MBA937: Causal Inference Models

Name: Om Ashish Mishra

Roll No: 23363025

5 MCQs:

Ans:

b. Multicollinearity

The estimation equation described essentially suffers from multicollinearity, as indicated by the
multiple terms in the equation that might correlate with each other. When variables in a regression
model are highly correlated, it becomes challenging to determine their individual effects, leading to
unstable estimates.

Here's a simple explanation of why the other answers are less likely:

a. Homoscedasticity: This refers to consistent variance in error terms across all levels of the
independent variable. The problem in the given equation does not suggest an issue with variance.

c. Non-random sample: This issue arises when the sample data is not representative of the overall
population. The provided context does not indicate sampling bias.

d. Non-normal distribution of the error term: This occurs when the residuals from the regression
model do not follow a normal distribution. There is no indication in the given context that this is a
concern.

Thus, the best answer to the problem outlined in the equation is b. Multicollinearity.
Ans:

c. A, C

The question presents a scenario where the relationship between BMI and Vitamin E intake is
studied, initially without controls, and then with gender and age as control variables. By looking at
the updated regression equation with controls, the following statements are evaluated:

A - Some of the relationship between BMI and vitamin E is explained by age and/or gender.

This is true. By adding age and gender as control variables, and seeing the coefficient for BMI
change, it indicates that some of the variability in Vitamin E intake that was attributed to BMI is
explained by these control variables.

B - Younger people are more likely to intake vitamin E.

This statement cannot be confirmed as true based on the given equations. The coefficient for age is
positive, suggesting that as age increases, so does Vitamin E intake.

C - The coefficient of BMI, i.e. 0.005, is a measure of that part of the relationship between BMI and
Vitamin E, that is explained by gender and age.

This is true. After controlling for age and gender, the coefficient for BMI (0.005) represents the
portion of the relationship between BMI and Vitamin E intake that is independent of age and
gender.

So the correct statements are A and C, making the correct choice from the given options c (A, C).
Ans:

1. b. Minimizes
2. b. Positive
3. b. related
Ans:

b (A, C)

The results of regressions were performed to study the impact of the number of locations on the
health inspection scores of restaurants, with and without controlling for the year of inspection.

Based on the given tables, here are the interpretations for each statement:

A - The addition of the Year of Inspection as a control did change the estimate.

This is true. The coefficient for the number of locations is the same in both models (-0.019), but the
addition of the year of inspection in the second model has a coefficient of its own (-0.065), which
means it has some explanatory power on the inspection score. However, the presence of the year of
inspection does not change the coefficient for the number of locations.

B - 6.5% of the variation in Inspection Score is predicted by the Number of Locations.

This statement seems to misunderstand the R2 value. The R2 value indicates the proportion of the
variance for a dependent variable that is explained by an independent variable(s) in a regression
model. In this case, the R2 value of 0.065 or 0.068 indicates that approximately 6.5% of the variation
in the inspection score is explained by the model as a whole, not just by the number of locations.

C - We infer that comparing two restaurants, the one that is part of a chain with one more location
than the other will on average have a lower inspection score.

This is true. The negative coefficient for the number of locations indicates that else equal, a
restaurant with one additional location is associated with a 0.019-point decrease in the inspection
score.

The correct statements are A and C, so the correct answer from the given options is b (A, C).

Ans:

b (A, B)
From the plot and the options given, the following conclusions can be drawn:

A - The regression equation for this graph should include an interaction term.

This is correct. The different slopes of the lines for urban, suburban, and rural communities suggest
that the relationship between income and attitude towards gun control varies by community type.
An interaction term would allow the model to account for these differences.

B - A significant interaction term means a better fit to the data and better predictions from the
regression equation. However, it also means uncertainty about the relative importance of main
effects.

This is also correct. An interaction term can improve the fit of the model to the data by capturing the
effect of one variable on the relationship between another variable and the outcome. However, it
can complicate the interpretation of the main effects because the effect of one predictor on the
outcome depends on the level of another predictor.

C - For ease of interpretation we drop the suburban community sub-sample from data. While
developing a linear model we also add an interaction term (Income * Community), where
community takes a value of 0 for urban and 1 for rural. The coefficient of the interaction term will
be positive.

This statement is not supported by the information provided. Dropping the suburban community
from the analysis could oversimplify the model and potentially lead to biased results. Furthermore,
the plot does not provide enough information to determine the sign of the interaction term's
coefficient.

Based on these points, statements A and B are correct, while C is not. The correct answer is
therefore b (A, B).

