Become an Actuary › Exams › Flashcards › Exam MAS-I › Survival & Reliability Models
Exam MAS-I · ~8-12%

Exam MAS-I: Survival & Reliability Models Flashcards

Survival and reliability analysis for CAS Exam MAS-I: the survival, density, hazard, and cumulative-hazard functions and the identities linking them; exponential, Weibull, and gamma lifetime models; series, parallel, and k-out-of-n system reliability for independent components; mean residual (excess) life; and empirical estimation from right-censored data via the Kaplan-Meier product-limit and Nelson-Aalen estimators - with fully worked hazards, system reliabilities, an MTTF, and a Kaplan-Meier calculation.

43 cards6 topicsFree · fact-checked · LaTeX math
Tap card or press Space to flip
Answer

Every deck is built into the Willys app

All of these decks, including the full practice problem banks, come built into Willys AI Flashcards & Quizzes for iPhone, iPad & Mac, with FSRS + SM-2 spaced repetition, streaks, and exam-date cram mode. 14-day free trial, then $14.99. To load a deck in the app: Settings → Library → Browse, then pick your exam and deck.

More Exam MAS-I decks:

Survival & Reliability Models Practice Credibility Credibility Practice Frequency, Severity & Aggregate Models Frequency, Severity & Aggregate Models Practice Markov Chains Markov Chains Practice Poisson Processes

← All Exam MAS-I decks

Browse all 43 cards as a list
  1. Survival & hazard
    Define the **survival function** $S(t)$ and give its relationship to the cdf $F(t)$ and density $f(t)$.
    For a nonnegative lifetime $T$, $S(t)=\Pr(T>t)=1-F(t)$ is the probability of surviving past time $t$. Properties: $S(0)=1$, $S$ is nonincreasing, and $S(\infty)=0$. The density is $f(t)=-\frac{d}{dt}S(t)=F'(t)$, so $S(t)=\int_t^\infty f(u)\,du$.
  2. Survival & hazard
    Define the **hazard (failure) rate** $h(t)$ and state the three equivalent forms.
    The hazard rate is the instantaneous failure intensity given survival to $t$: $h(t)=\lim_{\Delta t\to 0}\frac{\Pr(t<T\le t+\Delta t\mid T>t)}{\Delta t}$. Equivalent forms: $h(t)=\frac{f(t)}{S(t)}=-\frac{d}{dt}\ln S(t)$. It is a rate (can exceed $1$), not a probability. In reliability it is called the **force of mortality** or **failure rate**.
  3. Survival & hazard
    Define the **cumulative hazard** $H(t)$ and state how it recovers $S(t)$.
    $H(t)=\int_0^t h(u)\,du = -\ln S(t)$. Inverting, $S(t)=e^{-H(t)}$. Since $S$ decreases from $1$ to $0$, $H$ increases from $0$ to $\infty$. This is the master identity of survival analysis: any one of $S$, $f$, $h$, $H$ determines the other three.
  4. Survival & hazard
    Given $S(t)=e^{-0.02t}$ for $t\ge 0$, find $f(t)$, $h(t)$, and $H(t)$.
    Density: $f(t)=-S'(t)=0.02e^{-0.02t}$. Hazard: $h(t)=\frac{f(t)}{S(t)}=\frac{0.02e^{-0.02t}}{e^{-0.02t}}=0.02$ (constant). Cumulative hazard: $H(t)=-\ln S(t)=0.02t$. This is an exponential lifetime with rate $\lambda=0.02$; its constant hazard is the signature of the exponential.
  5. Survival & hazard
    A lifetime has hazard $h(t)=0.01+0.002t$ for $t\ge 0$. Find $S(t)$ and the probability of surviving past $t=10$.
    Cumulative hazard: $H(t)=\int_0^t(0.01+0.002u)\,du = 0.01t + 0.001t^{2}$. Survival: $S(t)=e^{-H(t)}=e^{-(0.01t+0.001t^{2})}$. At $t=10$: $H(10)=0.01(10)+0.001(100)=0.1+0.1=0.2$, so $S(10)=e^{-0.2}\approx 0.8187$. The linearly increasing hazard makes this an aging (wear-out) component.
  6. Survival & hazard
    Recover the survival function from a hazard: given $h(t)=\frac{2t}{1+t^{2}}$ for $t\ge 0$, find $S(t)$.
