In Interaction-based economics, sampling specifies which exchanges, markets, agent roles, and institutional settings are observed, how densely they are observed across space and time, and how the resulting subset stands in for the broader domain of strategic interaction. Sampling operates at the level of interaction populations, defining representativeness and the limits of valid inference about mechanisms and strategic behavior.
Ensuring Representative Data
Interaction sampling seeks to ensure that observed exchanges represent the intended population of interactions and institutional environments.
Sampling rules specify which markets, mechanisms, roles, and interaction types are included or excluded. Representativeness is defined relative to an explicit target domain—such as a class of auctions, bargaining protocols, or trading venues. Inference is valid only to the extent that sampled interactions align with this defined interaction space.
Broad Spatial and Temporal Coverage
Interaction sampling requires coverage across relevant venues and time horizons.
Sampling designs specify where interactions are observed—platforms, locations, or institutions—and over what temporal span. Temporal coverage determines whether samples capture transient behavior, steady-state patterns, or adaptive dynamics. Limited spatial or temporal coverage constrains generalization across settings or periods.
Sufficient Resolution and Frequency
Interaction sampling must be sufficiently dense to resolve strategic and institutional dynamics.
Sampling frequency determines the ability to observe sequencing, responses, price formation, congestion, or learning within interactions. Insufficient resolution risks obscuring critical adjustments or coordination effects. Sampling density is therefore matched to the expected speed and granularity of interaction dynamics.
Repetition and Large Sample Size
Interaction sampling relies on repeated observation and adequate sample size to establish reliability.
Sampling plans specify how many exchanges are observed per setting and how many settings or participant groups are included. Larger samples reduce sensitivity to idiosyncratic participants or events and support stable inference about interaction patterns within the target domain.
Sampling Across Different Conditions
Interaction sampling spans variation across relevant strategic and institutional conditions.
Sampling designs specify which parameters—rules, information regimes, participant composition, or enforcement structures—are varied and how interactions are sampled across those conditions. Coverage across conditions ensures that observed behavior reflects the mechanism’s structure rather than a narrow configuration.
Avoiding Bias with Randomization and Stratification
Interaction sampling uses randomization and stratification to limit selection bias.
Random selection of markets, sessions, or time windows reduces convenience and survivorship bias. Stratification ensures proportional representation of relevant roles, settings, or regimes. These mechanisms constrain bias introduced by selective observation of interactions.
Capturing Dynamic and Rare Phenomena
Interaction sampling must account for infrequent, transient, or extreme interaction outcomes.
Sampling designs may include extended observation windows or targeted oversampling to capture shocks, breakdowns, coordination failures, or rare equilibria. Without sufficient depth, such phenomena may be systematically missed, limiting inference about interaction dynamics.