Traditional network data collection methods can be generally divided in two modes: self-assessed and observational. Differences (logistical, practical, and theoretical) between these two translate into differences on how the networks can be analysed - they make different assumptions on how individuals connect to each other. The self-assessed mode can be further divided into two additional ways to ask survey-respondents for their connections: their close relationships, and the people they spend time with. This last method stems from the fact that people who might influence our behaviour are not necesarily the people we have more intense relationships with. The observational mode is considered a combination of these two self-assessed modes. We propose a simulation which allows the same group of individuals to connect in different ways. We explore the effect of the three different network representation modes on bayesian social learning outcomes.
Each one of the three methods have advantages depending on the expected outcome from the researchers. Self-assessment of relationships show [THIS IS WHAT RELATIONSHIPS SHOW]. Self-assessment of time shows that [WHAT DOES TIME SHOW]. Observational data collection is based on existing data structures, what is gained in independence from individual bias, can be lost in a more accurate version of type of connection which might affect an individual's actions.
The proposed simulation is designed to analyse the differences between the two different self-assessed data-gathering methods and the external, observational method. The first self-assessed method, in which respondants determine connections based on their close relationships, are defined as relation-based connections; the second one, in which respondents determine their connections based on the people they spend most of their time with, are defined as time-based connections. Each simulated individual has a 'relation' variable, and a 'time' variable, and the respective connections are based on closeness to other individuals on that same scale. In other words, we perceive our relation-based connections with people who are more in line with our emotional spectrum, and we perceive our time-based connections with people who are more in line with our time-spectrum. An external, independent observer will determine that our connections are a mix between our relation-based and our time-based links. We propose a Cobb-Douglass type of function to determine these connections.
We create networks based on proximity on the 'relation', 'time', and 'observable' scales. These are analysed using different methodologies, like link prediction using the Exponential Random Graph Model approach, and different measures of centrality and closure, to answer a specific set of questions, sampled from different well known network experiments. Results suggests that depending on the shape of the 'observation' function, it is the best way of answering the questions because of its more integrated analysis of a network.