Random Error Theory
Chapter 3
Probability
• The ratio of the number of times that an
event should occur to the total number of
possibilities
– If event A can occur m ways and fails to occur
n ways, then the probability of event A
occurring is m/(m + n)
• Probability of throwing a 1 on a fair die
– Should occur one time
– Will fail to occur 5 times
– Probability is 1/(1+5) = 1/6
• Probability of an event failing to occur is
1 − m/(m + n)
Probability
• Probability is always between 0 and 1
– 0 means there is no chance of occurrence
– 1 means that the event will absolutely occur
– The probability of all event occurring is 1
Compound Events
• Compound event: The simultaneous
occurrence of two or more events.
• The probability of a compound event is the
product of each probabilities of each
individual event occurring.
P = P1 ∙ P2 ∙ … ∙ Pn
Experiment
• Assume that we have 1 red ball in box 1 with 4
balls total and 2 red balls in box 2 with 5 balls
total. What is probability that 2 red balls will be
drawn when one ball is randomly selected from
each box?
Experiment
• Possible solution
– Construct a table of each possible draw.
– Compute the probability.
Experiment
Drawing Box 1 Box 2
1 White Ball 1 Red Ball 1
2 White Ball 1 Red Ball 2
3 White Ball 1 White Ball 3
4 White Ball 1 White Ball 4
5 White Ball 1 White Ball 5
6 Red Ball 2 Red Ball 1
7 Red Ball 2 Red Ball 2
8 Red Ball 2 White Ball 3
9 Red Ball 2 White Ball 4
10, and so on to 20 Red Ball 2 White Ball 5
Experiment
• There are 20 total events with only two producing
two red balls, so the probability is 2/20 = 0.1
• By formula
– Probability of drawing a red ball from Box 1 is 1/4
– Probability of drawing a red ball from Box 2 is 2/5
– Probability of drawing two red balls is
1 2 2
0.1
4 5 20
P
 
  
 
 
Application
• Least squares adjustments are the
probability of the unknowns (usually
coordinates) being a certain value based
on the simultaneous occurrence of all of
the observations.
– Distances, angles, azimuths, elevation
differences, etc.
Another Example
• Assume that a tape exist that can only
make a +1 ft or −1 ft when taping a
distance.
– Let t be the number of ways that each error
can occur
– And T be the total number of possibilities
– Then the possible occurrence of these
random errors is
Example
• For a distance of 1 tape length
– Only one +1 error or one −1 can occur
• For a distance of 2 tape lengths
– +1 and +1 occurs once with a resulting combined
error (ce) of +2
– +1 and −1 can occur twice (ce = 0)
– −1 and −1 occurs once (ce = −2)
• For a distance of 3 tape lengths
– +1, +1, +1 error occurs once (ce = +3)
– −1, +1, +1; +1, −1, +1; +1, +1, −1 (ce = +1) occurs 3 times
– −1, −1, +1; −1, +1, −1; +1, −1, −1, (ce = −1) occurs 3 times
– −1, −1, −1; (ce = −3) occurs 1
• And so on…
Number of
combining obs
Value of
resulting error
Frequency, t Total
possibilities
Probability
1 +1
−1
1
1
2 1/2
1/2
2 +2
0
−2
1
2
1
4 1/4
1/2
1/4
3 +3
+1
−1
−3
1
3
3
1
8 1/3
3/8
3/8
1/3
4 +4
+2
0
−2
−4
1
4
6
4
1
16 1/16
1/4
6/16
1/4
1/16
5 +5
+3
+1
And so on
1
5
10
32 1/32
5/32
10/32
Plots of Combining Observations
• Note the progression and fundamental shape of
the histograms
Shoot the Star-Out Game
• If you aimed at the center of the star, the
following 2D pattern of shots may occur.
Shoot the Star-Out Game
• Now
– Draw a single line across the target tabulating
the number of BBs that pierced the target
along the line
– Create a frequency plot of the hits versus
distance from the center of the target
• It may look as follows
Normal Distribution Curve
• Note that the previous
histograms were approaching
this shape.
• This is the normal distribution
curve
– So named b/c it occurs
frequently in nature
– AKA probability density curve
of a normal random variable
• The area under the curve
always equals 1.
Normal Distribution Curve
• Gauss derived the function that defines this curve
(see Appendix D) as
 
