may use various techniques to “fit” a mathematical
function to the data. One such method is called a Least Squares Analysis. When we plot a
graph, we should include our estimate of the uncertainty of the values. We do this by adding “er-
ror bars” to the points on the graph.

This is best illustrated with an example. Suppose we have some data of the variables D and
T . Let one of our points be D = 10.0 ± 0.5 at T = 5.0 , and let us plot D as the dependent vari-
able and T as the independent variable. We first plot the point at D = 10.0 , T = 5.0 . (See Fig-
ure 2.)

The actual value of D could be as high as 10.5, or as low as 9.5. It is misleading to plot the
point only as 10.0, even though that is our best estimate of the value of D . We can show the
possible range of values of D by drawing a line between 9.5 and 10.5, instead of just a point at
10.0. This would show that the value of D must be in this range. But that too is misleading,
since the value 10.0 is our best estimate of D . So we add to our line a distinct mark at 10.0. To
show clearly where the ends of the lines are, we add little “caps” to either end of the line.

Page 9 of 12
Last Update 8/20/2008
 Engineering Physics – Discussion of Measurements, Error
and Uncertainties

Figure 2 Graph showing error bars on a point

The lines extending up and down from our best estimate of 10.0 are called error bars, be-
cause they show the possible error in the value of the point. In this example there is no uncer-
tainty in the value of T . If there were, we would draw horizontal error bars of the appropriate
length similar to the ones for D .

Suppose you collect measurements of distance and time for a car moving along a highway.
You would summarize the data in a suitable table like the one below. In this example, the wrist-
watch used to measure time is marked only at the 12 hour positions, so the measurement uncer-
tainty in the time might be chosen to be ±1 min. The error in measuring the distance is less than
two car lengths, which is negligible in this situation.

Table 3 Distance versus Time Measurements

Distance (km) Time (min)

6.0 5 ±1
10.0 12 ±1
15.0 17 ±1
20.0 24 ±1

Using this data you can plot a graph of distance (dependent variable) versus time (independent
variable). Your graph might look like Figure 3 below.

Page 10 of 12
Last Update 8/20/2008
 Engineering Physics – Discussion of Measurements, Error
and Uncertainties
Some things that you will want to do every time you make a graph are
1) Give the graph a title, such a “Distance versus Time for Car”.

2) Scale your coordinate axes in such a way that you spread the graph out over the entire page.
This makes it easier to read values from the graph.

3) Label each scale, such as “Distance (km)” or “Time (min)”.

4) Draw a smooth curve through the region containing the data points. Do not merely connect
the dots.

5) Include error bars on your data points (if the uncertainties are larger than your point symbols)
to indicate the uncertainty of the data.

Figure 3 Plot of distance versus time data for a car

To determine the speed of the car from the graph, we need to find the best line that passes
through the points and compute the slope of that line. (Recall, speed = distance/time.) If you are
not familiar with mathematical techniques for determining the best line, you will need to make
your best guess. The best line serves as an average of the independent measurements. The best
line then “represents” the data, and in computing the speed of the car we simply compute the
slope of the line. From the graph, we find that the slope of the line is 0.75 ± 0.02 km/min. Thus
the velocity of the car is 0.75 ± 0.02 km/min, or 12.5 ± 0.3 m/s.

Page 11 of 12
Last Update 8/20/2008
 Engineering Physics – Discussion of Measurements, Error
and Uncertainties
A better and easier way to find the slope, intercept, and their uncertainties is to use the Linest
function in Excel. See the Excel tutorial on the lab website if you need help with the Linest func-
tion.

Page 12 of 12
Last Update 8/20/2008