1. What is the binary choice model?

What are the rationales of using the binary

choice model? Mention some of the examples.

Answer:

Binary choice models are statistical models used to analyze situations where the outcome
variable can take on only two possible values, typically coded as 0 and 1. These models are
widely used in various fields, including economics, finance, marketing, and social sciences.

Rationales for Using Binary Choice Models:

i. Modeling Binary Outcomes: These models are specifically designed to handle

situations where the outcome of interest is a binary decision or choice, such as:

o Whether or not to buy a product

o Whether or not to vote for a particular candidate

o Whether or not to default on a loan

o Whether or not to attend college

ii. Understanding Factors Influencing Choices: Binary choice models help identify the
factors that influence individuals' decisions. By analyzing the relationship between
these factors and the binary outcome, researchers can gain insights into the underlying
decision-making process.

iii. Making Predictions: These models can be used to predict the probability of an
individual making a particular choice based on their characteristics and the factors
influencing their decision.

iv. Policy Evaluation: Binary choice models can be used to evaluate the effectiveness of
policies or interventions aimed at influencing individual choices.

Examples of Binary Choice Models:

1. Logit Model: This is one of the most commonly used binary choice models. It assumes
that the probability of the outcome being 1 follows a logistic distribution.

2. Probit Model: Similar to the logit model, but it assumes that the probability of the
outcome being 1 follows a standard normal distribution.

3. Linear Probability Model (LPM): While simpler, the LPM has limitations as it can
predict probabilities outside the range of 0 to 1.
 2. What is Linear Probability Model (LPM)? What are the limitations of LPM?

Answer:

The Linear Probability Model (LPM) is a statistical method used to model the relationship
between a binary dependent variable (one that can only take two values, typically 0 or 1) and
one or more independent variables. It's a simple approach where the probability of the
dependent variable being 1 is modeled as a linear function of the independent variables.

Limitations of LPM

1. Predicted Probabilities Outside the Range of 0-1: The biggest issue with LPM is that it
can produce predicted probabilities that are less than 0 or greater than 1. Since
probabilities must lie between 0 and 1, this makes the model's predictions unrealistic
and difficult to interpret.

2. Heteroscedasticity: The variance of the error term in LPM is not constant, violating the
assumption of homoscedasticity required for valid statistical inference. This can lead to
unreliable estimates of the model's parameters and standard errors.

3. Non-Normality of Errors: The errors in LPM are not normally distributed, another
assumption violated by the model. This can affect the accuracy of hypothesis tests and
confidence intervals.

4. Questionable R2

3. What are the odds & log of odds?

Answer:

Odds and log of odds are related concepts used to express the likelihood of an event.

Odds are a way of expressing the likelihood of an event occurring. They represent the ratio of
the probability of the event happening to the probability of the event not happening.

• Formula: Odds = Probability of event / Probability of event not happening

• Example: If the probability of rain tomorrow is 0.4 (40%), then the odds of rain are 0.4
/ (1 - 0.4) = 0.4 / 0.6 = 2/3 or 2 to 3.

Log of Odds: The log of odds is the natural logarithm of the odds. It's a transformation used
in statistical modeling, particularly in logistic regression.3

• Formula: Log of Odds = ln(Odds) = ln(Probability of event / Probability of event not

happening)
 Why use log of odds?

o Linearity: In logistic regression, the log of odds has a linear relationship with the
predictor variables, making it easier to model.

o Range: The log of odds can take any value, unlike probability which is restricted
between 0 and 1.

4. What is a Logit model? Explain some of the features of the logit model?

Answer:

The Logit Model is a statistical model used to analyze the relationship between a binary
dependent variable (one that can take on only two values, typically 0 or 1) and one or more
independent variables.1 It's a popular alternative to the Linear Probability Model (LPM) and
overcomes many of its limitations.

Key Features of the Logit Model:

i. Based on Logistic Distribution: The Logit Model assumes that the probability of the
dependent variable being 1 follows a logistic distribution. This ensures that the
predicted probabilities always fall within the range of 0 to 1, unlike the LPM.

ii. Uses Log-Odds: The Logit Model models the log-odds of the event occurring.3 This
transformation allows for a linear relationship between the log-odds and the
independent variables, making it easier to estimate and interpret the model.

iii. Estimates Probabilities: The Logit Model estimates the probability of the dependent
variable being 1 for different values of the independent variables. This allows for
predictions and insights into the likelihood of the event occurring.

iv. Handles Heteroscedasticity: The Logit Model inherently addresses the issue of
heteroscedasticity, which is a common problem in the LPM.

v. Widely Applicable: The Logit Model has a wide range of applications in various fields,
including economics, finance, marketing, social sciences, and medicine.

