Model 5: Limited Dependent Variable Models and Sample Selection Corrections

Question 1

What is the linear probability model?

Answer 1

When the dependent variabe is a binary variable.

Question 2

Given the following regression, interpret train. Where,

em78 = 0.878+0.144train+0.005educ-0.528unem75

em78 = employed in 1978, train = recieved training, and…

unemp75 = unemployed in 1975

Answer 2

The probability of being employed in 1978 is 0.144 (or 14.4 percentage points) higher for men who received job training in 1977.

Question 3

Given the regresssion, interpret educ

em78 = 0.878+0.144train+0.005educ-0.528unem75

em78 = employed in 1978, train = recieved training, and…

unemp75 = unemployed in 1975

Answer 3

An additional year of education increases the probabilty that a man will be employed in 1978 by 0.5 percentage points.

Question 4

What are the drawbacks of LPM?

Answer 4

1)Predicted values may be greater or less than zero.
2)The probabilty increases by the same amount when xj increases by one unit
3)There is heteroskedasticity when y is a dummy

Question 5

What is the Maximum Likelihood Estimator?

Answer 5

When the estimators are chosen to maximize the probabilty that the model will produce predicted outomes that are equal to the actual observed y

Question 6

What are the properties of probit and logit models?

Answer 6

1) they are always within 0,1
2) Partial effects of explanatory variables are the largest when measured at the values of x that are close to average values and smallest when measuted at extremes
3) Assume homoskedasticity.

Question 7

T/F: In probit and logit model, the coefficients only show the direction of the effect and not the direc quantitative interpretation.

Answer 7

True

Question 8

What is the partial (marginal) effect?

Answer 8

It’s the predicted change in the probability that y = 1 when x increases by 1 unit, holding everything fixed

Question 9

What are the two ways to calculate the partial effects in the probit/logit model?

Answer 9

1) Evaluate the partial effects at the mean values of the other variables (margins, dydx() at means)
2) Calculate the average partial effects (APE) based on every value of the other variables (margins, dydx())

Question 10

T/F: We preform hypothesis testing for probit/logit on the partial effects.

Answer 10

False: It’s performed on the coefficients

Question 11

T/F: To compare probit/logit results, we need to compare the partial effects

Answer 11

True

Question 12

Why are truncated regressions (nonrandom sample selections) a problem?

Answer 12

Can generate biased estimates because only a certain set of people are selected