    $H(t)=\int_0^t \frac{2u}{1+u^{2}}\,du$. With substitution $w=1+u^{2}$, $dw=2u\,du$: $H(t)=\big[\ln(1+u^{2})\big]_0^t=\ln(1+t^{2})$. $S(t)=e^{-H(t)}=e^{-\ln(1+t^{2})}=\frac{1}{1+t^{2}}$. Check: $S(0)=1$ and $S(t)\to 0$, as required.
  7. Lifetime distributions
    State the **exponential** lifetime model: survival, hazard, mean, variance, and its defining property.
    For rate $\lambda>0$: $S(t)=e^{-\lambda t}$, $f(t)=\lambda e^{-\lambda t}$, $h(t)=\lambda$ (constant). Mean $E[T]=\frac{1}{\lambda}$ (the MTTF), variance $\frac{1}{\lambda^{2}}$. Defining property: **memorylessness** — $\Pr(T>s+t\mid T>s)=\Pr(T>t)$. A used component is stochastically as good as new.
  8. Lifetime distributions
    A component has an exponential lifetime with mean $500$ hours. Find the hazard rate and the probability it survives past $800$ hours.
    Mean $=\frac{1}{\lambda}=500$, so the hazard $\lambda=\frac{1}{500}=0.002$ per hour (constant). $S(800)=e^{-\lambda(800)}=e^{-0.002(800)}=e^{-1.6}\approx 0.2019$. So about a $20.2\%$ chance of lasting beyond $800$ hours.
  9. Lifetime distributions
    An exponential component has already survived $300$ hours. Given $\lambda=0.002$, find the probability it survives an **additional** $400$ hours.
    By memorylessness the past survival is irrelevant: $\Pr(T>300+400\mid T>300)=\Pr(T>400)=e^{-\lambda(400)}=e^{-0.002(400)}=e^{-0.8}\approx 0.4493$. The used component behaves like a brand-new one — a unique feature of the constant-hazard exponential.
  10. Lifetime distributions
    Define **MTTF** (mean time to failure) and give two ways to compute it from the survival function.
    MTTF $=E[T]$, the expected lifetime of a component. Direct: $E[T]=\int_0^\infty t\,f(t)\,dt$. Via survival (often easier): for $T\ge 0$, $E[T]=\int_0^\infty S(t)\,dt$. For the exponential, $\int_0^\infty e^{-\lambda t}\,dt=\frac{1}{\lambda}$, recovering MTTF $=1/\lambda$.
  11. Lifetime distributions
    A lifetime has $S(t)=\left(\frac{100}{100+t}\right)^{3}$ for $t\ge 0$ (a Pareto/Lomax form). Find the MTTF.
    Use MTTF $=\int_0^\infty S(t)\,dt = \int_0^\infty 100^{3}(100+t)^{-3}\,dt$. $=100^{3}\left[\frac{(100+t)^{-2}}{-2}\right]_0^\infty = 100^{3}\cdot\frac{(100)^{-2}}{2}=\frac{100^{3}}{2\cdot 100^{2}}=\frac{100}{2}=50$. MTTF $=50$. (General Pareto mean $=\frac{\theta}{\alpha-1}=\frac{100}{3-1}=50$.)
  12. Lifetime distributions
    State the **Weibull** lifetime model and how the shape parameter $\tau$ controls the hazard.
    Weibull survival: $S(t)=e^{-(t/\theta)^{\tau}}$, with scale $\theta>0$ and shape $\tau>0$. Hazard: $h(t)=\frac{\tau}{\theta}\left(\frac{t}{\theta}\right)^{\tau-1}\propto t^{\tau-1}$. **$\tau>1$:** increasing hazard (wear-out / aging). **$\tau<1$:** decreasing hazard (infant mortality / burn-in). **$\tau=1$:** constant hazard — reduces to the exponential.
  13. Lifetime distributions
    A Weibull lifetime has shape $\tau=2$ and scale $\theta=100$. Find the hazard at $t=50$ and the probability of survival past $t=100$.
