Chapter 5. Logistic Regression
Deyuan Li
School of Management, Fudan University
Feb. 28, 2011
1 / 113
Outline
•
5.1 Interpreting Parameters in Logistic Regression
•
5.2 Inference for Logistic Regression
•
5.3 Logit Models with Categorical Predictors
•
5.4 Multiple Logistic Regression
•
5.5 Fitting Logistic Regression Models
2 / 113
5.1 Interpreting Parameters in Logistic Regression
For a binary response variable
Y
and an explanatory variable
X
,let
π
(
x
)=
P
(
Y
=1

X
=
x
)=1
−
P
(
Y
=0

X
=
x
).
The logistic regression model is
π
(
x
exp(
α
+
β
x
)
1+exp(
α
+
β
x
)
.
Equivalently, the log odds, called the
logit
, has the linear
relationship
{
Insert .
.....
}
This equates the logit link function to the linear predictor.
3 / 113
5.1.1 Interpreting
β
: odds, probabilities, and linear
approximations
The sign of
β
determines whether
π
(
x
) is increasing or decreasing
as
x
increases.
The

β

determines the rate of climb or descent.
β
→
0:
the curve ﬂattens to a horizontal straight line.
β
:
Y
is independent of
X
.
For quantitative
x
with
β>
0, the curve for
π
(
x
)hastheshapeo
f
the cdf of the logistic distribution (recall Section 4.2.5).
Since the logistic density is symmetric,
π
(
x
) approaches 1 at the
same rate that it approaches 0.
4 / 113
This preview has intentionally blurred sections. Sign up to view the full version.
View Full Documentβ
and odds
Exponentiating both sides of the logit equation shows that the
odds are an exponential function of
x
:
odds =
π
(
x
)
/
[1
−
π
(
x
)] = exp(
α
+
β
x
)=
e
α
e
β
x
.
Interpretation for the magnitude of
β
:
•
the odds increase multiplicatively by
e
β
for every 1unit
increase in
x
;
•
e
β
is an odds ratio, i.e.,
odds at
X
=
x
+1
odds at
X
=
x
=
π
(
x
+1)
/
[1
−
π
(
x
+1)]
π
(
x
)
/
[1
−
π
(
x
)]
=
e
α
e
β
(
x
+1)
e
α
e
β
x
=
e
β
.
β
and linear approximation
Since
π
(
x
) changes with
x
and
∂π
(
x
)
/∂
x
=
βπ
(
x
)[1
−
π
(
x
)], the
rate of change in
π
(
x
) per unit change in
x
varies.
This expression also shows that the logistic density is symmetric,
i.e.,
π
(
x
) approaches a value
v
at the same rate that it approaches
(1
−
v
).
5 / 113
A straight line drawn tangent to the logistic regression curve at a
particular
x
value describes the rate of change at that point (see
Figure 5.1).
FIGURE 5.1
Linear approximation to logistic regression curve.
6 / 113
Example
π
(
x
)s
l
o
p
e
=
(
x
)[1
−
π
(
x
)]
1
/
2
β/
4
0
.
9o
r0
.
10
.
09
β
→
1o
→
0
The steepest slope occurs when
π
(
x
)=1
/
2
⇒
odds = 1
⇒
logit = 0
⇒
α
+
β
x
=0
⇒
x
=
−
α/β.
This
x
value (
−
α/β
) is sometimes called the
median efective level
and denoted by EL
50
.
In toxicology studies it is called LD
50
(LD = lethal dose), i.e., the
dose with a 50% chance of a lethal result.
7 / 113
From this linear approximation, near
x
where
π
(
x
/
2, a
change in
x
of 1
/β
corresponds to change in
π
(
x
) of roughly
(1
/β
)(
4) = 1
/
4;
that is, 1
/β
approximates the distance between
x
values where
π
(
x
)=0
.
25 or 0
.
75 (in reality, 0
.
27 and 0
.
73) and where
π
(
x
.
50.
The linear approximation works better for smaller changes in
x
.
This is the end of the preview.
Sign up
to
access the rest of the document.
 Spring '11
 DeyuanLi
 Regression Analysis, Logit

Click to edit the document details