2
2
2
1
2



 
x
f x e
Properties of the Normal
Distribution Curve
• The area under the normal distribution curve
provides the probability of the simultaneous
occurrence of errors.
– Mathematically speaking
2
2
2
area under curve
1
2





 

x
t
P
e dx
Probability of an Event
• If we wish to find the probability of a single
discrete event occurring at x
– Define a infinitesimally small region under the event
and compute the area of the region
   
2
2
2
1
( )
2


 
 
x
P x y dx e dx
Properties of the Normal
Distribution Function
• From calculus
So the first and second derivative are
2
2
where
2
u
u
de du x
e u
dx dx
  

and
2
2
2
2 2
2
2 2 2

  
   
 
 
 
  
 
x
dy x x
dx
d y x dy y
dx dx
e y
Properties of the Normal
Distribution Function
• Substituting the first derivative equation into the
second derivative equation yields
2 2
2 4 2
2
2 2
1
1
d y x
y y
dx
y x
 
 
 
 
 
 
 
Properties
• From calculus, we know that
– First derivative = slope of the line at
any point
– So dy/dx = 0 when either
• x = 0 or at y = 0
• So curve
– Is asymptotic to x axis where y = 0
– Peaks at x = 0
 
2
 

dy x
y
dx
Properties
• Second derivative provides
– The rate of change of the slope at a point
– Points of inflection can be found by setting the second
derivative equal to 0.
– Occurs when y = 0 (never happens) or
2 2
2 2 2
1 0
d y y x
dx
 
  
 
 
 
2
2
2
2 2
2
1 0 or
1 or
 

     

x
x
x x
68% of area
Inflection point Inflection point

σ +σ
Properties
• By observation, the maximum value for y occurs
when x = 0.
– Setting x = 0 in original function yields
– Note: As σ goes to 0, y becomes larger (i.e. higher
probability). So smaller σ means more precise data
and higher probability that sample mean is a good
estimate for the population mean!
68% of area
Inflection point Inflection point

σ +σ
2
2
2
0
1
2
1
2
1
2
x
y e
e



 

 

 
Standard Normal Distribution Table
• Note:
– Impossible to create a table for every value of μ and σ
– Thus a table is only developed for a population having
a mean of zero (μ = 0) and variance of 1 (σ2 = 1)
• Known as the standard normal variable
– Integration of normal distribution cannot be carried out in
closed form and thus is developed numerically.
– Table of values known as Standard Normal Distribution
Table (Table D.1) since μ = 0 and σ2 = 1. (pages 562–563)
Using Table D.1
• Table D.1 gives the value of t along the x axis
where the area under the curve is tabulated
going from −∞ to t.
– t is known as the critical value
– Thus area from −∞ to 1.68 is 0.95352
But What About My Values?
• Any normal random variable y with
variance σ2 can become a standard
normal random variable, z, with variance
σ2 = 1 by
– Computing z = (y – μ)/σ
– This will be used in SUR 441 to
• Analyze the statistical significance of variables
• Check data for blunders/outliers/mistakes
Probability Computations
• For any value z, the probability of
occurrence is
P(z < t) = Nz(t)
– To determine if z is between, say, a and b, we
write
P(a < z < b) = Nz(b) − Nz(a)
– Where Nz(t) is from Table D.1
Probability of the Standard Error
• AKA – Probable Error
• Since the variance = 1, σ = ±1
• So we need to know the probability that
• Table D.1
P(−1 < z < 1) = Nz(1) − Nz(−1)
=0.84134 − 0.15866
= 0.68268
• Note that about 68.3% of data is between ±σ
Properties of Distribution
• P(−t < z < t) = P(|z| < t)
= Nz(t) − Nz(−t)
– From the symmetry of the normal distribution, P(z > t)
= P(z < −t)
– Thus the 1 − Nz(t) = Nz(−t)
– So, P(|z| < t) = Nz(t) − [1 − Nz(t)]
= 2Nz(t) − 1
-z +z
-t +t
Linear Error Probable
• 50% Probable Error
P(|z| < t) = 0.50 = 2Nz(t) − 1
• So 1.5 = 2Nz(t)
• Thus 0.75 = Nz(t)
• We need to interpolate 0.75 from Table
D.1
Linear Error Probable (50%)
• From Table D.1
– Nz(0.67) = 0.7486 and Nz(0.68) = 0.7517
– So t = 0.67 + 0.0045 = 0.6745
0.75 0.7486
0.68 0.67 0.7517 0.7486
0.75 0.7486
0.01
0.7517 0.7486
0.004516
t
t
 

