Two streams of replies can rise and fall together because participants encounter one another’s messages, because both follow the same news cycle, or because they share a schedule. Each explanation has different consequences for what happens next. Measurement begins by recording the relations and times that let us distinguish them. The same inquiry applies across the information field, including where disinformation persists.
Record the Relation
Record when an item appeared, what earlier item it refers to, and when a linked response followed. Align the clocks across channels. Where reception is observed, record it separately. Publication to reply is an elapsed lag. Reception to reply describes the interval after an encounter. The observed links give the routes through which responses travel. Evidence of who encountered what gives a basis for testing influence.
Keep gaps in coverage distinct from intervals with no recorded activity. To compare response delays, select earlier items with complete observation through the same follow-up interval. Count how many receive a linked response within that interval and divide by the number of eligible items. Compare the delays and response fractions across similar conditions. Where exposure is part of the question, use records that establish exposure.
Choose a recurring relation that can be recognised across substitutions. A reply followed by a reply to that reply is one observable pattern. Record the links and their order, and specify which changes of participant, wording or channel preserve the pattern under examination. This first pass can proceed through times and recorded connections before interpreting what the replies mean. The choice of sources and observation intervals remains part of the description. See semantics follows frequency.
Estimate Recurrence and Relative Phase
Count the selected events in equal intervals of duration δt. Let X and Y denote the resulting series for two channels. Divide both series into windows of M complete bins with the same boundaries in both channels. Successive windows do not overlap. The Fourier frequency spacing is 1/(Mδt). Choose a bin duration that resolves the activity of interest and a window long enough to observe several cycles of the slowest rhythm under examination. Examine changes in mean activity and lag relationships before averaging across windows. Where those properties change, estimate spectra separately for the periods being compared.
In each window subtract the window mean from each series. Write xn for the resulting values in the first channel and yn for those in the second. Give each bin equal weight in the unnormalised Fourier transform.
DX(f) = ∑n=0M−1 xn e−2π i f n δt
Compute DY in the same way. With the convention in recursive harmonics, estimate the cross-spectrum by averaging the paired Fourier products across comparable windows. This uses the averaging principle of Bartlett (1950).
ŜXY(f) = (δt/M) average[DX*(f) DY(f)]
The star denotes complex conjugation, which reverses the phase of the first factor. The angle of the averaged cross-spectrum estimates the phase of Y relative to X wherever that average is nonzero. Use the same formula with the two factors drawn from one channel to estimate its power spectrum, ŜXX or ŜYY. With positive power in both channels, calculate the magnitude-squared coherence.
γ2(f) = |ŜXY(f)|2 / [ŜXX(f) ŜYY(f)]
The ratio ranges from zero to one and measures the strength of linear association at that frequency. Form it after averaging the products. With a single window it equals one wherever both powers are positive. Compare phase and association across periods and channel pairs, and report uncertainty using a procedure that accounts for dependence in the records.
A Worked Check
Use the delayed copy from [02] to check the direction of the phase estimate. For a four-bin cycle and a delay of one bin, the frequency is 1/(4δt). In windows containing whole cycles, the delayed series has DY = −iDX at that frequency. With nonzero power, the calculation gives a phase of −π/2 and magnitude-squared coherence of one. These are results for the specified delayed copy.
When observed streams exhibit this phase relationship, check the recorded lags and links. A quarter-cycle phase lag also corresponds to delays that differ by whole cycles. A regular external event can give two streams a stable phase relationship. Compare their dependence on that event with the sequence of encounters and responses that would carry influence between them.
Test Competing Explanations
Establish a comparison using records with similar coverage, ordinary activity and exposure to shared events. A baseline prediction can use each channel’s own recent activity, calendar rhythms and recorded common inputs. A second prediction also uses the earlier encounters or responses through which the proposed influence travels. Evaluate both on later windows held apart from those used to select the candidate frequencies and delays. An improvement shows what the linked history adds to prediction. A timing experiment tests whether changing the relation changes the response.
Follow both the time series and the routes that generate it. A rhythm can continue while replies move to different sources or channels. A familiar link pattern can acquire different timing as participants change. Comparing these views shows which relations remain consequential across substitutions. Where coverage is incomplete, mark continuity through the gap as unresolved. Apply the continuity criterion to the evidence about the intervening relations.
Once a sequence is identified, examine what its replies do in context. A rebuttal can return attention to an earlier claim and renew the responses that follow it. Record the evidence for that interpretation alongside the timing and link measurements. Assess accuracy and deceptive intent through evidence about the claim and its circumstances.
Observe Change in the Conditions of Recurrence
Use the same definitions and follow-up interval to compare what responses follow a specified history. If earlier exchanges change which sources are used, when replies arrive or which links are renewed, examine how the distribution of subsequent responses changes with them. Test the revised description on further observations. This is the empirical question raised by a transition rule that changes through its own activity.
A proposed frequency domain experiment varies timing while keeping wording fixed, then follows the responses and routes of recurrence. Track the effects across the channels and intervals through which they propagate. Control is always provisional. An intervention changes the conditions under which the next measurement is made. These observations inform relational intervention.
Reference
Bartlett, M.S. (1950) ‘Periodogram analysis and continuous spectra’, Biometrika, 37(1–2), pp. 1–16. doi: 10.1093/biomet/37.1-2.1.