It could be argued that data are never independent (or iid) other than as an artefact of sampling procedures. People rarely act independently of other people and a particularly apt example of this is network data, where we observe ties among social units. These objects, and dynamic networks in particular, are typically highly complex and rarely amenable to standard statistical analysis. Yet we want to draw meaningful and consistent conclusions from these data. Snijders introduced a novel approach to analysing longitudinal social networks that was inspired by agent-based modelling and simulations of social processes. A set of stochastic rules are set up that guides actors in a network when they form and break off ties to other actors, allowing for modelling complex outcomes yet has statistical parameters that are estimable from data. Simulation is not only an inferential tool, all statistical models may be thought of as simulation models for social processes. Using the building blocks of the stochastic actor-oriented model we illustrate this principle by defining and estimating a model for moves on the housing market that is couched in the form of a social process. This is an exemplar of how data science is what results when we are willing to eschew disciplinary boundaries in tackling the understanding of complex data.
Statistics Data Literacy
onsdag 2 november 2016
Generative and estimable models for longitudinal social networks
From RoyalStatSoc
Johan Koskinen,Manchester University
It could be argued that data are never independent (or iid) other than as an artefact of sampling procedures. People rarely act independently of other people and a particularly apt example of this is network data, where we observe ties among social units. These objects, and dynamic networks in particular, are typically highly complex and rarely amenable to standard statistical analysis. Yet we want to draw meaningful and consistent conclusions from these data. Snijders introduced a novel approach to analysing longitudinal social networks that was inspired by agent-based modelling and simulations of social processes. A set of stochastic rules are set up that guides actors in a network when they form and break off ties to other actors, allowing for modelling complex outcomes yet has statistical parameters that are estimable from data. Simulation is not only an inferential tool, all statistical models may be thought of as simulation models for social processes. Using the building blocks of the stochastic actor-oriented model we illustrate this principle by defining and estimating a model for moves on the housing market that is couched in the form of a social process. This is an exemplar of how data science is what results when we are willing to eschew disciplinary boundaries in tackling the understanding of complex data.
It could be argued that data are never independent (or iid) other than as an artefact of sampling procedures. People rarely act independently of other people and a particularly apt example of this is network data, where we observe ties among social units. These objects, and dynamic networks in particular, are typically highly complex and rarely amenable to standard statistical analysis. Yet we want to draw meaningful and consistent conclusions from these data. Snijders introduced a novel approach to analysing longitudinal social networks that was inspired by agent-based modelling and simulations of social processes. A set of stochastic rules are set up that guides actors in a network when they form and break off ties to other actors, allowing for modelling complex outcomes yet has statistical parameters that are estimable from data. Simulation is not only an inferential tool, all statistical models may be thought of as simulation models for social processes. Using the building blocks of the stochastic actor-oriented model we illustrate this principle by defining and estimating a model for moves on the housing market that is couched in the form of a social process. This is an exemplar of how data science is what results when we are willing to eschew disciplinary boundaries in tackling the understanding of complex data.