5. Why Logit and Probit model differs from the LPM model?

Answer:

The Logit and Probit models offer significant advantages over the Linear Probability Model
(LPM) when dealing with binary dependent variables (those that can only take on two values,
like 0 or 1). Here's a breakdown of their key differences:

1. Predicted Probabilities:
 • LPM: Can predict probabilities outside the 0-1 range. This is a major limitation as
probabilities must lie between 0 and 1.

• Logit & Probit: Ensure that predicted probabilities always fall within the 0-1 range.
This is because they are based on underlying distributions (logistic for Logit, normal
for Probit) that are bounded within these limits.

2. Heteroscedasticity:

• LPM: Suffers from heteroscedasticity, meaning the variance of the error term is not
constant across different values of the independent variables.1 This violates a key
assumption of linear regression.

• Logit & Probit: Inherently address the issue of heteroscedasticity, providing more
robust and reliable estimates.

3. Underlying Distributions:

• LPM: Assumes a linear relationship between the independent variables and the
probability of the outcome, which may not always hold in reality.

• Logit: Assumes that the probability of the outcome follows a logistic distribution.

• Probit: Assumes that the probability of the outcome follows a standard normal
distribution.

4. Interpretation of Coefficients:

• LPM: Coefficients directly represent the change in the probability of the outcome for
a one-unit change in the independent variable. However, as mentioned earlier, this can
lead to unrealistic predictions.

• Logit & Probit: Coefficients represent the change in the log-odds of the outcome for
a one-unit change in the independent variable.2 These coefficients are not as easily
interpretable as in LPM, but they provide a more accurate representation of the
relationship between the variables.

6. What is the probit model? How is it different from the logit model? Show
graphically & theoretically?

Answer:

The Probit Model is a statistical method used to analyze the relationship between a binary
dependent variable (one that can take on only two values, typically 0 or 1) and one or more
independent variables. It's a popular alternative to the Logit model and the Linear Probability
Model (LPM).
Differences between Probit and Logit Models

1. Underlying Distribution:

• Logit: Assumes the probability follows a logistic distribution.

• Probit: Assumes the probability follows a standard normal distribution.

2. Link Function:

• Logit: Uses the logit link function (log of odds).

• Probit: Uses the probit link function (inverse of the standard normal CDF).

3. Shape of the S-curve:

• Logit: The S-curve of the Logit model has slightly heavier tails than the Probit model.
This means it can handle extreme values in the predictors or outcomes slightly better.

• Probit: The S-curve of the Probit model is steeper near the center and approaches the
axes more quickly than the Logit model. This implies that the Probit model might be
less accommodating of outliers compared to the Logit model.

Applied Part

Answer:
a) Factors influencing Homeownership:

Several factors can significantly impact whether a household owns a house or not. Here are
some key ones:

• Family Income: Higher income generally increases the likelihood of homeownership.

• Household Size & Composition: Larger families or those with children may have a
greater need for space and thus be more likely to own a house.

• Education Level: Higher education levels are often associated with higher incomes
and, consequently, a greater likelihood of homeownership.

• Age of Household Head: Older individuals are more likely to have accumulated
wealth and, therefore, own a house.

• Location & Housing Market Conditions: Local housing prices, availability of

affordable housing, and the overall economic climate of the region all play a role.

b) Interpretation of the Equation & Coefficient Significance:

The equation is:

Ŷi = -0.9457 + 0.1021X

where:

• Ŷi is the predicted probability of homeownership for household i

• X is the family income in thousands of dollars

Interpretation:

• The intercept (-0.9457) suggests that for a household with zero income, the predicted
probability of homeownership is negative. This is not realistic in the context of
homeownership and highlights a limitation of the linear probability model (LPM).

• The coefficient of family income (0.1021) indicates that for every $1,000 increase in
income, the predicted probability of homeownership is expected to increase by
0.1021.