    Hazard: $h(t)=\frac{\tau}{\theta}\left(\frac{t}{\theta}\right)^{\tau-1}=\frac{2}{100}\left(\frac{t}{100}\right)^{1}=\frac{2t}{100^{2}}=\frac{2t}{10000}$. At $t=50$: $h(50)=\frac{2(50)}{10000}=0.01$ per unit (rising in $t$ since $\tau>1$). Survival: $S(100)=e^{-(100/100)^{2}}=e^{-1}\approx 0.3679$.
  14. Lifetime distributions
    Distinguish **infant mortality**, **random failure**, and **wear-out** on the bathtub hazard curve, and which $\tau$ models each.
    The bathtub curve plots $h(t)$ in three phases: **Infant mortality:** decreasing hazard (early defects burn in) — Weibull $\tau<1$. **Useful life:** roughly constant hazard (random shocks) — exponential, $\tau=1$. **Wear-out:** increasing hazard (aging/fatigue) — Weibull $\tau>1$. Reliability engineering blends these phases to describe a component's whole life.
  15. Reliability & systems
    Define the **reliability function** $R(t)$ and how it relates to the survival function and hazard.
    In reliability theory $R(t)=\Pr(T>t)=S(t)$ — the probability a component is still functioning at time $t$. The two terms are interchangeable. Thus $R(t)=e^{-H(t)}$, the failure rate is $h(t)=-\frac{R'(t)}{R(t)}$, and MTTF $=\int_0^\infty R(t)\,dt$.
  16. Reliability & systems
    State the **series system** reliability formula for $n$ independent components and explain when it fails.
    A series system works **only if every** component works, so it fails when **any** one fails. With independent components of reliability $R_i$, the system reliability is $R_S=\prod_{i=1}^{n} R_i = R_1 R_2 \cdots R_n$. Because each $R_i\le 1$, $R_S$ is never greater than the weakest component — series links degrade reliability.
  17. Series & parallel
    State the **parallel system** reliability formula for $n$ independent components and explain when it fails.
    A parallel (redundant) system fails **only if all** components fail, so it works if **any** one works. The system **un**reliability is the product of component unreliabilities: $R_P = 1 - \prod_{i=1}^{n}(1-R_i)$. Because each $(1-R_i)\le 1$, $R_P$ is never less than the strongest component — redundancy improves reliability.
  18. Series & parallel
    A series system has three independent components with reliabilities $0.95$, $0.90$, and $0.99$. Find the system reliability.
    Series: multiply the component reliabilities. $R_S = 0.95 \times 0.90 \times 0.99$. $0.95\times 0.90 = 0.855$; $0.855\times 0.99 = 0.84645$. $R_S \approx 0.8465$. Note the system is **less** reliable than its weakest part ($0.90$) — the hallmark of a series arrangement.
  19. Series & parallel
    Two independent components, each with reliability $0.80$, are placed in **parallel**. Find the system reliability.
    Parallel: the system fails only if both fail. Each failure probability is $1-0.80=0.20$. $R_P = 1 - (1-0.80)(1-0.80) = 1 - (0.20)(0.20) = 1 - 0.04 = 0.96$. The redundant pair ($0.96$) is far more reliable than a single unit ($0.80$).
  20. Series & parallel
    A system has two components in **series**, each of which is itself a **parallel** pair of units with unit reliability $0.90$. Find the overall reliability.
    First each parallel pair: $R_{\text{pair}}=1-(1-0.90)^{2}=1-0.01=0.99$. Then the two pairs are in series: $R_S = 0.99 \times 0.99 = 0.9801$. Overall reliability $\approx 0.9801$. Redundancy at the unit level lifts each block to $0.99$ before the series multiplication.
  21. Reliability & systems
    Each component of a $3$-component **series** system is exponential with the same rate $\lambda=0.01$. Find the system survival function and its MTTF.
    Series survival: $R_S(t)=\prod_i e^{-\lambda t}=e^{-3\lambda t}=e^{-0.03t}$. The system itself is exponential with combined rate $3\lambda=0.03$ (hazards of independent series components **add**). MTTF $=\frac{1}{3\lambda}=\frac{1}{0.03}\approx 33.33$, one-third the single-component MTTF of $100$.
  22. Reliability & systems
    Why do the **hazard rates add** for independent components in series, and what does this imply for the system MTTF?
