Deriving Transit-Chain Mean and Variance

What Determines a Transit Chain’s Completion Time?

A transit chain represents a delay as sequential movement through a specified number of stages. Each stage contributes time to the total, so the completion time depends both on how long a typical stage lasts and on how variable its duration is. Deriving these quantities makes the model’s delay implications explicit: a chain does not impose one exact travel time, but a distribution of possible completion times.

This tutorial considers a chain of \(m\) stages, where \(m\) is a positive integer. Each stage has the same constant transition rate \(\lambda\), measured in inverse time. For an individual item, let \(W_i\) be the time spent in stage \(i\), and let \(T\) be the total time from entry into the first stage until exit from the last.

Definition: An equal-rate exponential transit chain has stage times \(W_1,\ldots,W_m\) that are independent and identically distributed as exponential random variables with rate \(\lambda>0\). Thus, for \(w\geq 0\), each stage time has density \(f_W(w)=\lambda e^{-\lambda w}\), mean \(1/\lambda\), and variance \(1/\lambda^2\). The completion time is \(T=\sum_{i=1}^{m}W_i\).

The assumptions are substantive. A stage transition occurs at a constant rate while the item remains in that stage; the same rate applies to every stage; transitions occur in sequence without skipping or reversing; and stage waiting times are independent. The formulas below describe the completion time under these assumptions. They do not, by themselves, establish that a biological process follows this mechanism.

In a deterministic transit-compartment ODE model, the state variables represent amounts or fractions distributed across compartments. The predicted output over time can be interpreted using the completion-time distribution associated with the sequential exponential stages. That interpretation does not require treating the entire ODE solution as a random trajectory.

Deriving the Mean and Variance

For a single exponential stage with rate \(\lambda\), the moment-generating function is \(M_W(s)=\lambda/(\lambda-s)\), for \(s<\lambda\). Independence means that the moment-generating function of a sum is the product of the stage moment-generating functions. Therefore, the completion-time moment-generating function and its logarithm are

M_T(s)=\left(\frac{\lambda}{\lambda-s}\right)^m, \qquad K_T(s)=\log M_T(s)=-m\log\left(1-\frac{s}{\lambda}\right), \quad s<\lambda.

The first derivative of a log moment-generating function at zero is the mean. Its second derivative at zero is the variance: differentiating \(\log M_T(s)\) twice gives \(M_T''(s)/M_T(s)-[M_T'(s)/M_T(s)]^2\), which at zero is \(E[T^2]-(E[T])^2\). Here the derivatives are

K_T'(s)=\frac{m}{\lambda-s}, \qquad K_T''(s)=\frac{m}{(\lambda-s)^2}.

Evaluating at \(s=0\) yields the central result:

E[T]=\frac{m}{\lambda}, \qquad \operatorname{Var}(T)=\frac{m}{\lambda^2}.

The same result follows directly from sums of independent stage times: expectations add, and the variance of a sum of independent random variables is the sum of their variances. The moment-generating-function derivation also makes clear why the identical-rate and independence assumptions matter.

The standard deviation is \(\sqrt{m}/\lambda\), in units of time. Dividing it by the mean gives the coefficient of variation:

\mathrm{CV}(T) =\frac{\sqrt{m}/\lambda}{m/\lambda} =\frac{1}{\sqrt{m}}.

Thus, for equal-rate stages, the relative dispersion depends on the number of stages, not on the value of the rate. The rate sets the time scale: reducing \(\lambda\) lengthens both the mean and standard deviation. Increasing \(m\) at a fixed rate also increases the mean, while the coefficient of variation decreases.

Worked Example: Four Stages at a Common Rate

Suppose a hypothetical model has \(m=4\) sequential stages, each with rate \(\lambda=0.5\ \text{h}^{-1}\). Assume that the stage times are independent and exponential, as specified above. Find the completion-time mean, variance, standard deviation, and coefficient of variation.

The mean is

E[T]=\frac{m}{\lambda} =\frac{4}{0.5\ \text{h}^{-1}} =8\ \text{h}.