Statistical Significance:

Both coefficients have t-statistics that are much larger than the critical values for a typical
significance level (e.g., 0.05). This suggests that both the intercept and the slope coefficient
are statistically significant, meaning they are unlikely to have occurred by chance.

c) R² and Goodness of Fit:

R² of 0.8048 indicates that approximately 80.48% of the variation in homeownership can be
explained by the variation in family income. This suggests a strong relationship between
income and homeownership.

d) Interpretation if X=15:

If X (family income) is $15,000, the predicted probability of homeownership is:

Ŷi = -0.9457 + 0.1021 * 15 = 0.6468

This means that a household with an income of $15,000 has a predicted probability of
homeownership of approximately 64.68%.

e) Regressand and Zero-to-One Range:

In the LPM, the regressand (predicted probability) is not guaranteed to be within the 0-to-1
range. This is a major drawback of the LPM. In cases where the predicted probability falls
outside this range, the model's predictions become unrealistic.

f) Heteroscedasticity:

Heteroscedasticity refers to the situation where the variance of the error term is not constant
across1 different values of the explanatory variable(s). In the LPM, heteroscedasticity is often
a concern because the variance of the error term tends to be larger for extreme values of the
predicted probabilities (close to 0 or 1).

Solving Heteroscedasticity (Method):

One common method to address heteroscedasticity in the LPM is by using Weighted Least
Squares (WLS). WLS assigns different weights to observations based on their predicted
probabilities, giving more weight to observations with lower predicted probabilities and less
weight to those with higher predicted probabilities.

g) Marginal Effect and Explanatory Variable:

The marginal effect in the LPM is the coefficient of the explanatory variable (income in this
case). It represents the change in the predicted probability of homeownership for a one-unit
change in the explanatory variable.

In this model, the marginal effect of income is 0.1021, indicating that a $1,000 increase in
income is associated with a 0.1021 increase in the predicted probability of homeownership.
Answer:

a) Statistical Significance of Coefficients:

• Both the intercept and the income coefficient have very high t-statistics (in absolute
value) and are statistically significant at conventional levels (e.g., 5% or 1%). This
indicates that the estimated coefficients are unlikely to be zero by chance.

b) Interpretation of the Equation:

• Intercept: The intercept of -1.6587 represents the estimated log-odds of

homeownership for a household with zero income. However, this interpretation is not
directly meaningful in the context of homeownership as it's unlikely for a household
to have zero income.

• Income Coefficient: The coefficient of 0.0792 indicates that for every $1,000
increase in family income, the estimated log-odds of homeownership are expected to
increase by 0.0792.

• Interpretation in terms of Odds:

o To interpret the income coefficient in terms of odds, we can exponentiate it:

▪ exp(0.0792) ≈ 1.082

o This means that for every $1,000 increase in income, the odds of
homeownership are estimated to increase by a factor of 1.082 (or by about
8.2%).

c) Finding P when X = 20:

1. Calculate the estimated log-odds:

 o [\hat{L}_{i} = -1.6587 + 0.0792 * 20 = -1.6587 + 1.584 = -0.0747]

2. Calculate the odds:

o Odds = exp(-0.0747) ≈ 0.9277

3. Calculate the probability (P):

o P = Odds / (1 + Odds) = 0.9277 / (1 + 0.9277) ≈ 0.4813

o Therefore, the estimated probability of homeownership for a household with

an income of $20,000 is approximately 48.13%.

d) Marginal Effect when X = 25:

The marginal effect in a Logit model is not constant. It depends on the value of the
independent variable (X).

1. Calculate the estimated log-odds when X = 25:

o [\hat{L}_{i} = -1.6587 + 0.0792 * 25 = -1.6587 + 1.98 = 0.3213]

2. Calculate the marginal effect:

o Marginal Effect = Coefficient * (1 - Probability) * Probability

▪ = 0.0792 * (1 - 0.579) * 0.579 ≈ 0.0167

o When the income is $25,000, a $1,000 increase in income is estimated to

increase the probability of homeownership by approximately 0.0167.

e) Problems of using LPM instead of Logit:

• Predicted Probabilities Outside 0-1 Range: LPM can produce predicted

probabilities that are less than 0 or greater than 1, which are unrealistic.

• Heteroscedasticity: LPM assumes constant variance of the error term, which is often
violated in practice.

• Non-normality of Errors: The errors in LPM are not normally distributed, which can
affect the validity of statistical inferences.
Answer:

a. Interpretation of Coefficients:

• Intercept (1.1387): This represents the log-odds of a high murder rate when all the
independent variables (population size, growth rate, and reading quotient) are zero.
However, this interpretation might not be practically meaningful in this context as it's
unlikely to have an SMSA with zero population or growth rate.

