CURVE FITTING

Unit – I: Curve Fitting:Curve fitting by the method of least squares and fitting of the curves of
the form, y = ax + b, y = ax2 + bx + c, y = aebx and y = axb
Statistical Methods: Measures of central tendency and dispersion. Correlation-Karl Pearson’s
coefficient of correlation-problems. Regression analysis- lines of regression, problems. Rank
correlation.

Least Square method

The method of finding a specific relation y = f(x) for the data to satisfy as accurately as
possible and such an equation is called the best fitting equation or the curve of best fit.
Fitting of a straight line y = a + bx
Consider a set of n given values (x, y) for fitting the straight line y = a + bx where a and b are
parameters to be determined. We find the parameters a and b using the normal equations

na  b x   y

a  x  b x 2   xy

Fitting of a second degree parabolay = a + bx+ cx2

Consider a set of n given values (x, y) for fitting the curve y = a + bx + cx2 where a, b and c are
parameters to be determined. We find the parameters a, b and c using the normal equations

na  b x  c  x 2   y

a  x  b x 2  c x3   xy

a  x 2  b x 3  c  x 4   x 2 y

Note: The normal equations for fitting a straight line or parabola can be written instantly from
the desired equation of the curve as follows
We first apply summation (  ) to the desired equation keeping the constants a, b and c outside
the summation where the summation of pure constant terms like  a ,  b ,  c are to be written as
na, nb, nc respectively
We then multiply the given equation by the independent variable x and apply summation again.
This will be sufficient for fitting a straight line. However in the case of parabola we must also
multiply by x2 and apply summation.

1
Fitting of a curve of the form y  ab x

Consider y  ab x (1)

Taking log on both sides, we get

loge y  loge a  x loge b

or Y  A  BX (2)

where Y  loge y , A  loge a , B  log e b and X = x.

Which is the same as y  a  bx , the normal equations associated with equation (2) are as follows

nA  B  X   Y (3)

A X  B  X 2   XY (4)

Solving (3) and (4) we obtain A and B.

But we have log e a  A  a  e A

and log e b  B  b  e B

Substitution of the values of a and b in (1) give us the best fitting curve y  ab x in the least
square sense.

Note:We can also fit curves of the form y  aebx (Exponential curve), y  axb (Geometric curve)
in the similar way.
Working procedure for problems:
Method I: (Direct Method)
Step 1: We first write the normal equations appropriate to the curve of fit.
Step 2:We prepare the relevant table and the find the values of the summation present in the
normal equations. We substitute these values to arrive a system of equations in the unknown
parameters.
Step 3: We find the parameters by solving and substitute in the given equation.

2
Example: Fit a straight line y  a  bx in the least square sense for the data

x 1 3 4 6 8 9 11 14
y 1 2 4 4 5 7 8 9

Solution: The normal equations for y  a  bx are given by

na  b x   y

a  x  b x 2   xy Here n = 8

The relevant table is as follows

x y xy x2
1 1 1 1
3 2 6 9
4 4 16 16
6 4 24 36
8 5 40 64
9 7 63 81
11 8 88 121
14 9 126 196
 x = 56  y = 40  xy = 364 2
 x = 524

The normal equations becomes

8a  56b  40 (1)
56a  524b  364 (2)
Solving (1) and (2) we get a = 0.52, b = 0.64
The equation y  a  bx becomes y  0.52  0.64 x .

Example: Find a law of the form y  a  bx for the following data

x 100 120 140 160 180 200

y 45 55 60 70 80 85

Solution: The normal equations for y  a  bx are given by

na  b x   y

a  x  b x 2   xy Here n = 6

The relevant table is as follows

3
 x y xy x2
100 45 4500 10000
120 55 6600 14400
140 60 8400 19600
160 70 11200 25600
180 80 14400 32400
200 85 17000 40000
 x = 900  y = 395  xy = 62100 2
 x = 142000

The normal equations becomes

6a  900b  395 (1)
900a  142000b  62100 (2)
Solving (1) and (2) we get a = 4.7619, b = 0.4071
The equation y  a  bx becomes y  4.7619  0.4071x

Example: Fit a straight line for the data given below using the method of least squares
x 1 2 3 4 6 8
y 2.4 3 3.6 4 5 6

