Example of Advanced Regression Models Midterm

Exam (2024-2025)
Made by DIALLO Abou
January 01, 2025

Instructions
Answer the questions by ticking the correct box. Detailed calculations may be provided on
a separate sheet if needed.

Question 1
Consider the regression function m(x) = E(Y | X = x). A spline estimator m̂(x) with knots
t1 , . . . , tk is constructed using piecewise polynomials. Which statement is true?
(A) Splines require m(x) to be linear outside the knots. □ TRUE □ FALSE
(B) Cubic splines ensure continuity of the second derivative at the knots. □ TRUE
□ FALSE
(C) Natural cubic splines are linear beyond the boundary knots. □ TRUE □ FALSE
(D) Regression splines use a penalty term to control smoothness. □ TRUE □ FALSE

Question 2
The penalized spline estimator minimizes:
X n Z
2 2
(Yi − m(Xi )) + λ (m′′ (x)) dx.
i=1

Which of the following is true about λ?

1
Question 3
Which statement about B-splines is correct?

(A) B-splines use the truncated power basis for numerical stability. □ TRUE □ FALSE

(B) Each B-spline basis function has global support. □ TRUE □ FALSE

(C) B-splines are defined recursively via the Cox-de Boor formula. □ TRUE □ FALSE

(D) The number of B-spline basis functions equals the number of knots. □ TRUE
□ FALSE

Question 4
In a regression spline model with k knots and degree r, the basis dimension is:

(A) k + r □ TRUE □ FALSE

(B) k + r + 1 □ TRUE □ FALSE

(C) k × r □ TRUE □ FALSE

(D) k − r □ TRUE □ FALSE

Question 5
Which statement about smoothing splines is true?

(A) They use all data points as knots. □ TRUE □ FALSE

(B) They minimize the residual sum of squares without regularization. □ TRUE
□ FALSE

(C) The effective degrees of freedom equal the number of knots. □ TRUE □ FALSE

(D) They are equivalent to penalized splines with a truncated power basis. □ TRUE
□ FALSE

Question 6
Consider the mixed model representation of penalized splines:

Y = Xβ + Zu + ϵ, u ∼ N (0, σu2 I), ϵ ∼ N (0, σϵ2 I).

Which term corresponds to the smoothing parameter λ?

2
(A) λ = σu2 □ TRUE □ FALSE

(B) λ = σϵ2 /σu2 □ TRUE □ FALSE

(C) λ = σu2 /σϵ2 □ TRUE □ FALSE

(D) λ = σϵ2 □ TRUE □ FALSE

Question 7
For a cubic smoothing spline, the effective degrees of freedom (df λ ) is defined as:

(A) The number of knots. □ TRUE □ FALSE

(B) The trace of the smoother matrix Sλ . □ TRUE □ FALSE

(C) The rank of the design matrix. □ TRUE □ FALSE

(D) The sum of squared residuals. □ TRUE □ FALSE

Question 8
Which method avoids knot selection by using all data points as knots?

(A) Regression splines □ TRUE □ FALSE

(B) Penalized splines □ TRUE □ FALSE

(C) Smoothing splines □ TRUE □ FALSE

(D) B-splines □ TRUE □ FALSE

Question 9
Generalized cross-validation (GCV) is used to:

(A) Estimate the residual variance. □ TRUE □ FALSE

(B) Select the optimal number of knots. □ TRUE □ FALSE

(C) Choose the smoothing parameter λ. □ TRUE □ FALSE

(D) Determine the polynomial degree. □ TRUE □ FALSE

3
Question 10
In a natural cubic spline, the function is constrained to be:

(A) Linear beyond the boundary knots. □ TRUE □ FALSE

(B) Quadratic beyond the boundary knots. □ TRUE □ FALSE

(C) A degree-3 polynomial within each interval. □ TRUE □ FALSE

(D) Twice differentiable at all knots. □ TRUE □ FALSE

4
Answer Key
1. B (TRUE), C (TRUE)
- Cubic splines ensure continuity of the second derivative (Page 21).
- Natural cubic splines are linear beyond boundary knots (Page 23).

2. C (TRUE), D (TRUE)
- λ controls bias-variance trade-off (Page 29).
- λ is selected via GCV (Page 36).

3. C (TRUE)
- B-splines use the Cox-de Boor recursion (Page 44).

4. B (TRUE)
- Basis dimension: k + r + 1 (Page 40).

5. A (TRUE)
- Smoothing splines use all data as knots (Page 29).

6. B (TRUE)
- λ = σϵ2 /σu2 (Page 65).

7. B (TRUE)
- df λ = trace(Sλ ) (Page 34).

8. C (TRUE)
- Smoothing splines avoid knot selection (Page 29).

9. C (TRUE)
- GCV selects λ (Page 36).

10. A (TRUE)
- Natural cubic splines are linear beyond boundary knots (Page 23).