probit regression

84
Lecture 9: Logit/Probit Prof. Sharyn O’Halloran Sustainable Development U9611 Econometrics II

Upload: surya-kalsi

Post on 17-Jan-2016

52 views

Category:

Documents


1 download

DESCRIPTION

probit regssion for spss

TRANSCRIPT

Page 1: Probit Regression

Lecture 9: Logit/ProbitProf. Sharyn O’Halloran Sustainable Development U9611Econometrics II

Page 2: Probit Regression

Review of Linear EstimationSo far, we know how to handle linearestimation models of the type:

Y = β0 + β1*X1 + β2*X2 + … + ε ≡ Xβ + ε

Sometimes we had to transform or add variables to get the equation to be linear:

Taking logs of Y and/or the X’sAdding squared termsAdding interactions

Then we can run our estimation, do model checking, visualize results, etc.

Page 3: Probit Regression

Nonlinear EstimationIn all these models Y, the dependent variable, was continuous.

Independent variables could be dichotomous (dummy variables), but not the dependent var.

This week we’ll start our exploration of non-linear estimation with dichotomous Y vars.These arise in many social science problems

Legislator Votes: Aye/NayRegime Type: Autocratic/DemocraticInvolved in an Armed Conflict: Yes/No

Page 4: Probit Regression

Link FunctionsBefore plunging in, let’s introduce the concept of a link function

This is a function linking the actual Y to the estimated Y in an econometric model

We have one example of this already: logsStart with Y = Xβ + εThen change to log(Y) ≡ Y′ = Xβ + εRun this like a regular OLS equationThen you have to “back out” the results

Nidhi
Highlight
Page 5: Probit Regression

Link FunctionsBefore plunging in, let’s introduce the concept of a link function

This is a function linking the actual Y to the estimated Y in an econometric model

We have one example of this already: logsStart with Y = Xβ + εThen change to log(Y) ≡ Y′ = Xβ + εRun this like a regular OLS equationThen you have to “back out” the results

Differentβ’s here

Page 6: Probit Regression

Link FunctionsIf the coefficient on some particular X is β, then a 1 unit ∆X β⋅∆(Y′) = β⋅∆[log(Y))]

= eβ ⋅∆(Y) Since for small values of β, eβ ≈ 1+β , this is almost the same as saying a β% increase in Y(This is why you should use natural log transformations rather than base-10 logs)

In general, a link function is some F(⋅) s.t.F(Y) = Xβ + ε

In our example, F(Y) = log(Y)

Nidhi
Highlight
Nidhi
Highlight
Page 7: Probit Regression

Dichotomous Independent Vars.How does this apply to situations with dichotomous dependent variables?

I.e., assume that Yi œ {0,1}First, let’s look at what would happen if we tried to run this as a linear regressionAs a specific example, take the election of minorities to the Georgia state legislature

Y = 0: Non-minority electedY = 1: Minority elected

Page 8: Probit Regression

Dichotomous Independent Vars.0

1B

lack

Rep

rese

ntat

ive

Ele

cted

0 .2 .4 .6 .8 1Black Voting Age Population

The data look like this.

The only values Y can have are 0 and 1

Page 9: Probit Regression

Dichotomous Independent Vars.

And here’s a linear fit of the data

Note that the line goes below 0 andabove 1

01

0 .2 .4 .6 .8 1Black Voting Age Population

Black Representative Elected Fitted values

Nidhi
Highlight
Page 10: Probit Regression

Dichotomous Independent Vars.The line doesn’t fit the data very well.

And if we take values of Y between 0 and 1 to be probabilities, this doesn’t make sense

01

0 .2 .4 .6 .8 1Black Voting Age Population

Black Representative Elected Fitted values

Nidhi
Highlight
Page 11: Probit Regression

Redefining the Dependent Var.

