Communication Temporal Analysis
Overview
This Skill preserves communication as a process in time. It locks the clock, sequence,
risk set, temporal aggregation, and observation process before fitting a dynamic model.
Required start
- State whether time is the outcome, exposure, ordering device, moderator, or source of
dependence.
- Define the unit, clock, origin, interval, observation frequency, censoring, and
missingness in references/temporal-protocol.md.
- Separate within-unit change from between-unit differences and description from
causal intervention claims.
Method routing
- time-series and distributed-lag models address serial dependence and delayed relations;
- panel models separate within- and between-unit variation with attrition and clustering;
- event-history/survival models analyze event timing under a defined risk set;
- sequence analysis compares ordered states or actions without reducing them to totals;
- change-point models detect distributional breaks but do not identify their causes;
- interrupted series and event studies route to
communication-causal-inference when
they support intervention claims.
Validation
Check temporal order, stationarity or trend specification, autocorrelation, seasonality,
aggregation sensitivity, lag choice, irregular observation, informative missingness,
censoring, attrition, and multiple-event dependence. Compare results across defensible
time bins and lag windows.
Output shape
Return the temporal contract, method choice, diagnostics, missingness/censoring plan,
lag or sequence definition, sensitivity analyses, and a claim map that distinguishes
change, association, prediction, and causation.
Hard boundaries
- Before/after is not automatically causal.
- A detected change point does not identify the event that caused it.
- Aggregated time bins can create or erase relations.
- Lagged regression alone does not solve confounding or establish direction.
- Panel attrition and platform deletion can redefine the observed population over time.
1---2name: communication-temporal-analysis3description: Design and audit temporal communication research using time series, longitudinal panels, event history, survival, sequence analysis, change points, and distributed lags. Use for 时间序列/纵向面板/事件史/生存分析/序列分析/动态传播. Route causal intervention claims to communication-causal-inference.4---56# Communication Temporal Analysis78## Overview910This Skill preserves communication as a process in time. It locks the clock, sequence,11risk set, temporal aggregation, and observation process before fitting a dynamic model.1213## Required start14151. State whether time is the outcome, exposure, ordering device, moderator, or source of16 dependence.172. Define the unit, clock, origin, interval, observation frequency, censoring, and18 missingness in [references/temporal-protocol.md](references/temporal-protocol.md).193. Separate within-unit change from between-unit differences and description from20 causal intervention claims.2122## Method routing2324- time-series and distributed-lag models address serial dependence and delayed relations;25- panel models separate within- and between-unit variation with attrition and clustering;26- event-history/survival models analyze event timing under a defined risk set;27- sequence analysis compares ordered states or actions without reducing them to totals;28- change-point models detect distributional breaks but do not identify their causes;29- interrupted series and event studies route to `communication-causal-inference` when30 they support intervention claims.3132## Validation3334Check temporal order, stationarity or trend specification, autocorrelation, seasonality,35aggregation sensitivity, lag choice, irregular observation, informative missingness,36censoring, attrition, and multiple-event dependence. Compare results across defensible37time bins and lag windows.3839## Output shape4041Return the temporal contract, method choice, diagnostics, missingness/censoring plan,42lag or sequence definition, sensitivity analyses, and a claim map that distinguishes43change, association, prediction, and causation.4445## Hard boundaries4647- Before/after is not automatically causal.48- A detected change point does not identify the event that caused it.49- Aggregated time bins can create or erase relations.50- Lagged regression alone does not solve confounding or establish direction.51- Panel attrition and platform deletion can redefine the observed population over time.
haochengw372-hash