    Series survival is $R_S(t)=\prod_i e^{-H_i(t)}=e^{-\sum_i H_i(t)}$, so $H_S(t)=\sum_i H_i(t)$ and therefore $h_S(t)=\sum_i h_i(t)$ — the hazards add. For identical exponential parts each with rate $\lambda$, $h_S=n\lambda$ and the system MTTF $=\frac{1}{n\lambda}$, shrinking as components are added in series.
  23. Reliability & systems
    Define a **$k$-out-of-$n$** system and give the reliability formula for identical independent components.
    A $k$-out-of-$n$ system works if **at least $k$** of its $n$ components work. (Series is the $n$-out-of-$n$ case; parallel is $1$-out-of-$n$.) For i.i.d. components each with reliability $p$, the number working is Binomial$(n,p)$, so $R = \sum_{j=k}^{n}\binom{n}{j} p^{j}(1-p)^{n-j}$.
  24. Reliability & systems
    A $2$-out-of-$3$ system has identical independent components each with reliability $0.90$. Find the system reliability.
    Works if at least $2$ of $3$ work; let $X\sim\text{Bin}(3,0.9)$, need $\Pr(X\ge 2)$. $\Pr(X=2)=\binom{3}{2}(0.9)^{2}(0.1)=3(0.81)(0.1)=0.243$. $\Pr(X=3)=(0.9)^{3}=0.729$. $R=0.243+0.729=0.972$. The $2$-of-$3$ majority system ($0.972$) beats a single unit ($0.90$) without full triple redundancy.
  25. Series & parallel
    A bridge-free system has component A in **series** with a **parallel** block of B and C. Reliabilities: $R_A=0.98$, $R_B=0.85$, $R_C=0.85$. Find the system reliability.
    Parallel block B$\parallel$C: $R_{BC}=1-(1-0.85)(1-0.85)=1-(0.15)^{2}=1-0.0225=0.9775$. Then A in series with that block: $R_S = R_A \times R_{BC}=0.98\times 0.9775$. $=0.95795$. $R_S \approx 0.9580$.
  26. Series & parallel
    Two independent components have exponential lifetimes with rates $\lambda_1=0.01$ and $\lambda_2=0.02$, arranged in **parallel**. Find the probability the system survives past $t=50$.
    Component survivals at $t=50$: $R_1=e^{-0.01(50)}=e^{-0.5}\approx 0.6065$, $R_2=e^{-0.02(50)}=e^{-1}\approx 0.3679$. Parallel: $R_P = 1-(1-R_1)(1-R_2)=1-(0.3935)(0.6321)$. $(0.3935)(0.6321)\approx 0.2487$. $R_P \approx 1-0.2487 = 0.7513$.
  27. Reliability & systems
    Why is a **parallel** system's lifetime distribution generally NOT exponential even when its components are exponential?
    For a parallel pair the system lifetime is the **maximum** of the component lifetimes, $T_S=\max(T_1,T_2)$, with $R_P(t)=1-(1-e^{-\lambda_1 t})(1-e^{-\lambda_2 t})$. Differentiating shows the hazard $h_S(t)$ is **not constant** — it rises from $0$ as the system ages. The memoryless property is lost; redundancy creates an increasing failure rate even from constant-hazard parts.
  28. Mean residual life
    Define the **mean residual life** (mean excess life) $e(t)$ and give its integral formula.
    $e(t)=E[T-t\mid T>t]$ is the expected remaining lifetime given survival to $t$. Formula: $e(t)=\frac{1}{S(t)}\int_t^\infty S(u)\,du$. In loss models this is the **mean excess loss** $e_X(d)=E[X-d\mid X>d]$ above a deductible $d$. At $t=0$ it equals the MTTF: $e(0)=E[T]$.
  29. Mean residual life
    What is the mean residual life $e(t)$ of an **exponential** lifetime, and why?
    For an exponential with rate $\lambda$, $e(t)=\frac{1}{S(t)}\int_t^\infty e^{-\lambda u}\,du = \frac{1}{e^{-\lambda t}}\cdot\frac{e^{-\lambda t}}{\lambda}=\frac{1}{\lambda}$. The mean residual life is **constant** at $\frac{1}{\lambda}=$ MTTF for every $t$, a direct consequence of memorylessness: a survivor's expected remaining life never changes.
  30. Mean residual life
    A lifetime is uniform on $[0,100]$. Find the mean residual life $e(40)$.
