Four distinct objections, each with a corresponding improvement, are as follows.
  • Different patients imply different baseline risks. Raw death rates confound performance with case mix, including urgency, disease severity and comorbidity. Use a prespecified, validated risk-adjusted provider comparison, standardizing to comparable patients or comparing observed deaths with . Check overlap and model calibration rather than assuming adjustment removes all confounding.
  • Small annual volumes produce unstable rankings. A few deaths can move a surgeon far up or down a league table; zero deaths need not indicate zero risk. Display counts and suitably calibrated confidence intervals for binomial proportions, preferably in funnel plots against volume. Pool suitable periods or use a hierarchical Bayesian model to stabilize estimates, while allowing genuine time changes.
  • Many simultaneous comparisons generate chance outliers. With independent null comparisons, the probability of at least one false positive is , and ranks amplify selection of extremes. Use multiple testing adjustments or simultaneous control limits and seek confirmation before interpreting an apparent outlier. Dependence through the national comparator must also be accounted for.
  • The endpoint and unit of attribution can mislead. In-hospital death depends on discharge and transfer policy, and care is delivered by a team; follow-up after discharge is omitted. Use an independently audited fixed-horizon endpoint, such as a prespecified 30-day mortality outcome, linked across hospitals and to death registrations, and report an appropriate team or institutional comparison alongside any surgeon-level estimate. Reliable attribution and consistent inclusion criteria reduce incentives to manipulate coding or case selection.
A confidence interval containing the national rate does not demonstrate equivalence, and a raw ranking does not establish a causal performance difference. The improved presentation should convey precision and uncertainty as well as the estimated outcome.
Use the binomial distribution for independent outcomes in each nonoverlapping batch, and let the CUSUM accumulate the log-likelihood ratio favoring a specified change. The two numbered calculations below treat the batch score and the initial-response design separately.
At the last observed failure time , any individual still in the risk set must either fail there or be censored there or later. If there are no such censored observations, every remaining individual fails, so . Thus the final Kaplan–Meier estimator factor is zero, while the final Nelson–Aalen estimator jump is one:
There is no small-jump approximation at this endpoint. Without ties the final risk set has size one, so the same discrepancy follows from its last single failure.
Put . The Taylor series of the logarithm gives the small-jump comparison of cumulative hazard estimators:
Consequently, if , the two integrated hazard estimates obey
They are close when every event fraction is small. In particular, with distinct single failures and large risk sets, is small. With tied data, large risk sets alone are insufficient if a substantial fraction of the risk set fails together.
At a distinct event time , estimate the conditional probability of surviving that time by . Multiplication of these conditional survival probabilities gives the Kaplan–Meier estimator, and taking minus its logarithm gives an integrated hazard estimate:
The conditional failure count has a binomial likelihood whose maximizing event fraction is . Tied failures therefore enter directly as one count. Even ordering failures arbitrarily, with no intervening censoring, produces the same survivor function factor because
The failure-versus-censoring ordering convention still matters when both have the same recorded time. If , the survivor function becomes zero and the logarithmic integrated hazard is infinite.
Collect distinct observed failure times as , with failures and individuals in the risk set just before . The grouped Nelson–Aalen estimator is
All failures recorded at one time use the same pre-event risk set. When failures and censoring share a recorded time, the usual convention includes those censored at that time in the risk set and processes failures before removing censored individuals. A known different ordering should instead be honored. If ties arise solely through rounding of a continuous-time process, an alternative is to resolve the unobserved ordering: successive failures without intervening censoring would contribute . That is not the grouped jump ; the difference represents a choice about what the recorded ties mean.
Let count observed failures and let be the at-risk process, counting subjects under observation and still alive immediately before . Under independent censoring, a short interval of length has conditional expected failure count . The estimating equation obtained by replacing the count expectation by its observation is . Each distinct failure gives and , so the Nelson–Aalen estimator is
The same increment maximizes the local working likelihood for a Poisson distribution with mean : differentiating gives . The counting-process derivation explains why the risk set, rather than the initial sample size, supplies the denominator.
For and , the exponential distribution has probability density function . Integrating the probability density function gives the survivor function . The hazard function divides instantaneous failure density by the probability of having survived, and the integrated hazard accumulates this rate. Thus
The probability density function is zero for , and the survivor function is one there.
Normalizing the true density gives
a gamma distribution of shape and rate . Its mean is , and the Gamma integral gives
Put . Since , the positive adjustment coefficient satisfies
Both sides are positive in this domain. Squaring, multiplying by , and dividing out the zero root yields
The discriminant is . The negative constant term makes one root positive and the other negative. The polynomial is negative at zero and has value four at one, so the positive root lies in and is not an extraneous solution introduced by squaring. Therefore the adjustment coefficient for shape-one-half gamma claims is
To compare it with the exponential calculation, write . Their difference is
since . Thus the two proposed Lundberg inequality bounds are ordered as
Only is justified for the true claim law. The exponential calculation gives for a different model, not a valid guaranteed bound on the true ruin probability.