How to solve this problem?We need to transform the dichotomous Y into a continuous variable Y′ œ (-∞,∞)So we need a link function F(Y) that takes a dichotomous Y and gives us a continuous, real-valued Y′Then we can run

F(Y) = Y′ = Xβ + ε

saurabh khodke
Highlight
Page 12: Probit Regression

Redefining the Dependent Var.

0 1Original

Y

Page 13: Probit Regression

Redefining the Dependent Var.

0 1Original

Y

Y as aProbability 0 1

Page 14: Probit Regression

Redefining the Dependent Var.

0 1Original

Y

Y as aProbability 0 1

Y′œ (- ∞,∞)

?

- ∞ ∞

Page 15: Probit Regression

Redefining the Dependent Var.

What function F(Y) goes from the [0,1] interval to the real line?Well, we know at least one function that goes the other way around.

That is, given any real value it produces a number (probability) between 0 and 1.

This is the…

Page 16: Probit Regression

Redefining the Dependent Var.

What function F(Y) goes from the [0,1] interval to the real line?Well, we know at least one function that goes the other way around.

That is, given any real value it produces a number (probability) between 0 and 1.

This is the cumulative normal distribution ΦThat is, given any Z-score, Φ(Z) œ [0,1]

Nidhi
Highlight
Page 17: Probit Regression

Redefining the Dependent Var.So we would say that

Y = Φ(Xβ + ε)Φ−1(Y) = Xβ + ε

Y′ = Xβ + εThen our link function is F(Y) = Φ−1(Y) This link function is known as the Probit link

This term was coined in the 1930’s by biologists studying the dosage-cure rate linkIt is short for “probability unit”

Nidhi
Highlight
Nidhi
Highlight
Page 18: Probit Regression

Probit Estimation

After estimation, you can back out probabilities using the standard normal dist.

0.1

.2.3

.4

-4 -2 0 2 4

Page 19: Probit Regression

Probit Estimation

Say that for a given observation, Xβ = -1

0.1

.2.3

.4

-4 -2 0 2 4

Page 20: Probit Regression

0.1

.2.3

.4

-4 -2 0 2 4

Probit Estimation

Say that for a given observation, Xβ = -1

Page 21: Probit Regression

0.1

.2.3

.4

-4 -2 0 2 4

Probit Estimation

Say that for a given observation, Xβ = -1

Page 22: Probit Regression

0.1

.2.3

.4

-4 -2 0 2 4

Probit Estimation

Say that for a given observation, Xβ = -1

Prob(Y=1)

Page 23: Probit Regression

0.1

.2.3

.4

-4 -2 0 2 4

Probit Estimation

Say that for a given observation, Xβ = 2

Page 24: Probit Regression

0.1

.2.3

.4

-4 -2 0 2 4

Probit Estimation

Say that for a given observation, Xβ = 2

Page 25: Probit Regression

0.1

.2.3

.4

-4 -2 0 2 4

Probit Estimation

Say that for a given observation, Xβ = 2

Prob(Y=1)

Page 26: Probit Regression

Probit EstimationIn a probit model, the value of Xβ is taken to be the z-value of a normal distribution

Higher values of Xβ mean that the event is more likely to happen

Have to be careful about the interpretation of estimation results here

A one unit change in Xi leads to a βi change in the z-score of Y (more on this later…)

The estimated curve is an S-shaped cumulative normal distribution

Nidhi
Highlight
Nidhi
Highlight
Page 27: Probit Regression

Probit Estimation

• This fits the data much better than the linear estimation• Always lies between 0 and 1

0.2

.4.6

.81

Pro

babi

lity

of E

lect

ing

a B

lack

Rep

.

0 .2 .4 .6 .8 1Black Voting Age Population

Page 28: Probit Regression

Probit Estimation

• Can estimate, for instance, the BVAP at which Pr(Y=1) = 50%• This is the “point of equal opportunity”

0.2

.4.6

.81

Pro

babi

lity

of E

lect

ing

a B

lack

Rep

.