Solution: The normal equations for y  a  bx are given by

na  b x   y

a  x  b x 2   xy Here n = 6

The relevant table is as follows

x y xy x2
1 2.4 2.4 1
2 3 6 4
3 3.6 10.8 9
4 4 16 16
6 5 30 36
8 6 48 64
 x = 24  y = 24  xy = 113.2 2
 x = 130

The normal equations becomes

6a  24b  24 (1)
24a  130b  113.2 (2)

4
Solving (1) and (2) we get a = 1.9764, b = 0.5058
The equation y  a  bx becomes y  1.9764  0.5058 x .

Example: Find a law of the form y  a  bx for the following data

Year (x) 1911 1921 1931 1941 1951

Production (y) 8 10 12 10 6

Solution: The normal equations for y  a  bx are given by

 y  na  b x

 xy  a x  b x
2
Here n = 5

The relevant table is as follows

x y xy x2
1911 8 15288 3651921
1921 10 19210 3690241
1931 12 23172 3728761
1941 10 19410 3767481
1951 6 11706 3806401
 x = 9655  y = 46  xy = 88786 2
 x = 18644805

The normal equations becomes

46  5a  9655b (1)
88786  9655a  18644805b (2)
Solving (1) and (2) we get a = 86.44, b = 0.04
The equation y  a  bx becomes y   x

Alternative Method
The normal equations for y  a  bX are given by

 y  na  b X

 Xy  a  X  b X
2
Here n = 5 and X = x - 1931

5
The relevant table is as follows
x X = x - 1931 y Xy X2
1911 -20 8 -160 400
1921 -10 10 -100 100
1931 0 12 0 0
1941 10 10 100 100
1951 20 6 120 400
X=0  y = 46  Xy = -40 2
 X = 1000

The normal equations becomes

46  5a  0b
46
a  9.2
5
40  0a  1000b
40
b  0.04
1000
The equation y  a  bX becomes y  9.2  0.04 X

Put X = x - 1931
y  9.2  0.04  x  1931

y  86.44  0.04 x

Examples:
1.Find the equation of the beat fitting straight line for the following data
x 1 2 3 4 5
y 14 13 9 5 2
2. Fit a straight line for the data given below using the method of least squares
x 0 1 2 3 4 5
y 9 8 24 28 26 20
3. Fit a straight line for the data given below using the method of least squares
x 62 64 65 69 70 71 72
y 65.7 66.8 67.2 69.3 69.8 70.5 70.9
4. Find a law of the form y  a  bx for the following data

x 50 70 100 120
y 12 15 21 25

6
5. A simply supported beam carries a concentrated load P at its midpoint corresponding to
various values of Pthe maximum deflection Y is measured and is given below
P 100 120 140 160 180 200
Y 0.45 0.55 0.60 0.70 0.80 0.85
Find a law of the form Y  a  bP andhence estimate Y when P = 150.

Example: Fit a parabola of second degree y  a  bx  cx 2 for the data

x 0 1 2 3 4
y 1 1.8 1.3 2.5 2.3

Solution:The normal equations for y  a  bx  cx 2 are given by

na  b x  c  x 2   y

a  x  b x 2  c x3   xy

a  x 2  b x 3  c  x 4   x 2 y Here n = 5

The relevant table is as follows

x y xy x2 x2y x3 x4
0 1 0 0 0 0 0
1 1.8 1.8 1 1.8 1 1
2 1.3 2.6 4 5.2 8 16
3 2.5 7.5 9 22.5 27 81
4 2.3 9.2 16 36.8 64 256
 x = 10  y = 8.9  xy = 21.1 2
 x = 30
2
 x y = 66.3
3
 x = 100
4
 x = 354

The normal equations becomes

5a  10b  30c  8.9 (1)
10a  30b  100c  21.1 (2)
30a  100b  354c  66.3 (3)
Solving (1), (2) and (3) we get a = 1.078, b = 0.414 and c = - 0.021

The equation y  a  bx  cx 2 becomes y  1.078  0.414 x  0.021x2 .