This exponential misspecification of shape-one-half gamma claims is optimistic: the true variance is , whereas the assumed exponential variance is . Its tail also decays at rate , more slowly than the assumed rate . Equivalently, for , the true transform is larger than , since . This stronger transform growth makes the true adjustment root smaller. Comparing upper bounds alone does not by itself prove an ordering of the two actual ruin probabilities. At both displayed bounds equal one.
For an exponential distribution with mean , the claim moment-generating function is for . The calculated adjustment coefficient must solve
Subtract one, divide by the nonzero factor , and obtain . Thus
It belongs to for every , as required by the transform domain. The Poisson intensity cancels because the premium rate is specified using the same relative loading and mean claim cost.
In the classical risk model, the surplus with initial capital is
where is a Poisson process of rate , and the nonnegative independent claim sizes, of mean , are independent of that process. Positive relative safety loading makes expected premium income exceed expected claim outflow. With ruin time , the ultimate ruin probability is
The adjustment coefficient is the positive root in the finite domain of the claim moment-generating function satisfying
The zero root is excluded by definition.
Under the stated transform-divergence assumption this positive root exists and is unique. Indeed has , derivative , and strictly positive second derivative on its positive domain. At a finite endpoint the diverging transform makes positive eventually. At an infinite endpoint, choose with ; the bound makes it dominate the linear term. Strict convexity then gives exactly one positive crossing. This is the secant-slope existence criterion for an adjustment coefficient.
The Lundberg inequality is
For a short justification, the compound-Poisson transform and the defining equation make a nonnegative martingale starting at . Stop it at , a bounded stopping time. On its value is greater than one. The optional stopping theorem therefore gives ; let increase to infinity. No expectation identity at an unbounded ruin time is required.
Within the approximation that a factory has either zero or one accident, let be the accident count. The factories' independence gives . Given an accident, its cluster of claims has a Poisson distribution of mean , independently across factories. Its probability generating function is . Thus this is a binomial accident portfolio with Poisson clusters, with
A factory can contribute no claims either because it has no accident or because its accident's Poisson cluster is empty. Hence the no-claim probability differs from the no-accident probability .
For , the accident probability mass function is . For ,
Therefore its Panjer claim-count class parameters are
At the multiplier is zero, and afterwards both successive masses are zero, so the recurrence holds for all . Finally, and reproduce the same initialization .
Use the usual aggregate claims model assumption that the claim count is independent of the sequence of severities. Conditional on , the probability generating function of the sum is the product of the severity generating functions. The empty sum has value zero. Thus the law of total expectation gives the random-sum transform identity
for , with the usual continuous interpretation at .
For a claim count distribution in the Panjer claim-count class, multiply the count recurrence by and sum. With ,
Consequently , or
The chain rule applied to the composition then gives
For , compare the coefficients of . The two products give
Move the term to the left and combine the remaining coefficients, since . The Panjer recursion with zero severities is therefore
For the usual nondegenerate proper Panjer laws, , so . If a degenerate count is represented by redundant constants making this denominator zero, the aggregate is identically zero and must be handled directly rather than dividing by zero.
Because severities are nonnegative, the aggregate is zero exactly when every included severity is zero. Count-severity independence gives the initialization
In particular generally exceeds when zero severities are possible.
The count-severity independence is necessary and is implicit, rather than explicitly stated, in the printed random-sum setup. For a concrete counterexample without it, take independent Bernoulli- variables and put . Then , so , whereas . Independence of the among themselves alone cannot justify the requested formula.
First investigate what predicts dropout among variables actually observed, record reasons where possible, and seek additional outcome follow-up. Dependence on observed history alone can be missing at random; informative dependence on unseen responses or latent heterogeneity requires a missing not at random analysis. Observed data do not ordinarily distinguish MAR from all MNAR explanations.
An explicit selection model for informative longitudinal dropout can augment a joint response model by dropout hazards. Define as above and, among subjects retained through visit , put
Omit the previous-response term at the first visit. A nonzero makes dropout depend on the currently missing response. For monotone observation indicators , let
The appropriate observed-data likelihood is
Thus a patient's dropout pattern contributes information in the model; dropping the factor is not justified under this informative mechanism.
Another option is a shared-parameter model for informative dropout: retain the conditional Bernoulli response model, but let the dropout logit contain the same latent effect, for example . Conditional on the shared effect and observed history the processes can be independent, while marginally dropout remains associated with unseen responses. The joint likelihood must integrate the response factors and dropout factors together over . A response-only mixed-model fit does not generally account for this dependence.