0 .2 .4 .6 .8 1Black Voting Age Population

Page 29: Probit Regression

Probit Estimation

• Can estimate, for instance, the BVAP at which Pr(Y=1) = 50%• This is the “point of equal opportunity”

0.2

.4.6

.81

Pro

babi

lity

of E

lect

ing

a B

lack

Rep

.

0 .2 .4 .6 .8 1Black Voting Age Population

Page 30: Probit Regression

Probit Estimation

• Can estimate, for instance, the BVAP at which Pr(Y=1) = 50%• This is the “point of equal opportunity”

0.2

.4.6

.81

Pro

babi

lity

of E

lect

ing

a B

lack

Rep

.

0 .2 .4 .6 .8 1Black Voting Age Population

Page 31: Probit Regression

Probit Estimation

• This occurs at about 48% BVAP

0.2

.4.6

.81

Pro

babi

lity

of E

lect

ing

a B

lack

Rep

.

0 .2 .4 .6 .8 1Black Voting Age Population

Page 32: Probit Regression

Redefining the Dependent Var.Let’s return to the problem of transforming Y from {0,1} to the real lineWe’ll look at an alternative approach based on the odds ratioIf some event occurs with probability p, then the odds of it happening are O(p) = p/(1-p)

p = 0 O(p) = 0p = ¼ O(p) = 1/3 (“Odds are 1-to-3 against”)p = ½ O(p) = 1 (“Even odds”)p = ¾ O(p) = 3 (“Odds are 3-to-1 in favor”)p = 1 O(p) = ∞

Page 33: Probit Regression

Redefining the Dependent Var.

So taking the odds of Y occuring moves us from the [0,1] interval…

0 1Original

Y

Y as aProbability 0 1

Page 34: Probit Regression

Redefining the Dependent Var.

So taking the odds of Y occuring moves us from the [0,1] interval to the half-line [0, ∞)

0 1Original

Y

Y as aProbability 0 1

Odds of Y∞0

Page 35: Probit Regression

Redefining the Dependent Var.Original

Y

The odds ratio is always non-negativeAs a final step, then, take the log of the odds ratio

0 1Original

Y

Y as aProbability 0 1

Odds of Y∞0

Page 36: Probit Regression

Redefining the Dependent Var.

0 1Original

Y

Y as aProbability 0 1

Odds of Y∞0

Log-Odds of Y- ∞ ∞

Page 37: Probit Regression

Logit FunctionThis is called the logit function

logit(Y) = log[O(Y)] = log[y/(1-y)]Why would we want to do this?

At first, this was computationally easier than working with normal distributionsNow, it still has some nice properties that we’ll investigate next time with multinomial dep. vars.

The density function associated with it is very close to a standard normal distribution

Page 38: Probit Regression

Logit vs. Probit0

.05

.1.1

5.2

-4 -2 0 2 4

LogitNormal

The logit function is similar, but has thinner tails than the normal distribution

Nidhi
Highlight
Page 39: Probit Regression

Logit Function

This translates back to the original Y as:

( )

( )

β

β

ββ

ββ

ββ

β

β

β

X

X

XX

XX

XX

X

X

X

eeY

eYeeYeY

YeeYeYY

eYYYY

+=

=+

=+

−=

−=

=−

=⎟⎠⎞

⎜⎝⎛

1

1

11

1log

Page 40: Probit Regression

Latent Variables

For the rest of the lecture we’ll talk in terms of probits, but everything holds for logits tooOne way to state what’s going on is to assume that there is a latent variable Y* such that

( )2* ,0~ , σεεβ NY += X

Page 41: Probit Regression

Latent Variable Formulation

For the rest of the lecture we’ll talk in terms of probits, but everything holds for logits tooOne way to state what’s going on is to assume that there is a latent variable Y* such that