The variance is

\operatorname{Var}(T)=\frac{m}{\lambda^2} =\frac{4}{(0.5\ \text{h}^{-1})^2} =\frac{4}{0.25\ \text{h}^{-2}} =16\ \text{h}^2.

Therefore the standard deviation is \(\sqrt{16\ \text{h}^2}=4\ \text{h}\), and the coefficient of variation is \(4\ \text{h}/8\ \text{h}=0.5000\). As a check, each stage has mean \(1/0.5=2\ \text{h}\) and variance \(1/(0.5)^2=4\ \text{h}^2\). Four independent stages therefore have mean \(4(2)=8\ \text{h}\) and variance \(4(4)=16\ \text{h}^2\).

Worked Example: A Specified Stage Mean

In a hypothetical five-stage chain, each stage has a mean residence time of \(1.5\ \text{h}\). Determine the stage rate and the completion-time summaries. The assumptions of equal rates, exponential waiting times, and independence are taken to hold.

For an exponential stage, the mean is \(1/\lambda\). Thus,

\lambda=\frac{1}{1.5\ \text{h}} =\frac{2}{3}\ \text{h}^{-1}.

The chain mean is \(5(1.5\ \text{h})=7.5\ \text{h}\). Each stage variance is \((1.5\ \text{h})^2=2.25\ \text{h}^2\), so the chain variance is \(5(2.25)=11.25\ \text{h}^2\). Its standard deviation is \(\sqrt{11.25}\ \text{h}\approx 3.354\ \text{h}\). The coefficient of variation is \(\sqrt{11.25}/7.5=1/\sqrt{5}\approx 0.4472\), which is dimensionless.

Worked Example: Unequal Stage Rates

Consider a hypothetical two-stage sequence in which the first stage has rate \(0.5\ \text{h}^{-1}\), the second has rate \(1\ \text{h}^{-1}\), and the two stage times are independent and exponential. This is sequential, but it is not an equal-rate chain.

The first-stage mean and variance are \(1/0.5=2\ \text{h}\) and \(1/(0.5)^2=4\ \text{h}^2\). The second-stage mean and variance are \(1/1=1\ \text{h}\) and \(1/1^2=1\ \text{h}^2\). Since the stages are independent, the completion-time mean is \(2+1=3\ \text{h}\), and its variance is \(4+1=5\ \text{h}^2\). The standard deviation is \(\sqrt{5}\ \text{h}\approx2.236\ \text{h}\).

Using the equal-rate formula with \(m=2\) and one arbitrarily chosen rate would give incorrect results. For independent exponential stages with distinct rates \(\lambda_i\), the corresponding sums are \(E[T]=\sum_i 1/\lambda_i\) and \(\operatorname{Var}(T)=\sum_i 1/\lambda_i^2\); they reduce to the equal-rate formulas only when all \(\lambda_i=\lambda\).

How to Check the Model’s Implications

The mean and variance are useful summaries, but they should be checked against the model structure and its intended interpretation. In an application, examine whether stages represent distinct sequential processes, whether a constant transition rate is plausible within each stage, and whether a common rate across stages is defensible. Independence is also an assumption: if stage durations share an unmodeled source of variation, their dependence can change the total variance.

For any stage times, the general variance identity is

\operatorname{Var}\left(\sum_{i=1}^{m}W_i\right) = \sum_{i=1}^{m}\operatorname{Var}(W_i) + 2\sum_{i<j}\operatorname{Cov}(W_i,W_j).

The covariance terms vanish under independence. Positive dependence increases the total variance relative to the independent case with the same stage variances; negative dependence decreases it. Consequently, simply adding stage variances is not justified when stage times are dependent.

Numerical checks can help catch implementation errors. For a simulated sample of completion times, the sample mean and variance should approach \(m/\lambda\) and \(m/\lambda^2\) as the number of simulated chains grows, subject to Monte Carlo variability. In an ODE implementation, test the transition structure, units, and limiting cases: a rate has units of inverse time, while the calculated mean has units of time and variance has units of time squared. A discrepancy in units or a failure to approach the predicted summaries under a controlled model setup signals a problem to investigate, not evidence that the theoretical formulas have changed.