    For Uniform$[0,100]$, $S(t)=\frac{100-t}{100}$ for $0\le t\le 100$. $\int_t^{100} S(u)\,du = \int_t^{100}\frac{100-u}{100}\,du = \frac{(100-t)^{2}}{200}$. $e(t)=\frac{1}{S(t)}\cdot\frac{(100-t)^{2}}{200}=\frac{100}{100-t}\cdot\frac{(100-t)^{2}}{200}=\frac{100-t}{2}$. $e(40)=\frac{100-40}{2}=30$. (Expected remaining life of a uniform survivor is half the time left.)
  31. Mean residual life
    How does the **sign of the hazard's trend** relate to whether mean residual life increases or decreases (DFR vs IFR)?
    **Increasing hazard (IFR, e.g. Weibull $\tau>1$):** an aging component's expected remaining life **decreases** with $t$. **Decreasing hazard (DFR, e.g. Weibull $\tau<1$):** mean residual life **increases** with $t$ — survivors that pass the burn-in phase are sturdier. **Constant hazard (exponential):** $e(t)$ is constant. Thus $e'(t)$ and $h'(t)$ generally have opposite signs.
  32. Mean residual life
    A lifetime has $S(t)=e^{-0.05t}$. A maintenance contract pays for the **expected remaining life** of a unit that has survived to $t=20$. What value does it use?
    This is exponential with $\lambda=0.05$, so by memorylessness $e(20)=\frac{1}{\lambda}=\frac{1}{0.05}=20$. The contract values the expected remaining life at $20$ time units — exactly the original MTTF, regardless of the $20$ units already survived.
  33. Mean residual life
    Relate the mean residual life $e(d)$ to the **expected value with a deductible** in loss models.
    Splitting the unconditional mean of losses above $d$: $E[(X-d)_+] = E[X-d\mid X>d]\cdot \Pr(X>d) = e(d)\,S(d)$. So the expected payment per loss with deductible $d$ equals the mean excess loss times the survival probability. Equivalently $e(d)=\frac{E[(X-d)_+]}{S(d)}$ — the per-payment severity above the deductible.
  34. KM & Nelson-Aalen
    What problem do the **Kaplan-Meier** and **Nelson-Aalen** estimators solve, and what is **right censoring**?
    They estimate $S(t)$ (or $H(t)$) **nonparametrically** from lifetime data without assuming a distribution. **Right censoring:** for some subjects we only know the lifetime exceeds their last observed time (study ended, lost to follow-up) — the exact failure time is unknown but bounded below. KM and NA correctly use this partial information through the at-risk set, rather than discarding censored observations.
  35. KM & Nelson-Aalen
    State the **Kaplan-Meier** (product-limit) estimator and define $r_i$ and $s_i$.
    At each distinct **death time** $t_i$, let $r_i$ = number **at risk** (alive and uncensored) just before $t_i$ and $s_i$ = number of **deaths** at $t_i$. $\hat S(t)=\prod_{t_i\le t}\left(1-\frac{s_i}{r_i}\right)$. The estimate is a right-continuous step function, dropping only at observed death times; censored times reduce later $r_i$ but cause no drop.
  36. KM & Nelson-Aalen
    State the **Nelson-Aalen** estimator of the cumulative hazard and how it gives an alternative estimate of $S(t)$.
    $\hat H(t)=\sum_{t_i\le t}\frac{s_i}{r_i}$ — a sum (not a product) of the per-time hazard contributions. An alternative survival estimate is $\tilde S(t)=e^{-\hat H(t)}$. Nelson-Aalen is preferred for estimating $H$ directly and tends to be more stable with small samples; it generally gives slightly **higher** survival estimates than KM.
  37. KM & Nelson-Aalen
    Ten units are on test. Failures occur at $t=2$ ($1$ unit), $t=5$ ($2$ units), and $t=8$ ($1$ unit); no censoring. Compute the Kaplan-Meier estimate $\hat S(8)$.
    Build the risk sets: $t=2$: $r_1=10$, $s_1=1 \Rightarrow 1-\frac{1}{10}=0.9$. $t=5$: $r_2=9$, $s_2=2 \Rightarrow 1-\frac{2}{9}=\frac{7}{9}\approx 0.7778$. $t=8$: $r_3=7$, $s_3=1 \Rightarrow 1-\frac{1}{7}=\frac{6}{7}\approx 0.8571$. $\hat S(8)=0.9\times 0.7778\times 0.8571 \approx 0.6000$.