( )2* ,0~ , σεεβ NY += X Normal = Probit

Page 42: Probit Regression

Latent Variables

For the rest of the lecture we’ll talk in terms of probits, but everything holds for logits tooOne way to state what’s going on is to assume that there is a latent variable Y* such that

In a linear regression we would observe Y* directlyIn probits, we observe only

⎩⎨⎧

>≤

=0 if 10 if 0

*

*

i

ii y

yy

( )2* ,0~ , σεεβ NY += X Normal = Probit

Page 43: Probit Regression

Latent Variables

For the rest of the lecture we’ll talk in terms of probits, but everything holds for logits tooOne way to state what’s going on is to assume that there is a latent variable Y* such that

In a linear regression we would observe Y* directlyIn probits, we observe only

⎩⎨⎧

>≤

=0 if 10 if 0

*

*

i

ii y

yy

( )2* ,0~ , σεεβ NY += X Normal = Probit

These could be anyconstant. Later we’llset them to ½.

Page 44: Probit Regression

Latent Variables

This translates to possible values for the error term:

Similarly,

( ) ( ) ( )

⎟⎠⎞

⎜⎝⎛ ′−

Φ=

⎟⎠⎞

⎜⎝⎛ ′−

>=

′−>===>

′−>⇒>+′⇒>

σβ

σβ

σε

βε

βεεβ

i

ii

iiiiii

iiiii

yy

y

x

xxxx

xx

Pr

Pr|1Pr|0Pr

00*

*

( ) ⎟⎠⎞

⎜⎝⎛ ′−

Φ−==σβ i

iiyxx 1|0Pr

Page 45: Probit Regression

Latent Variables

Look again at the expression for Pr(Yi=1):

We can’t estimate both β and σ, since they enter the equation as a ratioSo we set σ=1, making the distribution on ε a standard normal density.One (big) question left: how do we actually estimate the values of the b coefficients here?

(Other than just issuing the “probit” command in Stata!)

( ) ⎟⎠⎞

⎜⎝⎛ ′−

Φ==σβ i

iiyxx|1Pr

Page 46: Probit Regression

Maximum Likelihood Estimation

Say we’re estimating Y=Xβ+ε as a probitAnd say we’re given some trial coefficients β′.

Then for each observation yi, we can plug in xi and β′ to get Pr(yi=1)=Φ(xi β′).

For example, let’s say Pr(yi=1) = 0.8Then if the actual observation was yi=1, we can say its likelihood (given β′) is 0.8But if yi=0, then its likelihood was only 0.2

And conversely for Pr(yi=0)

Page 47: Probit Regression

Let L(yi | β) be the likelihood of yi given βFor any given trial set of β′ coefficients, we can calculate the likelihood of each yi.Then the likelihood of the entire sample is:

Maximum likelihood estimation finds the β’s that maximize this expression.Here’s the same thing in visual form

Maximum Likelihood Estimation

( ) ( ) ( ) ( ) ( )∏=

=⋅⋅⋅⋅n

iin yyyyy

1321 LLLLL K

Page 48: Probit Regression

0

½

1y* = xβ + ε, Probit: ℜ1#[0,1]P(y=1) = Probit(y*)

Maximum Likelihood Estimation

Page 49: Probit Regression

0

½

1y* = xβ′ + ε, Probit: ℜ1#[0,1]P(y=1) = Probit(y*)

Maximum Likelihood Estimation

Given estimates β′ of β, the distance from yi to the line P(y=1) is 1-L(yi | β′)

Page 50: Probit Regression

0

½

1y* = xβ′ + ε, Probit: ℜ1#[0,1]P(y=1) = Probit(y*)

Maximum Likelihood Estimation

1-L(y3)

Given estimates β′ of β, the distance from y3 to the line P(y=1) is 1-L(y3 | β′)

y3

Page 51: Probit Regression

0

½

1y* = xβ′ + ε, Probit: ℜ1#[0,1]P(y=1) = Probit(y*)

Maximum Likelihood Estimation

1-L(y3)

