Nonlinear Relationships

Students in a physics class are studying free-fall to determine the relationship between the distance an object has fallen and the amount of time since release.

Time
(sec)

Distance
(cm)

Time
(sec)

Distance
(cm)

.16

12.1

.57

150.2

.24

29.8

.61

182.2

.25

32.7

.61

189.4

.30

42.8

.68

220.4

.30

44.2

.72

250.4

.32

55.8

.72

261.0

.36

63.5

.83

334.5

.36

65.1

.88

375.5

.50

124.6

.89

399.1

.50

129.7

 

 

1.     Draw a scatter plot of the data.

2.     Even though it doesn’t look linear find the regression line.

3.     Find the median-median line.

4.     Look at the residuals for both equations and find the standard error.

In a residual plot, positive and negative values should occur randomly.

The relationship between y and x need not be linear.  We can examine the shape of the relationship with a scatter plot and look for more detailed information by plotting the residuals from the median-median line.  If either the original or residual plot shows a bend and if the y-versus-x plot shows a generally consistent trend either up or down rather than a cup shape, we may be able to straighten the y-versus-x relationship by re-expressing one or both variables.

From the summary points of the median-median line, we compute the half-slopes:

and then find the half-slope ratio: .  If the half-slopes are equal, then the relationship is linear and the half-slope ratio is 1.  If the half-slope ratio is not close to one, then re-expressing x or y or both may help.  If the half-slope ratio is negative, the half-slopes have different signs, and re-expression will not help.

If the half-slopes are not equal, the two line segments will meet and form an angle.  We can think of the angle as an arrowhead that points toward re-expressions on the ladder of powers that might make the relationship straighter.  To determine how we might re-express y, we need to know if the arrow points more upward—towards higher values of y—or more downward—toward lower y-values.  To determine how we might re-express x, we need to know if the arrow points more to the right—toward higher x values—or more to the left—toward lower x values.

The rule for selecting a re-expression to straighten a plot is that we consider moving the expression of y or x in the direction the arrow points.  That is if the arrow points down toward lower y we might try re-expressions of y lower in the ladder of powers.  If the arrow points to the right, toward higher x, we might try re-expressions of x higher on the ladder of powers.

The half-slopes will suggest re-expressions for both x and y.  We may choose to re-express either y or x or both. 

1.     Do this for our data.

Ladder of Powers

p

Re-expression

Name

Notes

 

 

Higher powers can be used.

3

Cube

 

2

Square

One of the most commonly used powers

1

“Raw”

No re-expression necessary

½

Square root

A commonly used power, especially for counts

(0)

Logarithm

 

–½

Reciprocal root

The minus sign preserves order.

-1

Reciprocal

 

-2

Reciprocal square

 

 

 

Lower powers can be used.

 

1.     Let’s re-express our data.

2.     Re try the fit.

3.     There are other ways to think of re-expression.


Gas Mileage and Displacement for Some 1976 Automobiles

Automobile

mpg

Displacement

Mazda RX-4

21.0

160.0

Mazda RX-4 Wagon

21.0

160.0

Datsun 710

22.8

108.0

Hornet 4-Drive

21.4

258.0

Hornet Sportabout

18.7

360

Valiant

18.1

225

Plymouth Duster

14.3

360

Mercedes 240D

24.4

146.7

Mercedes 230

22.8

140.8

Mercedes 280

19.2

167.6

Mercedes 280C

17.8

167.6

Mercedes 450SE

16.4

275.8

Mercedes 450SL

17.3

275.8

Mercedes 450SLC

15.2

275.8

Cadillac Fleetwood

10.4

472

Lincoln Continental

10.4

460

Chrysler Imperial

14.7

440

Fiat 128

32.4

78.7

Honda Civic

30.4

75.7

Toyota Corolla

33.9

71.1

Toyota Corona

21.5

120.1

Dodge Challenger

15.5

318.0

AMC Javelin

15.2

304

Camaro Z28

13.3

350

Pontiac Firebird

19.2

400

Fiat X1-9

27.3

79

Porsche 914-2

26.0

120.3

Lotus Europa

30.4

95.1

Ford Pantera L

15.8

351

Ferrari Dina 1973

19.7

145

Maserati Bora

15.0

301

Volvo 142E

21.4

121