Introduction to
Stochastic Actor-oriented Models
From variables to effectsSpecial Needs Assistance
András Vörös
Department of Social Statistics, University of Manchester
SOST71032 Social Network Analysis
* Thanks to Christoph Stadtfeld for many nice figures and examples.
The heart of the SAOM: the “ministep”
 In each step of the simulation of a chain of network
changes, an actor is chosen and prompted to make one
change in his/her outgoing ties
 Which tie to change? – a multinomial choice (as many
options as the number of peers)
 And no change is also an option
 What determines which of the options an actor will
choose?
What determines how actors change ties?
What determines how actors change ties?
 Given the structure of the data, the possible factors:
 network structure around the actor, “Ego”
(~ temporal autocorrelation in the network)
 attributes of Ego
 attributes of others (“Alters”)
 any combination of the above
(e.g. attributes of Alters whom Ego is tied to in the network)
 These factors can affect tie changes by actors
– but HOW exactly?
 E.g. how does my current set of friends affect my
friendship choice in the next moment?
From variables to effects
 Variables we use: the network, actor attributes, etc.
 The role of even the most simple variable in network
change is ambiguous
 How does gender affect whether I like someone (Bob) or not?
From variables to effects
 Variables we use: the network, actor attributes, etc.
 The role of even the most simple variable in network
change is ambiguous
 How does gender affect whether I like someone (Bob) or not?
 Whose gender?
 Mine?
 Bob’s?
 Both?
 The number of my male friends?
 The number of Bob’s female friends?
 The number of male friends we have in common with Bob?
 And so on…
From variables to effects
 Even such a basic thing as gender is not a meaningful
variable to be included in our model as it is
 We need to be more specific about the effect of gender
on network evolution
 gender of Ego, gender of Alter, same gender of Ego and Alter,
etc.
 “Effects” in SIENA models:
 the specific operationalizations of a variable
 i.e. the exact ways in which a variable affects actor decisions
 one variable (network, attribute) can affect actor decisions in
multiple ways – multiple effects may be specified for each var.
 these effects are meaningful for the model
 in SIENA models, we often say “effects” when we talk about
the technical “variables” that are included in the actual model
Please continue with the next topic.