2 R Output Interpretations:

First one:
Ans:

The R output provided from a linear regression analysis contains several pieces of information:

1. Regression Equation: The call indicates the model is `Performance ~ Studyh + Pre_sco + extra +
sleep_h + Sample`. This means that the performance index (Performance) is the dependent variable,
and the hours of study (Studyh), previous scores (Pre_sco), extracurricular activities (extra), sleep
hours (sleep_h), and the number of sample papers practiced (Sample) are the independent
variables.

2. Residuals: The residuals, which are the differences between observed and predicted values, range
from a minimum of -8.6333 to a maximum of 8.7932, with median close to zero, which is expected in
a well-fitting model.

3. Coefficients: The estimates show how much the dependent variable is expected to increase when
the independent variable increases by one unit, all else being equal.

- The intercept is -34.07558, but without context, it’s not clear what this means since it's unlikely
that the variables would all be zero.

- Studyh has a coefficient of 2.852982, suggesting that each additional hour of study is associated
with an increase in the performance index by approximately 2.85 points.

- Pre_sco has a coefficient of 1.018434, indicating that for each additional point in previous scores,
the performance index is expected to increase by approximately 1.02 points.

- Extracurricular activities (extra) and sleep hours (sleep_h) have positive effects on performance
with coefficients of 0.612898 and 0.480560, respectively.

- Sample papers practiced (Sample) also have a positive effect with a coefficient of 0.193802.

4. Statistical Significance: The 'Pr(>|t|)' column shows p-values for the hypothesis test of each
coefficient being different from zero. All variables have very low p-values (indicated by `<2e-16`),
meaning that all the coefficients are statistically significant at common significance levels.

5. Fit of the Model: The model has an extremely high R-squared value of 0.9888, which indicates
that 98.88% of the variability in the performance index is explained by the model. However, such a
high R-squared value in a real-world dataset should be approached with caution as it may indicate
overfitting.

6. Residual Standard Error: This value provides an estimate of the standard deviation of the
residuals; in this case, it is 2.038. This gives us a measure of the typical size of the errors.

7. Degrees of Freedom: The model was fit using 10,000 observations (as indicated by the residual
degrees of freedom, 9994, which is total observations minus the number of estimated parameters).

8. F-statistic: The F-statistic and its associated p-value test the null hypothesis that all of the
regression coefficients are equal to zero. The very low p-value (near zero) suggests that the model is
statistically significant.
Second one:

Ans:

The R output presented indicates the results of a multiple linear regression analysis that seeks to
understand how different variables affect insurance charges. Here's the interpretation of the key
elements from the output:

1. Regression Formula: The model is predicting `charges` using `age`, `gender`, `bmi`, `children`,
`smo` (presumably smoking status), and regions (`northeast`, `northwest`, `southeast`, `southwest`).

2. Coefficients and Significance:

- `age`: For each additional year of age, insurance charges increase by approximately $256.86,
highly significant.

- `gender`: The coefficient for gender is not significant (p > 0.05), suggesting gender does not have
a statistically significant effect on insurance charges in this model.

- `bmi`: For each unit increase in BMI, charges increase by approximately $339.19, highly
significant.

- `children`: For each additional child, insurance charges increase by $475.50, significant.

- `smo`: Being a smoker is associated with an increase in insurance charges by $23848.53, which is
highly significant.
 - Regions are compared to a baseline (probably `region_southwest`, as its coefficient is not shown).
`region_northeast`, `region_northwest`, and `region_southeast` have their own coefficients
indicating how much more or less the insurance charges are in comparison to the baseline. Only
`region_northeast` is somewhat significant (p < 0.05).

3. Residuals: The residuals range widely, from -11304.9 to 29992.8. The median near -982.1 suggests
there might be a slight skew in the residuals since it is not close to zero.

4. Fit of the Model:

- The multiple R-squared of 0.7509 indicates that about 75.09% of the variability in insurance
charges is explained by the variables in the model.

- The adjusted R-squared of 0.7494 accounts for the number of predictors in the model and is very
close to the multiple R-squared, which indicates that most of the variables contribute information.

- The F-statistic is very large, and its corresponding p-value is less than 2.2e-16, indicating that the
model is statistically significant.

5. Residual Standard Error: The RSE of 6062 on 1329 degrees of freedom gives an estimate of the
standard deviation of the residuals; it's quite high, indicating a considerable variation in the charges
that the model doesn't explain.

6. Issue of Singularities: The note about singularities indicates a potential issue with multicollinearity
or a perfect linear relationship between some of the predictors, which can interfere with the model's
ability to estimate the individual effects of the predictors.

Overall, the model seems to be significant with a good proportion of the variance explained by the
included variables, but the significant residual standard error indicates there's still a large amount of
unexplained variability. The issue with singularities should be investigated, possibly through variance
inflation factor (VIF) analysis or checking for data entry errors.

2 Datasets:

Ans:

Notebooks Attached!