7
Example: Fit a parabola y  a  bx  cx 2 by the method of least square for the data

x 2 4 6 8 10
y 3.07 12.85 31.47 57.38 91.29

Solution: The normal equations for y  a  bx  cx 2 are given by

na  b x  c  x 2   y

a  x  b x 2  c x3   xy

a  x 2  b x 3  c  x 4   x 2 y Here n = 5

Examples:

1. Fit a parabola y  a  bx  cx 2 by the method of least square to the following data

x 1.0 1.5 2.0 2.5 3.0 3.5 4.0

y 1.1 1.3 1.6 2.0 2.7 3.4 4.1

2. Fit a parabola of second degree y  a  bx  cx 2 in the least square sense for the data

x 0 1 2 3 4 5 6
y 14 18 27 29 36 40 46

3. Fit a second degree parabola y  a  bx  cx 2 in the least square sense for the following data
and hence estimate y at x = 6
x 1 2 3 4 5
y 10 12 13 16 19

4. Fit a parabola of second degree y  a  bx  cx 2 in the least square sense for the data

x 1 2 3 4 5
y 25 28 33 39 46

5. Fit a parabola of second degree y  a  bx  cx 2 in the least square sense for the data

x 10 20 30 40 50 60
y 157 179 210 252 302 361

8
6. Fit a parabola of second degree y  a  bx  cx 2 in the least square sense for the data

x 0 1 2 3 4
y 1 5 10 22 38
7. Fit a curve of the form y  a0  a1 x  a2 x to the data
2

x 0 1 2 3 4
y 1 1.8 1.3 2.5 6.3
by the method of least squares

8. Fit a parabola y  a  bx  cx 2 to the data

x 1 2 3 4
y 1.7 1.8 2.3 3.2
by the method of least squares

9. Fit a parabola of the form y  a  bx  cx 2 for the following data

x -2 -1 0 1 2
y -3.150 -1.390 0.620 2.886 5.378

10. Fit a parabola y  a  bx  cx 2 by the method of least square to the following data

x -3 -2 -1 0 1 2 3
y 4.63 2.11 0.67 0.09 0.63 2.15 4.58

11. Fit a second degree polynomial of the form y  a  bx  cx 2 for the data

x 0 1 2 3 4 5
y 1 3 7 13 21 31

9
Example: Fit a curve of the form y  ab x in the least square sense for the following data

x 0 2 4 5 7 10
y 100 120 256 390 710 1600

Solution: Consider y  ab x (1)

Take log on both sides

loge y  loge a  x loge b

Let us write this in the form

Y  A  BX

where Y  loge y , A  loge a , B  log e b , X  x

The associated normal equations are

nA  B  X   Y

A X  B  X 2   XY Here n = 6

The relevant table is as follows

X=X y Y= loge y XY X2
0 100
120
4 256
5 390
7 710
10 1600
 X = 28 Y =  XY = 2
 X = 194
34.8471 181.3214
The normal equations becomes
6 A  28B  34.8471 (1)
28 A  194 B  181.3214 (2)
Solving (1) and (2) we get A = 4.4298, B = 0.2953

But A  log e a  a  e A  a  e 4.4298  83.9146

B  log e b  b  e B  b  e 0.2953  1.3435

Thus the required curve is y  (83.9146)(1.3435) x

10
Example: Fit a curve of the form y  ab x in the least square sense for the following data

x 1 2 3 4 5 6 7 8
y 1.0 1.2 1.8 2.5 3.6 4.7 6.6 9.1

Example: Fit a curve of the form y  ab x in the least square sense for the following data and
hence estimate y when x = 8.
x 0 1 2 3 4 5 6
y 32 47 65 92 132 190 275

Example: Fit a curve of the form y  ab x in the least square sense for the following data

x 1 2 3 4 5 6 7
y 87 97 113 129 202 195 193

Example: At constant temperature, the pressure P and the volume V of a gas are connected by
the relation PV  = constant. Find the best fitting equation of this form to the following data and
estimate V when P = 4
P(Kg/Sq. cm 0.5 1.0 1.5 2.0 2.5 3.0
V(c.c) 1620 1000 750 620 520 460
Example: Fit a curve of the form y  ae for the data
bx

x 0 2 4
y 8.12 10 31.82
Example: Fit a curve of the form y  aebx for the data

x 5 6 7 8 9 10
y 133 55 23 7 2 2
Example: Fit a curve of the form y  axb for the data

x 1 2 3 4 5 6
y 2.98 4.26 5.21 6.1 6.8 7.5

11
Correlation:
Suppose two variables x and y are related in such a way that an increase in one is accompanied
by an increase or decrease in the other. Such a relationship is called correlation (or covariation).
If x and y increase or decrease together, then we say that x and y are positively (directly)
correlated. On the other hand, if y decreases as x increases or vice-versa then we say that x and y
are negatively (inversely) correlated.
For example demand and price of a commodity are positively correlated, whereas supply and
price are negatively correlated.
The numerical measure of correlation between two variables x and y is known as the co-efficient
of correlation and it is defined as

