THE LOGIT AND PROBIT MODELS of Regression Analysis

Damodar Gujarati
Econometrics by Example, second edition
CHAPTER 8
THE LOGIT AND PROBIT MODELS

QUALITATIVE RESPONSE REGRESSION MODELS
Regression models involving nominal scale
dependent variables are among a broader class
of models known as qualitative response
regression models.
Here we will consider the simplest of such
models, the binary or dichotomous or dummy
dependent variable regression models.
Damodar Gujarati

PROBLEMS WITH LINEAR PROBABILITY MODEL
 The linear probability model (LPM) uses the OLS method to
determine the probability of an outcome.
 Problems:
 1. The LPM assumes that the probability of the outcome moves linearly with
the value of the explanatory variable, no matter how small or large that value
is.
 2. The probability value must lie between 0 and 1, yet there is no guarantee
that the estimated probability values from the LPM will lie within these
limits.
 3. The usual assumption that the error term is normally distributed cannot
hold when the dependent variable takes only values of 0 and 1, since it
follows the binomial distribution.
 4. The error term in the LPM is heteroscedastic, making the traditional
significance tests suspect.
Damodar Gujarati

THE LOGIT MODEL
 If the outcome decision depends on an unobservable utility index,
then this index can be expressed as:
 If a person’s utility index I exceeds the threshold level I*, the
dichotomous outcome Y=1, and if not, then Y=0:
Yi = 1 if
Yi = 0 if
 Therefore:
Damodar Gujarati
*
i i i
I BX u
 
*
Pr( 1) Pr( 0)
Pr( ) 0
Pr[ ( )]
i
i i
i i
Y I
BX u
u BX
  
  
 
*
0
i
I 
*
0
i
I 

THE LOGIT MODEL (CONT.)
 If the probability is symmetric around zero, then:
 Therefore:
 ui is assumed to follow a cumulative probability distribution,
where the probability that Y=1 is given as:
where + ui
 P(Y=0) is given as:
Damodar Gujarati
Pr( ) Pr( )
i i i i
u BX u BX
  
Pr( 1)
Pr( )
i i
i i
P Y
u BX
 
 
1
1 i
i Z
P
e 


i i
Z BX

1
1
1 i
i Z
P
e
 


THE LOGIT MODEL (CONT.)
 The ratio of the two probabilities is the odds ratio in favor of the
outcome:
 The log of the odds ratio (logit) is given as:
+ ui
Damodar Gujarati
1
1 1
i
i
i
Z
Z
i
Z
i
P e
e
P e

 
 
ln
1
i
i i i
i
P
L Z BX
P
 
  
 

 

CHARACTERISTICS OF LOGIT MODEL
 1. As Pi goes from 0 to 1, Li goes from -∞ to ∞.
 2. Although Li is linear in Xi, the probabilities themselves are not.
 3. If Li is positive, when the value of the explanatory variable(s) increases, the odds
of the outcome increase. If it negative, the odds of the outcome decrease.
 4. The interpretation of the logit model is as follows: Each slope coefficient shows
how the log of the odds in favor of the outcome changes as the value of the X
variable changes by a unit.
 5. Once the coefficients of the logit model are estimated, we can easily compute
the probabilities of the outcome.
 6. In the LPM the slope coefficient measures the marginal effect of a unit change in
the explanatory variable on the probability of the outcome, holding other variables
constant. In the logit model, the marginal effect of a unit change in the explanatory
variable not only depends on the coefficient of that variable but also on the level of
probability from which the change is measured. The latter depends on the values
of all the explanatory variables in the model.
Damodar Gujarati

THE PROBIT MODEL
 In the probit model, the error term follows a normal distribution.
 Given the assumption of normality, the probability that is less than or
equal to Ii can be computed from the standard normal cumulative
distribution function (CDF) as:
 F, the standard normal CDF, can be written as:
 If we multiply the probit coefficient by about 1.81 ( ), we will get
approximately the logit coefficient.
 The marginal effect is given by the coefficient of the variable multiplied by
the value of the normal density function evaluated for all the X values for
that individual.
Damodar Gujarati
*
i
I
*
Pr( 1 ) Pr( ) Pr( ) ( )
i i i i i i
P Y X I I Z BX F BX
      
2
/2
1
( )
2
i
BX
z
i
F I e dz


 
 
/ 3



LOGIT vs PROBIT
 Logit and probit models generally give similar results.
 The main difference between the two models is that the
logistic distribution has slightly fatter tails.
 The conditional probability Pi approaches 0 or 1 at a slower rate in
logit than in probit.
 In practice there is no compelling reason to choose one
over the other.
 Many researchers choose the logit over the probit because
of its comparative mathematical simplicity.
Damodar Gujarati

THE LOGIT AND PROBIT MODELS of Regression Analysis

More Related Content

Similar to THE LOGIT AND PROBIT MODELS of Regression Analysis (20)

More from Ameya Patekar (12)

Recently uploaded (20)

THE LOGIT AND PROBIT MODELS of Regression Analysis