Stat 101L: Lecture 14
Regression Wisdom
 Sifting
Residuals for Groups
–Display residuals versus the
explanatory variable.
–Look at the distribution of
residuals.
1
Example
 y,
Life Expectancy (years)
 x, Wealth Index
r  0 . 874
yˆ  2 . 41  7 . 71 x
2
10
6
Count
8
4
2
-15
-10
-5
0
5
10
153
Residual Life Expectancy
1
Stat 101L: Lecture 14
Interpretation
 There
appear to be several
groups of residuals.
– two large negative residuals.
– a large group between –5 and 0.
– another group between 5 and 10.
4
90
Life Expectancy
80
70
60
50
40
30
5
6
7
8
9
Wealth Index
10
7
10
11
5
15
Residual
10
5
0
-5
-10
-15
5
6
8
9
Wealth Index
11
6
2
Stat 101L: Lecture 14
Getting the “Bends”
A
fundamental assumption is that
the relationship is a straight line.
 What looks straight on a scatter
plot may show a curve when one
looks at the plot of residuals
versus the explanatory variable.
7
Example
 y,
Stopping distance (feet)
 x, Speed (miles per hour)
R 2  0 . 984
yˆ   62 . 8  3 . 48 x
8
200
Distance
150
100
50
0
0
10
20
30
40
Speed
50
60
70
80
9
3
Stat 101L: Lecture 14
10
Residual
5
Under predict
Under predict
0
Over predict
-5
-10
0
10
20
30
40
50
Speed
60
70
80
10
Interpretation
 There
is a curved pattern in the
residuals.
– under predicts
– over predicts
– under predicts
11
Interpretation
 Although
the straight line does
a very good job explaining the
variation in stopping distance,
a curved relationship model
would do even better.
12
4
Stat 101L: Lecture 14
Dangers of Extrapolation
 Suppose
we use the least
squares equation relating speed
to stopping distance for a
vehicle traveling at 5 mph?
 The predicted stopping
distance is –45.4 feet.
13
Special Points
 Outlier
– In regression, this is
a point with a large residual.
 Leverage – In regression a
point has high leverage if it is
an extreme value for the
explanatory variable.
14
Influence
 Outliers
and high leverage
points can be influential
points, that is, they can greatly
influence what the intercept
and the slope of the least
squares line will be.
15
5