1-L(y9)

Given estimates β′ of β, the distance from y9 to the line P(y=1) is 1-L(y9 | β′)

y9

Page 52: Probit Regression

0

½

1Probit(xβ′)

Impact of changing β′…

Maximum Likelihood Estimation

Page 53: Probit Regression

0

½

1Probit(xβ′′)

Maximum Likelihood Estimation

Impact of changing β′ to β′′

Page 54: Probit Regression

0

½

1Probit(xβ′′)

Maximum Likelihood Estimation

Remember, the object is to maximize the product of the likelihoods L(yi | β)

Page 55: Probit Regression

0

½

1Probit(xβ′′)

Maximum Likelihood Estimation

Using β′′may bring regression line closer to some observations, further from others

Page 56: Probit Regression

0

½

1

Error Terms for MLE

( ) ( ) ( )βε 11*11 01 xPyPyP −≥=≥==

Maximum Likelihood Estimation

Page 57: Probit Regression

0

½

1

Time Series

t = 1 2 3… T

Maximum Likelihood Estimation

Page 58: Probit Regression

0½1

Country 1

0½1

Country 2

0½1

Country 3

0½1

Country 4

Time Series Cross Section

Page 59: Probit Regression

Maximum Likelihood Estimation

Recall that a likelihood function is:

To maximize this, use the trick of taking the log firstSince maximizing the log(L) is the same as maximizing L

( ) ( ) ( ) ( ) ( ) LLLLLL ≡=⋅⋅⋅⋅ ∏=

n

iin yyyyy

1321 K

( ) ( )

( )[ ]in

i

n

ii

y

y

L

LL

=

=

=

=

1

1

log

loglog

Page 60: Probit Regression

Maximum Likelihood EstimationLet’s see how this works on some simple examplesTake a coin flip, so that Yi=0 for tails, Yi=1 for heads

Say you toss the coin n times and get p headsThen the proportion of heads is p/n

Since Yi is 1 for heads and 0 for tails, p/n is also the sample meanIntuitively, we’d think that the best estimate of p is p/n

If the true probability of heads for this coin is ρ, then the likelihood of observation Yi is:

( )

( ) ii yy

i

ii y

yy

−−⋅=

⎩⎨⎧

=−=

=

11

0 if 11 if

ρρ

ρρ

L

Page 61: Probit Regression

Maximum Likelihood EstimationMaximizing the log-likelihood, we get

To maximize this, take the derivative with respect to ρ

( )[ ] ( )[ ]

( ) ( ) ( )ρρ

ρρρ

−−+=

−⋅=

∑∑

=

=

=

1log1log

1loglogmax

1

1

1

1

i

n

ii

n

i

yyn

ii

yy

|ρy iiL

( ) ( ) ( )

( )ρρ

ρ

ρρ

ρ

−−−=

⎥⎦

⎤⎢⎣

⎡−−+

=

=

=

1111

1log1logdlogd

1

1

i

n

ii

i

n

ii

yy

yyL

Page 62: Probit Regression

Maximum Likelihood EstimationFinally, set this derivative to 0 and solve for ρ

( )

( ) ( )[ ]

( )

( )[ ]

n

y

yn

yyy

yy

yy

n

ii

n

ii

n

iiii

n

iii

n

i

ii

=

=

=

=

=

=

=

=−+−−

=−

−−−

=⎥⎦

⎤⎢⎣

⎡−−

1

1

1

1

1

01

01

11

011

ρ

ρ

ρρρ

ρρ

ρρ

ρρ

Page 63: Probit Regression

Maximum Likelihood EstimationFinally, set this derivative to 0 and solve for ρ

( )

( ) ( )[ ]

( )

( )[ ]

n

y

yn

yyy

yy

yy

n

ii

n

ii

n

iiii

n

iii

n

i

ii

=

=

=

=

=

=

=

=−+−−

=−

−−−

=⎥⎦

⎤⎢⎣

⎡−−

1

1

1

1

1

01

01

11

011

ρ

ρ

ρρρ

ρρ

ρρ

ρρ

Magically, the value of ρ that maximizes the likelihood function is the sample mean, just as we thought.

