Does shifting the release of a response change how often a linked sequence recurs? To answer the question posed in the timing comparison, connect the assigned schedule to the actual release, recorded encounters and subsequent replies. The same records provide the basis for testing a model and reviewing the decision it informs.
Record Events and Relations
| Record | Fields |
|---|---|
| Event | Unique event identifier, channel, time on a common clock and event type. |
| Link | Earlier item identifier, later item identifier and relation type, such as reply or reference. |
| Encounter | Item identifier, receiving context, reception time and route where observed. |
| Coverage | Channel, type of record collected, and start and end of each interval with known coverage. |
| Assignment | Trial and exchange identifiers, earlier item identifier, assigned schedule, assigned release time and actual response event identifier when available. |
The identifiers connect a result to the records from which it was calculated. Keep the selection rule, bin duration and observation horizon with each derived series. A link records a relation between items. An encounter supplies evidence of receipt and of the route where that information is captured.
Construct the Outcome
For each assigned item, follow the links during the same elapsed interval h from its assigned release. Record whether a new reply to the earlier item receives a further reply within that interval. Count the item once when the sequence occurs. Retain the channels and times of both replies, including links that cross between assigned exchanges. Mark an outcome as unresolved when a gap in coverage leaves it undetermined.
Use the assigned schedule when comparing recurrence fractions, and report actual release and exposure alongside them. Count the same events in aligned bins if the question concerns changes in lag or frequency. The measurement and diagnostics chapter gives the paired spectral estimate. These observations can begin with times and links. Judgements about meaning, accuracy and deceptive intent draw on the communication and its circumstances.
Fit and Simulate a Transition Rule
For this outcome, divide h into equal bins. At the end of each bin record the progress of the specified sequence.
- Stage 0. A new direct reply to the earlier item is still awaited.
- Stage 1. At least one new direct reply has appeared and the sequence is awaiting a reply to one of those replies.
- Stage 2. The two linked steps have occurred within the observation interval.
Let s describe the stage at the start of a bin and the elapsed bin number. A bin can contain both replies, allowing a move from stage 0 to stage 2. Stage 2 remains recorded once reached because the outcome is completion by a specified time. Add information about encounters or channels to s when the proposed explanation depends on it.
Let ns,g,j count transitions from state s under schedule g to next stage j. Include bins with enough coverage to determine both stages. When the total count is positive, use the conditional proportions as a starting estimate.
p̂(j | s, g) = ns,g,j / ∑ℓ ns,g,ℓ
The sum runs over the possible next stages. In a simulation, start at stage 0 for the period after assigned release. Draw the next stage from the probabilities for the current state and schedule, advance the clock and repeat through h. The fraction of simulated paths that reach stage 2 estimates the model’s recurrence probability under that schedule. Show which states have recorded support and which require assumptions.
Fit on earlier exchanges and check the predicted fractions on later exchanges held apart from fitting. Compare them with observed recurrence and with a baseline using ordinary activity and shared events. Vary the supported assumptions about timing and exposure to see which conclusions depend on them. Forecasts about a move between channels require states and transitions that represent those routes. The relation between a general principle and its local mechanism is part of domain practice.
Revise the Model and the Decision
Re-estimating a probability from fresh records updates the analyst’s description. A model of recursive change specifies how realised exchanges change the transition rule within the process. Use the recursive transition rule to test such an account against a rule held fixed. A reply that renews exposure can change the conditions under which a further reply becomes likely. The model must specify how that change enters later transitions.
Record the purpose of the decision, the observed effects, the range of outcomes under the modelled alternatives and the conditions that warrant review. Follow changes in access, recurrence and the work displaced to other channels. Choosing which effects matter requires judgement about the people and relations affected.
Control is always provisional. An analysis becomes part of the field when its use changes access, timing or response. The observations produced under those altered conditions can require changes to the model, the intervention and what is measured next. The synthesis takes up the research questions that follow.