 ( x  x )( y  y )
r
n x y

x y
where n is the number of observations, x  is mean of x, y  is mean of y,
n n
 (x  x )
2 2
x
x    ( x ) 2 is the standard deviation of x and
n n
 ( y  y)
2 2
y
y    ( y ) 2 is the standard deviation of y.
n n
Alternate form (1):
If X  x  x and Y  y  y

 (x  x )
2 2
X
x  
n n

 ( y  y)
2 2
Y
y  
n n
2 2
X Y
or  x y   n x y   X 2  Y 2
n n

 XY
Therefore r 
2 2
X Y

Alternate form (2):

 x2   y2   x2 y
r
2 x y

25
Property:
The co-efficient of correlation numerically does not exceed unity.
Proof:
We have to show that 1  r  1
2 2
1 X Y  1 X Y 
Let S     and S     
2n   x  y  2n   x  y 

where X  x  x and Y  y  y

Obviously both S and S  are ≥ 0.

1  X 2 Y 2 2 XY 
Now S       0
2 n   x2  y2  x y 

1  X2 Y2 XY 
S   2   2  2   0
2n   x y  x y 

1 1 X2 1 Y 2  XY 
S  2  2 2   0
2x n y n n x y 

1 1 2 1 2 
S  2  x  2  y  2 r   0
2 x y 
1
S 1  1  2r   0
2
1
S  2  2r   0
2
 1 r  0
 1  r (1)
1
Similarly we can obtain S    2  2r   0
2
 1 r  0
 r 1 (2)
From (1) and (2) 1  r  1

26
Regression
Regression is an estimation of one independent variable in terms of the other. If x and y are
correlated, the best fitting straight line in the least square sense gives reasonably a good relation
between x and y.
The best fitting straight line of the form y = ax + b (x being the independent variable) is called
the regression line of y on x and x = ay +b (y being the independent variable) is called the
regression line of x on y.
The regression line of y on x
y
y y r (x  x )
x

(or) r  bxy  byx

 y  XY
where byx  r  ,
x  X 2

X  x  x and Y  y  y

The regression line of x on y

x
xx r (y  y)
y

(or) x  x  bxy ( y  y )

 x  XY
where bxy  r  ,
 y Y 2

Note:
y 
The values byx  r and bxy  r x are known as the regression co-efficients. Their product is
x y
equal to r2

i.e., r  bxy  byx

27
Example:

 x y  1  r 2 
Show that θ is the angle between the lines of regression then tan   2  
 x   y2  r 

Solution:

We know that if θ is acute the angle between the lines y  m1 x  c1 and y  m2 x  c2 is given by
m2  m1
tan  
1  m1m2

We have the lines of regression

y
y y r (x  x ) (1)
x

x
and x  x  r ( y  y ) we write this equation as
y

y
y y  (x  x ) (2)
r x

Slope of equations (1) and (2) are respectively given by

y y
m1  r and m2 
x r x

Substituting these in the formula for tanθ, we have

y  y 1 
r y r
r x x  x  r 
tan   
y y 2
1 r 1  y2
 x r x x

 y  1 r2   1 r2 
 x y 
 x  r   r 

 
 x2   y2  x2   y2
 x2

 x y  1  r 2 
tan    
 x2   y2  r 

28
Example:
Calculate the co-efficient of correlation and obtain the lines of regression for the following data
x 1 2 3 4 5 6 7 8 9
y 9 8 10 12 11 13 14 16 15
Obtain an estimate for y which corresponds to x = 6.2.
Solution:
Here n = 9

 x 45  y 108
x   5 and y    12
n 9 n 9
We prepare the following table
x X  xx X2 y Y  y y Y2 XY
1 -4 16 9 -3 9 12
2 -3 9 8 -4 16 12
3 -2 4 10 -2 4 4
4 -1 1 12 0 0 0
5 0 0 11 -1 1 0
6 1 1 13 1 1 1
7 2 4 14 2 4 4
8 3 9 16 4 16 12
9 4 16 15 3 9 12
∑X2 = 60 2
∑ Y = 60 ∑XY = 57