Page 64: Probit Regression

Maximum Likelihood EstimationCan do the same exercise for OLS regression

The set of β coefficients that maximize the likelihood would then minimize the sum of squared residuals, as before

This works for logit/probit as wellIn fact, it works for any estimation equation

Just look at the likelihood function L you’re trying to maximize and the parameters β you can change Then search for the values of β that maximize L(We’ll skip the details of how this is done.)

Maximizing L can be computationally intense, but with today’s computers it’s usually not a big problem

Page 65: Probit Regression

Maximum Likelihood EstimationThis is what Stata does when you run a probit:. probit black bvap

Iteration 0: log likelihood = -735.15352Iteration 1: log likelihood = -292.89815Iteration 2: log likelihood = -221.90782Iteration 3: log likelihood = -202.46671Iteration 4: log likelihood = -198.94506Iteration 5: log likelihood = -198.78048Iteration 6: log likelihood = -198.78004

Probit estimates Number of obs = 1507LR chi2(1) = 1072.75Prob > chi2 = 0.0000

Log likelihood = -198.78004 Pseudo R2 = 0.7296

------------------------------------------------------------------------------black | Coef. Std. Err. z P>|z| [95% Conf. Interval]

-------------+----------------------------------------------------------------bvap | 0.092316 .5446756 16.95 0.000 0.081641 0.102992

_cons | -0.047147 0.027917 -16.89 0.000 -0.052619 -0.041676------------------------------------------------------------------------------

Page 66: Probit Regression

Maximum Likelihood EstimationThis is what Stata does when you run a probit:. probit black bvap

Iteration 0: log likelihood = -735.15352Iteration 1: log likelihood = -292.89815Iteration 2: log likelihood = -221.90782Iteration 3: log likelihood = -202.46671Iteration 4: log likelihood = -198.94506Iteration 5: log likelihood = -198.78048Iteration 6: log likelihood = -198.78004

Probit estimates Number of obs = 1507LR chi2(1) = 1072.75Prob > chi2 = 0.0000

Log likelihood = -198.78004 Pseudo R2 = 0.7296

------------------------------------------------------------------------------black | Coef. Std. Err. z P>|z| [95% Conf. Interval]

-------------+----------------------------------------------------------------bvap | 0.092316 .5446756 16.95 0.000 0.081641 0.102992

_cons | -0.047147 0.027917 -16.89 0.000 -0.052619 -0.041676------------------------------------------------------------------------------

Maximizing the log-likelihoodfunction!

Page 67: Probit Regression

Maximum Likelihood EstimationThis is what Stata does when you run a probit:. probit black bvap

Iteration 0: log likelihood = -735.15352Iteration 1: log likelihood = -292.89815Iteration 2: log likelihood = -221.90782Iteration 3: log likelihood = -202.46671Iteration 4: log likelihood = -198.94506Iteration 5: log likelihood = -198.78048Iteration 6: log likelihood = -198.78004

Probit estimates Number of obs = 1507LR chi2(1) = 1072.75Prob > chi2 = 0.0000

Log likelihood = -198.78004 Pseudo R2 = 0.7296

------------------------------------------------------------------------------black | Coef. Std. Err. z P>|z| [95% Conf. Interval]

-------------+----------------------------------------------------------------bvap | 0.092316 .5446756 16.95 0.000 0.081641 0.102992

_cons | -0.047147 0.027917 -16.89 0.000 -0.052619 -0.041676------------------------------------------------------------------------------

Maximizing the log-likelihoodfunction!

Coefficients aresignificant

Page 68: Probit Regression

Marginal Effects in ProbitIn linear regression, if the coefficient on x is β, then a 1-unit increase in x increases Y by β.But what exactly does it mean in probit that the coefficient on BVAP is 0.0923 and significant?

