Statistical Methods
Check State Probabilities Across Repeated Model Cycles
Define the task
Use a transition scenario to audit bookkeeping in a model with named states and a fixed cycle length. Place origin states in rows and destination states in columns. Each row must contain nonnegative probabilities summing to one. Preserve the same state order in the initial distribution. Before projecting, justify a time-homogeneous Markov assumption: the next-state probabilities are fixed and depend on the present state, not unrecorded history.
Carry out the check
For a row-vector distribution, update p(next)=p(current)P. Repeated updates equal multiplication by the integer matrix power Pᵏ; k=0 leaves the initial distribution unchanged. Check conservation of total probability after each cycle. This does not validate the transition estimates. Changing risks, duration dependence, absorbing states, competing events and parameter uncertainty require separate modeling decisions. Do not read the illustrative tool as an individualized clinical forecast.
Worked hypothetical example
Hypothetical example: start at [1,0] with transition rows [0.8,0.2] and [0.1,0.9]. One step gives [0.8,0.2]. Two steps give [0.8×0.8+0.2×0.1, 0.8×0.2+0.2×0.9]=[0.66,0.34]. Both totals equal one. Report the cycle length alongside the number of steps: two monthly transitions and two annual transitions are not interchangeable. Keep scenario assumptions with the output.
Method checked 22 November 2026 against Primary method documentation: numpy.linalg.matrix_power.