 XY
Now r 
2 2
 X Y

57 57
r   0.95
60  60 60
2 2
X 60 Y 60
 x2    6.6667  y2    6.6667
n 9 n 9

 x  2.582  y  2.582

Therefore the regression co-efficients are

y 2.582  2.582
byx  r  0.95   0.95 bxy  r x  0.95   0.95
x 2.582 y 2.582

29
Therefore the line of regression of y on x is
y  y  byx ( x  x )

y  12  0.95( x  5)

y  0.95 x  7.25 (1)

The line of regression of y on x is

x  x  bxy ( y  y )

x  5  0.95( y  12)

x  0.95 y  6.4 (2)

When x = 6.2 we find from equation (1) that y = 13.14.

Example: Obtain the lines of regression and hence find the co-efficient of correlation for the
following data
x 1 3 4 2 5 8 9 10 13 15
y 8 6 10 8 12 16 16 10 32 32

Solution:
Here n = 10

 x 70  y 150
x   7 and y    15
n 10 n 10

We prepare the following table

x X  xx X2 y Y  y y Y2 XY
1 -6 36 8 -7 49 42
3 -4 16 6 -9 81 36
4 -3 9 10 -5 25 15

30
 2 -5 25 8 -7 49 35
5 -2 4 12 -3 9 6
8 1 1 16 1 1 1
9 2 4 16 1 1 2
10 3 9 10 -5 25 -15
13 6 36 32 17 289 102
15 8 64 32 17 289 136
∑X2 = ∑Y 2 = ∑XY =
204 818 360

The regression co-efficients are

 XY 360  XY 360
byx  2
  1.7647 bxy  2
  0.44
X 204 Y 818

Therefore the line of regression of y on x is

y  y  byx ( x  x )

y  15  1.7647( x  7)

y  1.7647 x  2.6471 (1)

The line of regression of x on y is

x  x  bxy ( y  y )

x  7  0.44( y  15)

x  0.44 y  0.4 (2)

Co-efficient of correlation is

r  bxy  byx  0.44  1.7647  0.8812

31
Example: A person while calculating the co-efficient of correlation between two variables x, y
from a set of 25 observations obtain the following results ∑x = 125, ∑y = 100, ∑xy = 508,
x :8 6
∑x2 = 650, ∑y2 = 460. But it was later found that the pair of values where wrongly
y :12 8
x :8 6
copied as .
y :14 6

Obtain the correct value for the correlation co-efficient.

Solution: Computation for wrong values is as follows
x y xy x2 y2
6 14 84 36 196
8 6 48 64 36
∑x = 14 ∑y = 20 ∑xy = 132 ∑ x2 = 100 ∑ y2 = 132

Computation for correct values is as follows

x y xy x2 y2
8 12 96 64 144
6 8 48 36 64
∑x = 14 ∑y = 20 ∑xy = 144 ∑ x2 = 100 ∑ y2 = 208

It may be observed that the summations ∑x, ∑y, ∑x 2 are unchanged even after the correction.
However we have
correct∑xy = 508 – 132 + 144 = 520
correct ∑y2 = 460 – 232 + 208 = 436
Therefore correct values of the mean and standard deviation of x and y are as follows

 x 125  y 100
x  5 y  4
n 25 n 25
2
x 650 2
x   ( x )2  5  1 1
n 25
2
y 436 2
y   ( y )2   4  1.44  1.2
n 25

 ( x  x )( y  y )
We have r 
n x y

32
  xy   xy   xy   x y

n x y

1   xy y  x x  y nx y 
     
 x y  n n n n 

1   xy 
   x y  x y  x y
 x y  n 

1   xy 
   x y
 x y  n 

1  520 
   (5  4)   0.6666  0.67
(1)(1.2)  25 
Hence the correct value of r = 0.67.
Examples:
1. Find the correlation co-efficient and the regression lines for the following data
x 1 2 3 4 5
y 2 5 3 8 7
Find the best estimate for y when x = 3.5 and the best estimate for x when y = 3.5.
2. Calculate the co-efficient of correlation and obtain the lines of regression for the following
data
x 3 5 6 9 10 12 15 20 22 28
y 10 12 15 18 20 22 27 30 32 34

