Suppose that a given individual in a population has a survival time which is exponential with a hazard rate θ. Each individual’s hazard rate θ is potentially different and is sampled from a Weibull distribution with probability density function :
f(θ) = (α/λ)(θ/λ)^(α−1) exp(−θ/λ)^α, 0 ≤ θ < ∞
Let X be the life length of a randomly chosen individual of this population.
(i) Find the survival function of X.
(ii) Find the hazard rate of X. What is the shape of the hazard rate ?
10 marks · directive: Find
The core of a good answer
S(x) = E[e^(−θx)] = ∫₀^∞ e^(−θx) f(θ) dθ = ∫₀^∞ exp(−λx u^(1/α)) e^(−u) du (the Laplace transform of the Weibull law; for α = 1 it is 1/(1 + λx)); the hazard is h(x) = E[θ e^(−θx)]/E[e^(−θx)] = E[θ | X > x], and since h′(x) = −Var(θ | X > x) < 0 the hazard is strictly decreasing, from λΓ(1 + 1/α) at x = 0 towards 0.
An outline written by upsc.help, not an official UPSC answer; every fact checked by a separate review. How we work
What the question asks
A mixture (frailty) problem. Condition on θ, use P(X > x | θ) = e^(−θx), and average over the Weibull density: that gives the survival function. The hazard is −S′(x)/S(x). The shape question is answered by differentiating the hazard and recognising the result as minus a conditional variance, so the mixture always has a decreasing failure rate whatever the Weibull shape α. Reading of the density: the paper prints exp(−θ/λ)^α with the exponent outside the bracket. Taken literally this is e^(−αθ/λ), and the printed f(θ) then integrates to Γ(α + 1)/α^α, which equals 1 only when α = 1; after normalising it is a gamma density, which gives S(x) = (1 + λx/α)^(−α) and h(x) = αλ/(α + λx), also decreasing. Since the question names the Weibull distribution, the outline uses the standard Weibull density f(θ) = (α/λ)(θ/λ)^(α−1) exp{−(θ/λ)^α}; state the reading used. The paper prints the mark split 4+2+4 = 10 against only two numbered sub-parts, so the split cannot be matched to the parts with certainty.
Answer outline
Introduction
Given: X | θ ~ exponential with rate θ, so P(X > x | θ) = e^(−θx) and the conditional density is θe^(−θx); θ ~ Weibull(shape α, scale λ). X is the unconditional life length.
Step 1: Survival function (law of total probability)
- S(x) = P(X > x) = ∫₀^∞ P(X > x | θ) f(θ) dθ = ∫₀^∞ e^(−θx) (α/λ)(θ/λ)^(α−1) exp{−(θ/λ)^α} dθ, x ≥ 0.
- Put u = (θ/λ)^α, so θ = λu^(1/α) and du = (α/λ)(θ/λ)^(α−1) dθ:
- S(x) = ∫₀^∞ exp(−λx u^(1/α)) e^(−u) du = E[e^(−θx)], the Laplace transform of the Weibull variable θ at x.
- Expanding e^(−θx) and using E(θ^k) = λ^k Γ(1 + k/α): S(x) = Σₖ (−λx)^k Γ(1 + k/α)/k! (convergent for α > 1; for α ≤ 1 use the integral form).
- Closed form for α = 1 (θ exponential with mean λ): S(x) = ∫₀^∞ e^(−θx)(1/λ)e^(−θ/λ) dθ = 1/(1 + λx), a Lomax (Pareto II) law.
Step 2: Density and hazard rate
- g(x) = −S′(x) = ∫₀^∞ θ e^(−θx) f(θ) dθ = E[θ e^(−θx)].
- h(x) = g(x)/S(x) = E[θ e^(−θx)] / E[e^(−θx)].
- Interpretation: the posterior density of θ among survivors to x is e^(−θx) f(θ)/S(x), so h(x) = E[θ | X > x], the mean hazard of those still alive.
- α = 1: h(x) = λ/(1 + λx).
Step 3: Shape of the hazard
- Differentiate the ratio: h′(x) = −E[θ² e^(−θx)]/S(x) + {E[θ e^(−θx)]/S(x)}².
- So h′(x) = −{E(θ² | X > x) − [E(θ | X > x)]²} = −Var(θ | X > x).
- θ is non-degenerate (it has a Weibull density), so the conditional variance is positive and h′(x) < 0 for all x: the hazard is strictly decreasing (DFR).
- h(0) = E(θ) = λΓ(1 + 1/α); h(x) → 0 as x → ∞ because survivors are increasingly those with θ near 0.
- Numerical check (λ = 1): α = 2 gives h(0) = 0.886, h(1) = 0.701, h(5) = 0.335; α = 0.5 gives h(0) = 2.000, h(1) = 0.292; α = 1 gives h(2) = 1/3 = λ/(1 + 2λ). All decreasing (computed by numerical integration in Python).
Conclusion
Each individual has a constant hazard, yet the population hazard of X falls with age: the frail (high θ) die early and the survivors are progressively more robust. This selection effect (h′(x) = −Var(θ | X > x) < 0) makes every non-degenerate mixture of exponentials DFR, so the hazard of X is monotone decreasing regardless of α.
Examples and data to cite
- α = 1 special case: θ exponential with mean λ gives S(x) = 1/(1 + λx) and h(x) = λ/(1 + λx), a Lomax distribution (an exponential compounded with a gamma rate gives a Lomax law).
Facts to remember
- Law of total probability: S(x) = ∫ P(X > x | θ) f(θ) dθ.
- Exponential lifetime: P(X > x | θ) = e^(−θx), constant hazard θ.
- Hazard rate h(x) = g(x)/S(x) = −d/dx log S(x).
- Weibull moments: E(θ^k) = λ^k Γ(1 + k/α).
- A mixture of exponential distributions has a decreasing failure rate, since h′(x) = −Var(θ | X > x).