Rate versus mean. A larger rate means faster stage transitions and a shorter stage mean, \(1/\lambda\). Confusing rate with mean reverses the direction of the effect.
Variance versus standard deviation. The variance is \(m/\lambda^2\) and has squared-time units; the standard deviation is \(\sqrt{m}/\lambda\) and has time units.
Independence. Adding stage variances without checking dependence omits covariance terms.
Equal rates. The formula \(m/\lambda^2\) presumes one common rate. With unequal rates, calculate the stage-specific contributions instead.
Deterministic delay. A chain with finite \(m\) has nonzero variance; it is not an exact fixed delay.

How the Chain Compares With Other Delay Descriptions

A fixed delay assigns every item the same completion time \(\tau\). Its completion-time variance is zero. An equal-rate transit chain instead produces a spread of completion times, with mean \(m/\lambda\) and variance \(m/\lambda^2\). If the mean is held at a value \(\tau\) by setting \(\lambda=m/\tau\), these become

E[T]=\tau, \qquad \operatorname{Var}(T)=\frac{\tau^2}{m}.

This relationship describes the dispersion implied by the chain under a matched mean; it is not, on its own, a rule for choosing \(m\). A general distributed delay can represent shapes that do not arise from equal-rate exponential stages. Unequal-rate sequential stages are another alternative, with variance determined by their individual rates when independent. These models may be more appropriate when the biological mechanism or observed delay pattern does not support identical exponential stages.

A transit chain is therefore a useful, interpretable representation when its assumptions are reasonable, but its summaries are consequences of those assumptions rather than universal properties of delays. Its simple mean and variance do not establish that the full delay distribution is suitable for a given QSP application.

Key takeaway: For \(m\) independent, sequential exponential stages with common rate \(\lambda\), the completion-time mean is \(m/\lambda\) and its variance is \(m/\lambda^2\). Check the equal-rate, exponential, and independence assumptions before applying these formulas.

Check Your Understanding

Answer each question, then compare your reasoning with the solutions below.

A hypothetical chain has six independent, equal-rate stages with \(\lambda=0.3\ \text{h}^{-1}\). What are its completion-time mean, variance, standard deviation, and coefficient of variation?
A four-stage equal-rate chain has a mean completion time of \(12\ \text{h}\). What is the stage rate and what are the completion-time variance and standard deviation?
Two independent exponential stages each have rate \(0.2\ \text{min}^{-1}\). Find the total mean and variance, with units.
Two independent exponential stages have rates \(2\ \text{day}^{-1}\) and \(0.5\ \text{day}^{-1}\). Find the completion-time mean and variance. Explain why the equal-rate formula does not apply.
If stage durations are dependent, what term must be considered when calculating the variance of their sum?

Solutions. (1) The mean is \(6/0.3=20\ \text{h}\). The variance is \(6/(0.3)^2=66.67\ \text{h}^2\), the standard deviation is \(\sqrt{66.67}\ \text{h}\approx8.165\ \text{h}\), and the coefficient of variation is \(1/\sqrt{6}\approx0.4082\). (2) Since the mean is \(m/\lambda\), \(\lambda=4/(12\ \text{h})=1/3\ \text{h}^{-1}\). The variance is \(12^2/4=36\ \text{h}^2\), and the standard deviation is \(6\ \text{h}\). (3) Each stage has mean \(1/0.2=5\ \text{min}\) and variance \(1/(0.2)^2=25\ \text{min}^2\). The total mean is \(10\ \text{min}\), and the variance is \(50\ \text{min}^2\). (4) The means are \(1/2=0.5\ \text{day}\) and \(1/0.5=2\ \text{days}\); the variances are \(1/2^2=0.25\ \text{day}^2\) and \(1/(0.5)^2=4\ \text{day}^2\). Thus the total mean is \(2.5\ \text{days}\) and the variance is \(4.25\ \text{day}^2\). The rates differ, so there is no common \(\lambda\) for the equal-rate formulas. (5) Include the pairwise covariance terms: the variance of a sum is the sum of the individual variances plus twice the sum of covariances over all distinct stage pairs.

Deriving Transit-Chain Mean and Variance

What You'll Learn