3.Find the co-efficient of correlation by obtaining the lines of regression

x 80 45 55 56 58 60 65 68 70 75 85
y 52 56 50 48 60 62 64 65 70 74 90

4. Find the co-efficient of correlation by obtaining the lines of regression

x 17 18 19 19 20 20 21 22 21 23
y 12 16 14 11 15 19 22 15 16 20
5. Find the correlation coefficient and the regression lines y on x and x on y for the following
data
x 1 2 3 4 5
y 2 5 3 8 7

33
6.Find the correlation coefficient and the regression lines y on x and x on y for the following data
x 2 4 6 8 10
y 5 7 9 8 11

7. Find the correlation coefficient between x and y from the given data
x 78 89 97 69 59 79 68 57
y 125 137 156 112 107 138 123 108

8. Find the co-efficient of correlation by obtaining the lines of regression

x 1 2 3 4 5 6 7
y 9 8 10 12 11 13 14
9. The two lines of regression for the variables x and y are given by x = 19.3 – 0.87y and y =
11.64 – 0.90x. Find
(i) the mean values of x and y
(ii) co-efficient of correlation between x and y.
10. In a partially destroyed lab record of analysis of correlation data, the following results are
available, variance of x is 9. Regression equations are 8x – 10y + 66 = 0 and 40x – 18y – 214 =
0. Find x , y ,  y and correlation co-efficient.

11. The regression equations are 8x – 10y + 66 = 0 and 20x – 9y – 107 = 0. Find x , y ,  y and
correlation co-efficient.

34
Rank Correlation
A group of n individuals may be arranged in order to merit with respect to some characteristics.
The same group would give different orders for different characteristics. Considering the orders
corresponding to two characteristics A and B, the correlation between these n pairs of ranks is
called rank correlation in the characteristics A and B for that group of individuals.
Let xi, yi be the ranks of the ith individuals in A and B respectively. Assuming that no two
individuals are bracketed equal in either case, each of the variables taking the values 1, 2, 3, 4,
…..n, we have
Rank correlation between A and B is given by

6 d i2
  1
n3  n

where di  xi  yi difference between the ranks of the ith individuals in A and B respectively.

Example: Ten participants in a contest are ranked by two judges as follows

x 1 6 5 10 3 2 4 9 7 8
y 6 4 9 8 1 2 3 10 5 7
Calculate the rank correlation coefficient ρ.

Solution: If di  xi  yi , then di  5, 2, 4, 2, 2,0,1, 1, 2,1

 d i  25  4  16  4  4  0  1  1  4  1  60
2

6 d i2 6  60 360
Hence   1   1 3  1  0.6 nearly.
n n
3
10  10 990
Example: Three judges, A, B, C, give the following ranks. Find which pair of judges has
common approach
A 1 6 5 10 3 2 4 9 7 8
B 3 5 8 4 7 10 2 1 6 9
C 6 4 9 8 1 2 3 10 5 7
 ( x, y )  0.2,  ( y , z )  0.3,  ( z , x )  0.6

Since  ( z , x )  0.6 is maximum, the pair of judges A and C have the nearest common approach.

Example: Calculate the rank correlation coefficient from the following data showing ranks of 10
students in two subjects
Maths 3 8 9 2 7 10 4 6 1 5
Physics 4 9 10 1 8 7 3 4 2 6

35
The Spearman rank correlation for repeated ranks is given by

 2 2 
6  d 2  m1 (m1  1)  m2 (m2  1)  ...
  1  12
3
12 
n n
Where m1 , m2 ,... are the number of items whose ranks are common.

Example: Find rank correlation for the following data

x 56 42 72 36 63 47 55 49 38 42 68 60
y 147 125 160 118 149 128 150 145 115 140 152 155

Example: Find rank correlation for the following data showing rank of 10 students in two tests
Student A B C D E F G H I J
Test 1 70 68 67 55 60 60 75 63 60 72
Test 2 65 65 80 60 68 58 75 63 60 70

36