• Population Size (0.0014): This coefficient suggests that a one-thousand-unit increase

in the 1980 population size is associated with a 0.0014 increase in the log-odds of a
high murder rate, holding other variables constant.

• Population Growth Rate (0.0561): This coefficient indicates that a one-unit increase
in the population growth rate is associated with a 0.0561 increase in the log-odds of a
high murder rate, holding other variables constant.

• Reading Quotient (-0.4050): This negative coefficient suggests that a one-unit

increase in the reading quotient is associated with a 0.4050 decrease in the log-odds of
a high murder rate, holding other variables constant. This implies that higher reading
quotients are associated with lower odds of high murder rates.

b. Statistically Significant Coefficients:

To determine statistical significance, we can use the t-statistic, which is calculated as the
coefficient divided by its standard error. As a rule of thumb, if the absolute value of the t-
statistic is greater than 2, we can generally consider the coefficient statistically significant at
the 5% level.

• Population Size: t-statistic = 0.0014 / 0.0009 ≈ 1.56. Not statistically significant.

• Population Growth Rate: t-statistic = 0.0561 / 0.0227 ≈ 2.47. Statistically significant.

• Reading Quotient: t-statistic = -0.4050 / 0.1568 ≈ -2.58. Statistically significant.

Therefore, the coefficients for population growth rate and reading quotient are statistically
significant at the 5% level.

c. Effect of a Unit Increase in Reading Quotient on Odds:

To find the effect on the odds, we can exponentiate the coefficient:

• exp(-0.4050) ≈ 0.667

This means that a one-unit increase in the reading quotient is associated with a decrease in the
odds of a high murder rate by a factor of approximately 0.667.

d. Effect of a Percentage Point Increase in Population Growth Rate on Odds:

Since the population growth rate is already in percentage points, the coefficient itself
represents the effect on the log-odds for a one percentage point increase.

• exp(0.0561) ≈ 1.057

This means that a one percentage point increase in the population growth rate is associated
with an increase in the odds of a high murder rate by a factor of approximately 1.057.

4. For the Bernoulli distribution show that 𝑣𝑎𝑟(𝑢𝑖)=𝑝(1−𝑝).

Answer:

Bernoulli Distribution: A Bernoulli random variable, often denoted as X, can only take on
two values: 0 or 1. It represents a single trial with two possible outcomes: success (X=1) with
probability p, and failure (X=0) with probability (1-p).

Expected Value (E[X]):

o E[X] = (0 * (1-p)) + (1 * p) = p

Expected Value of X squared (E[X^2]):

o E[X^2] = (0^2 * (1-p)) + (1^2 * p) = p

Variance (Var(X)):

o Var(X) = E[X^2] - (E[X])^2

o Var(X) = p - p^2
 o Var(X) = p(1-p)

Therefore, we have shown that Var(X) = p(1-p) for a Bernoulli distribution.

This makes sense intuitively. The variance is highest when p = 0.5 (meaning success and
failure are equally likely), and it decreases as p gets closer to 0 or 1.

5. Show the marginal effect of logit model.

Answer:

In a logit model, the probability of an event occurring (Y = 1) is modeled as:

P(Y = 1) = 1 / (1 + exp(-Z))

where:

• Z = β₀ + β₁X₁ + β₂X₂ + ... + βₖXₖ

• β₀ is the intercept

• β₁, β₂, ..., βₖ are the coefficients for the independent variables X₁, X₂, ..., Xₖ

Marginal Effect

The marginal effect of a variable (let's say X₁) represents the change in the probability of the
event (Y = 1) for a one-unit increase in that variable, holding other variables constant.

Derivation

1. Expressing Probability in Terms of Z:

o Let P(Y = 1) = P

o P = 1 / (1 + exp(-Z))

o Rearranging:

▪ 1/P - 1 = exp(-Z)

▪ exp(Z) = (1/P) - 1

▪ Z = ln((1/P) - 1)

2. Differentiating Z with respect to X₁:

o dZ/dX₁ = β₁
 3. Differentiating P with respect to Z:

o dP/dZ = P * (1 - P)

o This can be derived using the chain rule and the properties of the logistic
function.

4. Applying the Chain Rule:

o dP/dX₁ = (dP/dZ) * (dZ/dX₁)

o dP/dX₁ = P * (1 - P) * β₁

Therefore, the marginal effect of X₁ on the probability of the event (Y = 1) in a logit

model is:

Marginal Effect = β₁ * P * (1 - P)