Because or can be weakly identified from the observed outcomes, report sensitivity of the treatment conclusion to plausible informative-dropout assumptions. One can fix a range of these dependence parameters and refit, or use pattern-mixture sensitivity analysis that shifts the imputed post-dropout log odds by a specified amount from the MAR prediction. Present population risks and treatment contrasts across these scenarios with their uncertainty. These are explicit identifying assumptions, not an empirical test proving which missingness mechanism is true. Neither a complete-case analysis nor carrying forward the last response supplies such a justification.
The population-averaged logistic regression describes after averaging over unmeasured individual heterogeneity. Its intercept is a marginal log odds at reference baseline values and time zero, its treatment coefficient is a population conditional-on-baseline log odds ratio, and its time coefficient is the change in that marginal log odds per month. Time zero is an extrapolation here because the recorded visits begin later; centring time at the first visit would give a more interpretable intercept.
In the random-intercept logistic model, the intercept is instead the conditional log odds for a subject with . The treatment and time coefficients apply at fixed ; the subject's actual intercept is . Even with a treatment-independent random intercept, the conditional and marginal coefficients generally differ through noncollapsibility of the odds ratio. Indeed, if , marginalization gives
which is not generally a logistic curve with the same linear coefficients. To quantify random-intercept attenuation of marginal logistic slopes, let . Then
For a nondegenerate finite random intercept this derivative lies strictly between zero and one and usually varies with . Thus the marginal treatment contrast is an integrated, attenuated version of the conditional contrast, and the marginal time slope need not be constant. One should not equate the two sets of coefficients, even in a randomized controlled trial without treatment confounding.
For missing completely at random, an unweighted observed-response GEE with cluster-robust standard errors can consistently fit the marginal mean, and a correctly specified integrated GLMM likelihood is valid as well. Under missing at random depending on previous observed outcomes, ordinary unweighted GEE is generally biased because the remaining responses are selected by informative observed history. Correct inverse-observation-weighted estimating equations for longitudinal dropout recover the marginal target under the sequential MAR and positivity assumptions already stated. Covariate-only observation mechanisms are a simpler special case in which correctly conditioned unweighted mean equations can remain valid.
A correctly specified joint response GLMM, fitted by observed-data likelihood, is valid under ignorable missing at random with distinct parameters, including dependence of missingness on observed response history. This is a likelihood property, not a claim that every mixed model automatically fixes missingness. Under missing not at random, neither ordinary GEE, MAR-based weights nor a response-only GLMM is generally valid; the unseen-outcome dependence requires further modelling and sensitivity assumptions.

Pinned article: Introduction to the OurBigBook Project

Welcome to the OurBigBook Project! Our goal is to create the perfect publishing platform for STEM subjects, and get university-level students to write the best free STEM tutorials ever.
Everyone is welcome to create an account and play with the site: ourbigbook.com/go/register. We belive that students themselves can write amazing tutorials, but teachers are welcome too. You can write about anything you want, it doesn't have to be STEM or even educational. Silly test content is very welcome and you won't be penalized in any way. Just keep it legal!
We have two killer features:
  1. topics: topics group articles by different users with the same title, e.g. here is the topic for the "Fundamental Theorem of Calculus" ourbigbook.com/go/topic/fundamental-theorem-of-calculus
    Articles of different users are sorted by upvote within each article page. This feature is a bit like:
    • a Wikipedia where each user can have their own version of each article
    • a Q&A website like Stack Overflow, where multiple people can give their views on a given topic, and the best ones are sorted by upvote. Except you don't need to wait for someone to ask first, and any topic goes, no matter how narrow or broad
    This feature makes it possible for readers to find better explanations of any topic created by other writers. And it allows writers to create an explanation in a place that readers might actually find it.
    Figure 1.
    Screenshot of the "Derivative" topic page
    . View it live at: ourbigbook.com/go/topic/derivative
  2. local editing: you can store all your personal knowledge base content locally in a plaintext markup format that can be edited locally and published either:
    This way you can be sure that even if OurBigBook.com were to go down one day (which we have no plans to do as it is quite cheap to host!), your content will still be perfectly readable as a static site.
    Figure 5. . You can also edit articles on the Web editor without installing anything locally.
    Video 3.
    Edit locally and publish demo
    . Source. This shows editing OurBigBook Markup and publishing it using the Visual Studio Code extension.
  3. https://raw.githubusercontent.com/ourbigbook/ourbigbook-media/master/feature/x/hilbert-space-arrow.png
  4. Infinitely deep tables of contents:
    Figure 6.
    Dynamic article tree with infinitely deep table of contents
    .
    Descendant pages can also show up as toplevel e.g.: ourbigbook.com/cirosantilli/chordate-subclade
All our software is open source and hosted at: github.com/ourbigbook/ourbigbook
Further documentation can be found at: docs.ourbigbook.com
Feel free to reach our to us for any help or suggestions: docs.ourbigbook.com/#contact