 

 


Linear Interpolation
• Method of determining a value between
two tabulated values.
• Given:
• Find n as
t1 n1
tn n
t2 n2
1
1 2 1
2 1
( )
n
n n
t t t t
n n

  

Example
• Recall the example from Chapter 2 with mean = 23.5 ±1.37
20.1 20.5 21.2 21.7 21.8
21.9 22.0 22.2 22.3 22.3
22.5 22.6 22.6 22.7 22.8
22.8 22.9 22.9 23.0 23.1
23.1 23.2 23.2 23.3 23.4
23.5 23.6 23.7 23.8 23.8
23.8 23.9 24.0 24.1 24.1
24.2 24.3 24.4 24.6 24.7
24.8 25.0 25.2 25.3 25.3
25.4 25.5 25.9 25.9 26.1
• 50% of observations should be between
23.5 − 0.6745(1.37) = 22.58 < z < 24.42 = 23.5 + 0.6745(1.37)
• By inspection, 27 of 50 are between these values or 54%. So close
for a sample set. Why is it not exactly 50%?
Class practice
• What is the multiplier for
– E90?
– E95?
– E99?
– E99.7?
– E99.97?
95% Probable Error
• No surveying business is willing to reject 50% of
their observations as blunders.
• 95% is typically used in all surveying standards
such as ALTA-ACSM
• Determine t for 95% probable error
0.95 = P(|z| < t) = 2Nz(t) − 1
So 1.95 = 2Nz(t)
Nz(t) = 0.975
95% Probable Error
• From Table D.1
• Nz(1.96) = 0.9750
– no interpolation required!
Example
• Again using example from Chapter 2 with mean = 23.5 ±1.37
20.1 20.5 21.2 21.7 21.8
21.9 22.0 22.2 22.3 22.3
22.5 22.6 22.6 22.7 22.8
22.8 22.9 22.9 23.0 23.1
23.1 23.2 23.2 23.3 23.4
23.5 23.6 23.7 23.8 23.8
23.8 23.9 24.0 24.1 24.1
24.2 24.3 24.4 24.6 24.7
24.8 25.0 25.2 25.3 25.3
25.4 25.5 25.9 25.9 26.1
• 95% of observations should be between
23.5 − 1.96(1.37) = 20.81 < z < 26.18 = 23.5 + 1.96(1.37)
• By inspection, 48 of 50 are between these values or 96%. So close
for a sample set. Is there reason to be concerned about data set?
E99.7
• 0.997 = P(|z| < t) = 2Nz(t) − 1
• 1.997 = 2Nz(t)
• Nz(t) = 0.9985
• From Table D.1
– Nz(2.96) = 0.99846 and Nz(2.97) = 0.99851
– Interpolate
0.9985 0.99846
2.96 0.01
0.99851 0.99846
2.968
t

 
   

 

Other Probable Errors
Symbol Multiplier % Error Name
E50 0.6745σ 50% Linear Error Probable
E90 1.645σ 90% 90% confidence level
E95 1.960σ 95% 95% confidence level
E95.4 2σ 95.4% 2-sigma error
E99.73 3σ 99.73% 3-sigma error
E99.9 3.29σ 99.9% “Ivory soap clean” ☺