  38. KM & Nelson-Aalen
    Eight units are tested. Deaths at $t=3,7,9$; a unit is **censored** at $t=5$. Risk counts: $r=8$ at $t=3$, then $r=6$ at $t=7$, $r=5$ at $t=9$ (one death each). Compute $\hat S(9)$.
    Censoring at $t=5$ removes a unit from the risk set without a survival drop. $t=3$: $1-\frac{1}{8}=0.8750$. $t=7$: $1-\frac{1}{6}\approx 0.8333$ (the death at $t=3$ and the censoring at $t=5$ both leave $6$ at risk just before $t=7$). $t=9$: $1-\frac{1}{5}=0.8000$. $\hat S(9)=0.8750\times 0.8333\times 0.8000 \approx 0.5833$.
  39. KM & Nelson-Aalen
    Using the same data — $10$ units, deaths $1$ at $t=2$, $2$ at $t=5$, $1$ at $t=8$, no censoring — compute the Nelson-Aalen $\hat H(8)$ and the implied $\tilde S(8)=e^{-\hat H(8)}$.
    Sum the hazard contributions $\frac{s_i}{r_i}$: $\hat H(8)=\frac{1}{10}+\frac{2}{9}+\frac{1}{7}=0.1000+0.2222+0.1429=0.4651$. $\tilde S(8)=e^{-0.4651}\approx 0.6281$. This exceeds the Kaplan-Meier $\hat S(8)\approx 0.6000$, illustrating that Nelson-Aalen typically gives a slightly higher survival estimate.
  40. KM & Nelson-Aalen
    How does a **censored** observation at time $c$ enter the Kaplan-Meier calculation, and why does it not cause $\hat S$ to drop?
    A unit censored at $c$ stays counted in the at-risk set $r_i$ for all death times $t_i\le c$, then is removed before any later death time. Because no death is recorded at $c$, there is no factor $\left(1-\frac{s_i}{r_i}\right)$ for it — $\hat S$ does not step down there. Its only effect is to shrink the later risk sets, which makes subsequent death-time drops slightly larger. This is how KM extracts information from incomplete (right-censored) lifetimes.
  41. Lifetime distributions
    Define the **maximum likelihood estimator** of an exponential rate from complete lifetime data, and give it for a sample mean of $250$.
    For i.i.d. exponential$(\lambda)$ data $t_1,\dots,t_n$, the log-likelihood is $\ell(\lambda)=n\ln\lambda - \lambda\sum t_i$. Setting $\ell'(\lambda)=\frac{n}{\lambda}-\sum t_i=0$ gives $\hat\lambda=\frac{n}{\sum t_i}=\frac{1}{\bar t}$. With $\bar t=250$: $\hat\lambda=\frac{1}{250}=0.004$, so the estimated MTTF is $\frac{1}{\hat\lambda}=250$.
  42. Lifetime distributions
    A test runs $20$ identical exponential units; total observed run-time across all units is $4{,}000$ hours and $8$ failures occur (Type I censoring). Estimate the failure rate $\hat\lambda$ and MTTF.
    For exponential data with censoring, the MLE of the rate is $\hat\lambda=\frac{\text{number of failures}}{\text{total time on test}}=\frac{8}{4000}=0.002$ per hour. MTTF $=\frac{1}{\hat\lambda}=\frac{1}{0.002}=500$ hours. Only the failure count and aggregate exposure are needed — a standard reliability life-test result.
  43. Reliability & systems
    Compare **series** and **parallel** systems of identical components on reliability and on system MTTF (assume constant-hazard parts).
    **Series ($n$ identical, reliability $p$):** $R_S=p^{n}\le p$; reliability **falls** as parts are added. For exponential parts the MTTF $=\frac{1}{n\lambda}$ — adding parts shortens system life. **Parallel ($n$ identical):** $R_P=1-(1-p)^{n}\ge p$; reliability **rises** toward $1$. For exponential parts the MTTF $=\frac{1}{\lambda}\sum_{j=1}^{n}\frac{1}{j}$ — adding redundancy lengthens system life. Series trades reliability for nothing; parallel buys reliability at the cost of redundant hardware.