It means that a 1% increase in BVAP will raise the z-score of Pr(Y=1) by 0.0923.And this coefficient is different from 0 at the 5% level.

So raising BVAP has a constant effect on Y′.But this doesn’t translate into a constant effect on the original Y.

This depends on your starting point.

Page 69: Probit Regression

Marginal Effects in ProbitFor instance, raising BVAP from .2 to .3 has little appreciable impact on Pr(Black Elected)

0.2

.4.6

.81

Pro

babi

lity

of E

lect

ing

a B

lack

Rep

.

0 .2 .4 .6 .8 1Black Voting Age Population

Page 70: Probit Regression

Marginal Effects in ProbitBut increasing BVAP from .5 to .6 does have a big impact on the probability

0.2

.4.6

.81

Pro

babi

lity

of E

lect

ing

a B

lack

Rep

.

0 .2 .4 .6 .8 1Black Voting Age Population

Page 71: Probit Regression

Marginal Effects in ProbitSo lesson 1 is that the marginal impact of changing a variable is not constant.Another way of saying the same thing is that in the linear model

In the probit model

ii

nn

xY

xxxY

β

ββββ

=∂∂

++++= so ,11110 K

( )

( )nnii

nn

xxxxY

xxxY

ββββφβ

ββββ

++++=∂∂

++++Φ=

K

K

11110

11110 so ,

Page 72: Probit Regression

Marginal Effects in ProbitThis expression depends on not just βi, but on the value of xi and all other variables in the equationSo to even calculate the impact of xi on Y you have to choose values for all other variables xj.

Typical options are to set all variables to their means or their medians

Another approach is to fix the xj and let xi vary from its minimum to maximum values

Then you can plot how the marginal effect of xi

changes across its observed range of values

Page 73: Probit Regression

Example: Vote Choice

Model voting for/against incumbent as( )

challengerQuality name s'challenger recallCan name sincumbent' recallCan

situation financial Personal conditions economic National

incumbent as ame IDParty Constant

where, Probit

7

6

5

4

3

2

1

=======

+=

i

i

i

i

i

i

i

xxxxx

sxx

Y εβX

Page 74: Probit Regression

Example: Vote Choice

Page 75: Probit Regression

Example: Vote Choice

SignificantCoefficients

Page 76: Probit Regression

Example: Vote Choice

This backs out the marginal impact of a 1-unit change in the variable on the probability of voting for the incumbent.

Page 77: Probit Regression

Example: Vote Choice

This backs out the marginal impact of a 1-unit change in the variable on the probability of voting for the incumbent.

big

big

Page 78: Probit Regression

Example: Vote ChoiceOr, calculate the impact of facing a quality challenger by hand, keeping all other variables at their median.

Page 79: Probit Regression

Example: Vote ChoiceOr, calculate the impact of facing a quality challenger by hand, keeping all other variables at their median.

Φ(1.52)

Page 80: Probit Regression

Example: Vote ChoiceOr, calculate the impact of facing a quality challenger by hand, keeping all other variables at their median.

Φ(1.52)From standardnormal table

Page 81: Probit Regression

Example: Vote ChoiceOr, calculate the impact of facing a quality challenger by hand, keeping all other variables at their median.

Φ(1.52)From standardnormal table

Φ(1.52-.339)

So there’s an increase of .936 - .881 = 5.5% votes in favor of incumbents who avoid a quality challengers.

Page 82: Probit Regression

Example: Senate ObstructionModel the probability that a bill is passed in the Senate (over a filibuster) based on:

The coalition size preferring the bill be passedAn interactive term: size of coalition X end of session

Page 83: Probit Regression

Example: Senate ObstructionGraph the results for end of session = 0

Page 84: Probit Regression

Example: Senate ObstructionGraph the results for end of session = 1