Random Error Theory

  • 1.
  • 2.
    Probability • The ratioof the number of times that an event should occur to the total number of possibilities – If event A can occur m ways and fails to occur n ways, then the probability of event A occurring is m/(m + n) • Probability of throwing a 1 on a fair die – Should occur one time – Will fail to occur 5 times – Probability is 1/(1+5) = 1/6 • Probability of an event failing to occur is 1 − m/(m + n)
  • 3.
    Probability • Probability isalways between 0 and 1 – 0 means there is no chance of occurrence – 1 means that the event will absolutely occur – The probability of all event occurring is 1
  • 4.
    Compound Events • Compoundevent: The simultaneous occurrence of two or more events. • The probability of a compound event is the product of each probabilities of each individual event occurring. P = P1 ∙ P2 ∙ … ∙ Pn
  • 5.
    Experiment • Assume thatwe have 1 red ball in box 1 with 4 balls total and 2 red balls in box 2 with 5 balls total. What is probability that 2 red balls will be drawn when one ball is randomly selected from each box?
  • 6.
    Experiment • Possible solution –Construct a table of each possible draw. – Compute the probability.
  • 7.
    Experiment Drawing Box 1Box 2 1 White Ball 1 Red Ball 1 2 White Ball 1 Red Ball 2 3 White Ball 1 White Ball 3 4 White Ball 1 White Ball 4 5 White Ball 1 White Ball 5 6 Red Ball 2 Red Ball 1 7 Red Ball 2 Red Ball 2 8 Red Ball 2 White Ball 3 9 Red Ball 2 White Ball 4 10, and so on to 20 Red Ball 2 White Ball 5
  • 8.
    Experiment • There are20 total events with only two producing two red balls, so the probability is 2/20 = 0.1 • By formula – Probability of drawing a red ball from Box 1 is 1/4 – Probability of drawing a red ball from Box 2 is 2/5 – Probability of drawing two red balls is 1 2 2 0.1 4 5 20 P         
  • 9.
    Application • Least squaresadjustments are the probability of the unknowns (usually coordinates) being a certain value based on the simultaneous occurrence of all of the observations. – Distances, angles, azimuths, elevation differences, etc.
  • 10.
    Another Example • Assumethat a tape exist that can only make a +1 ft or −1 ft when taping a distance. – Let t be the number of ways that each error can occur – And T be the total number of possibilities – Then the possible occurrence of these random errors is
  • 11.
    Example • For adistance of 1 tape length – Only one +1 error or one −1 can occur • For a distance of 2 tape lengths – +1 and +1 occurs once with a resulting combined error (ce) of +2 – +1 and −1 can occur twice (ce = 0) – −1 and −1 occurs once (ce = −2) • For a distance of 3 tape lengths – +1, +1, +1 error occurs once (ce = +3) – −1, +1, +1; +1, −1, +1; +1, +1, −1 (ce = +1) occurs 3 times – −1, −1, +1; −1, +1, −1; +1, −1, −1, (ce = −1) occurs 3 times – −1, −1, −1; (ce = −3) occurs 1 • And so on…
  • 12.
    Number of combining obs Valueof resulting error Frequency, t Total possibilities Probability 1 +1 −1 1 1 2 1/2 1/2 2 +2 0 −2 1 2 1 4 1/4 1/2 1/4 3 +3 +1 −1 −3 1 3 3 1 8 1/3 3/8 3/8 1/3 4 +4 +2 0 −2 −4 1 4 6 4 1 16 1/16 1/4 6/16 1/4 1/16 5 +5 +3 +1 And so on 1 5 10 32 1/32 5/32 10/32
  • 13.
    Plots of CombiningObservations • Note the progression and fundamental shape of the histograms
  • 14.
    Shoot the Star-OutGame • If you aimed at the center of the star, the following 2D pattern of shots may occur.
  • 15.
    Shoot the Star-OutGame • Now – Draw a single line across the target tabulating the number of BBs that pierced the target along the line – Create a frequency plot of the hits versus distance from the center of the target • It may look as follows
  • 16.
    Normal Distribution Curve •Note that the previous histograms were approaching this shape. • This is the normal distribution curve – So named b/c it occurs frequently in nature – AKA probability density curve of a normal random variable • The area under the curve always equals 1.
  • 17.
    Normal Distribution Curve •Gauss derived the function that defines this curve (see Appendix D) as   2 2 2 1 2      x f x e
  • 18.
    Properties of theNormal Distribution Curve • The area under the normal distribution curve provides the probability of the simultaneous occurrence of errors. – Mathematically speaking 2 2 2 area under curve 1 2         x t P e dx
  • 19.
    Probability of anEvent • If we wish to find the probability of a single discrete event occurring at x – Define a infinitesimally small region under the event and compute the area of the region     2 2 2 1 ( ) 2       x P x y dx e dx
  • 20.
    Properties of theNormal Distribution Function • From calculus So the first and second derivative are 2 2 where 2 u u de du x e u dx dx     and 2 2 2 2 2 2 2 2 2                    x dy x x dx d y x dy y dx dx e y
  • 21.
    Properties of theNormal Distribution Function • Substituting the first derivative equation into the second derivative equation yields 2 2 2 4 2 2 2 2 1 1 d y x y y dx y x              
  • 22.
    Properties • From calculus,we know that – First derivative = slope of the line at any point – So dy/dx = 0 when either • x = 0 or at y = 0 • So curve – Is asymptotic to x axis where y = 0 – Peaks at x = 0   2    dy x y dx
  • 23.
    Properties • Second derivativeprovides – The rate of change of the slope at a point – Points of inflection can be found by setting the second derivative equal to 0. – Occurs when y = 0 (never happens) or 2 2 2 2 2 1 0 d y y x dx            2 2 2 2 2 2 1 0 or 1 or           x x x x 68% of area Inflection point Inflection point  σ +σ
  • 24.
    Properties • By observation,the maximum value for y occurs when x = 0. – Setting x = 0 in original function yields – Note: As σ goes to 0, y becomes larger (i.e. higher probability). So smaller σ means more precise data and higher probability that sample mean is a good estimate for the population mean! 68% of area Inflection point Inflection point  σ +σ 2 2 2 0 1 2 1 2 1 2 x y e e           
  • 25.
    Standard Normal DistributionTable • Note: – Impossible to create a table for every value of μ and σ – Thus a table is only developed for a population having a mean of zero (μ = 0) and variance of 1 (σ2 = 1) • Known as the standard normal variable – Integration of normal distribution cannot be carried out in closed form and thus is developed numerically. – Table of values known as Standard Normal Distribution Table (Table D.1) since μ = 0 and σ2 = 1. (pages 562–563)
  • 26.
    Using Table D.1 •Table D.1 gives the value of t along the x axis where the area under the curve is tabulated going from −∞ to t. – t is known as the critical value – Thus area from −∞ to 1.68 is 0.95352
  • 27.
    But What AboutMy Values? • Any normal random variable y with variance σ2 can become a standard normal random variable, z, with variance σ2 = 1 by – Computing z = (y – μ)/σ – This will be used in SUR 441 to • Analyze the statistical significance of variables • Check data for blunders/outliers/mistakes
  • 28.
    Probability Computations • Forany value z, the probability of occurrence is P(z < t) = Nz(t) – To determine if z is between, say, a and b, we write P(a < z < b) = Nz(b) − Nz(a) – Where Nz(t) is from Table D.1
  • 29.
    Probability of theStandard Error • AKA – Probable Error • Since the variance = 1, σ = ±1 • So we need to know the probability that • Table D.1 P(−1 < z < 1) = Nz(1) − Nz(−1) =0.84134 − 0.15866 = 0.68268 • Note that about 68.3% of data is between ±σ
  • 30.
    Properties of Distribution •P(−t < z < t) = P(|z| < t) = Nz(t) − Nz(−t) – From the symmetry of the normal distribution, P(z > t) = P(z < −t) – Thus the 1 − Nz(t) = Nz(−t) – So, P(|z| < t) = Nz(t) − [1 − Nz(t)] = 2Nz(t) − 1 -z +z -t +t
  • 31.
    Linear Error Probable •50% Probable Error P(|z| < t) = 0.50 = 2Nz(t) − 1 • So 1.5 = 2Nz(t) • Thus 0.75 = Nz(t) • We need to interpolate 0.75 from Table D.1
  • 32.
    Linear Error Probable(50%) • From Table D.1 – Nz(0.67) = 0.7486 and Nz(0.68) = 0.7517 – So t = 0.67 + 0.0045 = 0.6745 0.75 0.7486 0.68 0.67 0.7517 0.7486 0.75 0.7486 0.01 0.7517 0.7486 0.004516 t t          
  • 33.
    Linear Interpolation • Methodof determining a value between two tabulated values. • Given: • Find n as t1 n1 tn n t2 n2 1 1 2 1 2 1 ( ) n n n t t t t n n     
  • 34.
    Example • Recall theexample from Chapter 2 with mean = 23.5 ±1.37 20.1 20.5 21.2 21.7 21.8 21.9 22.0 22.2 22.3 22.3 22.5 22.6 22.6 22.7 22.8 22.8 22.9 22.9 23.0 23.1 23.1 23.2 23.2 23.3 23.4 23.5 23.6 23.7 23.8 23.8 23.8 23.9 24.0 24.1 24.1 24.2 24.3 24.4 24.6 24.7 24.8 25.0 25.2 25.3 25.3 25.4 25.5 25.9 25.9 26.1 • 50% of observations should be between 23.5 − 0.6745(1.37) = 22.58 < z < 24.42 = 23.5 + 0.6745(1.37) • By inspection, 27 of 50 are between these values or 54%. So close for a sample set. Why is it not exactly 50%?
  • 35.
    Class practice • Whatis the multiplier for – E90? – E95? – E99? – E99.7? – E99.97?
  • 36.
    95% Probable Error •No surveying business is willing to reject 50% of their observations as blunders. • 95% is typically used in all surveying standards such as ALTA-ACSM • Determine t for 95% probable error 0.95 = P(|z| < t) = 2Nz(t) − 1 So 1.95 = 2Nz(t) Nz(t) = 0.975
  • 37.
    95% Probable Error •From Table D.1 • Nz(1.96) = 0.9750 – no interpolation required!
  • 38.
    Example • Again usingexample from Chapter 2 with mean = 23.5 ±1.37 20.1 20.5 21.2 21.7 21.8 21.9 22.0 22.2 22.3 22.3 22.5 22.6 22.6 22.7 22.8 22.8 22.9 22.9 23.0 23.1 23.1 23.2 23.2 23.3 23.4 23.5 23.6 23.7 23.8 23.8 23.8 23.9 24.0 24.1 24.1 24.2 24.3 24.4 24.6 24.7 24.8 25.0 25.2 25.3 25.3 25.4 25.5 25.9 25.9 26.1 • 95% of observations should be between 23.5 − 1.96(1.37) = 20.81 < z < 26.18 = 23.5 + 1.96(1.37) • By inspection, 48 of 50 are between these values or 96%. So close for a sample set. Is there reason to be concerned about data set?
  • 39.
    E99.7 • 0.997 =P(|z| < t) = 2Nz(t) − 1 • 1.997 = 2Nz(t) • Nz(t) = 0.9985 • From Table D.1 – Nz(2.96) = 0.99846 and Nz(2.97) = 0.99851 – Interpolate 0.9985 0.99846 2.96 0.01 0.99851 0.99846 2.968 t           
  • 40.
    Other Probable Errors SymbolMultiplier % Error Name E50 0.6745σ 50% Linear Error Probable E90 1.645σ 90% 90% confidence level E95 1.960σ 95% 95% confidence level E95.4 2σ 95.4% 2-sigma error E99.73 3σ 99.73% 3-sigma error E99.9 3.29σ 99.9% “Ivory soap clean” ☺

Editor's Notes