Skip to content

Autologistic Actor Attribute Model

A conditional exponential-family model of binary actor outcomes on an observed social network, with statistics that can express outcome dependence along ties.

Version
v1 · 2026-10-31 · History
Domain-specific #
14465
Domain group
Social Sciences
Origin domain
Sociology & Anthropology
Subdomains
Social Network Analysis, Statistical Network Models → Sociology & Anthropology
Aliases
ALAAM, Autologistic Actor Attribute Models

Core Idea

An autologistic actor attribute model (ALAAM) describes possible patterns of a binary outcome across actors in a social network.[1][2] The network of ties is held fixed; the model assigns probabilities to different 0/1 outcome patterns.[1][2] For example, the outcome could be whether each pupil smokes or whether each music-service user likes jazz.[2]

The model can give weight to an overall tendency toward the outcome, actor characteristics, and configurations involving connected actors.[1][2] A fitted result describes an association under the specified network and model terms.[2] Cross-sectional data alone cannot establish that one actor caused another to change.[2]

How would you explain it like I'm…

Friend Map Patterns

Imagine a map that shows who is friends with whom in a class. Some kids like jazz and some don't. This tool keeps the friend map the same and figures out how likely different patterns of jazz-liking are, like whether friends tend to match. Finding that friends match doesn't prove one friend made the other like jazz.

Fixed Friendships, Changing Traits

Scientists who study groups of people sometimes draw a map of who is connected to whom, like friendships. An Autologistic Actor Attribute Model keeps that map fixed and looks at a yes-or-no trait each person has, like smoking or liking jazz. It works out how likely different patterns of that trait are across the whole map, using things like how common the trait is, facts about each person, and whether connected people tend to share it. It can show that friends share a trait, but by itself it can't prove one friend caused the other to change.

Fixed-Network Trait Model

An Autologistic Actor Attribute Model (ALAAM) is a statistical model for social networks. It takes the network of ties as fixed and treats a yes/no outcome for each person (say, whether they smoke) as the random thing to be explained. Contrast this with an exponential random graph model, which does the reverse: it treats the ties themselves as random. An ALAAM can include terms for how common the outcome is overall, for individual characteristics, and for whether connected people share the outcome. Its results describe associations given the network and the measured characteristics. A one-time snapshot can't show that a friend caused someone to change behavior. The same method has been used for very different data, from smoking among a small group of girls to liking jazz in large online networks.

 

An autologistic actor attribute model (ALAAM) holds an observed network fixed and assigns a joint probability distribution over possible configurations of a binary actor-level outcome on that network. It is a family of conditional exponential-family models in which weighted statistics capture baseline outcome prevalence, effects of actor covariates, and patterns of the outcome along network ties (e.g., contagion-like configurations). A specific model can drop the tie-dependence term as a nested comparison while still belonging to the family. The key contrast is with exponential random graph models, where ties are the random variable; in ALAAMs the ties are given and actor outcomes vary. Coefficients describe associations conditional on the specified network and covariates. A cross-sectional fit alone does not establish that a neighbor's outcome caused an actor's outcome. The same formal structure has been applied across very different settings, such as smoking among 50 girls in a friendship network and jazz preference among Deezer users in national online networks.

Scope of Application

Use this model family when the question concerns a binary actor outcome on a given social network, including whether connected actors show an outcome together after accounting for chosen covariates.[1] The network is a condition of the outcome model, not the random object being predicted.[3][2] Different datasets can call for different configuration terms and fitting checks.[2]

The bounded entry describes the standard binary-outcome form in the inspected papers.[3][1] The screened claim that ordinal or continuous dependent outcomes belong to standard ALAAM is excluded from this bounded binary-outcome entry because the inspected definitions specify binary outcomes.[3][1] A separately specified model would require its own primary review.

Clarity

Ask what varies in the model. For an ALAAM, possible actor-outcome vectors vary while the ties are fixed.[3][2] For an ordinary exponential random graph model, the ties themselves vary.[3][2] A connected-pair outcome term may be included to examine network dependence, but a nested comparison can omit it without redefining the model family.[1]

Manages Complexity

The model collects many possible actor-outcome assignments into one conditional probability family.[1][2] Statistics summarize features of each assignment, and coefficients determine how strongly those features contribute to its relative probability.[1][2] This lets an analyst compare a tie-outcome pattern with baseline prevalence and other included features. The result remains conditional on how the network, outcome and terms were specified.

Richer specifications can be harder to estimate.[2] In the Deezer application, the authors removed some triangle terms to obtain converged models for two national networks.[2]

Abstract Reasoning

Hold the social ties fixed and consider alternative 0/1 assignments to the actors.[1] For each assignment, count the chosen configurations, weight those counts by model coefficients, and normalize over the possible assignments to obtain a conditional probability.[1][2] A comparison between a null fit and a fit with a tie-outcome term tests a specified association; it is not a reconstruction of when or why the actors acquired the outcome.[1][2]

Knowledge Transfer

The same modeling orientation can be applied to different social networks and outcomes. One paper illustrates smoking in a 50-girl friendship-network excerpt and separately analyzes jazz preference in online friendship networks.[2] The actor, tie, outcome and term roles transfer; an estimated coefficient does not transfer unchanged between the cases.

Relations

It specifies a parameterized family of probability laws over a declared binary actor-outcome space.[1] This makes it a strict kind of the live Statistical Model. It also presupposes Social Networks as its actor-and-tie carrier. The carrier can exist without this model, and the model is not itself a social network. Random Graph instead makes the graph structure random.[3][2] Contagion is a possible phenomenon under investigation, not an automatic conclusion from a fitted association.[2]

Examples

School friendship and smoking. A paper fits an ALAAM to an illustrative public excerpt of 50 girls.[2] Friendship nominations are the fixed ties, and smoking is the binary outcome.[2] The reported fit includes positive smoking co-occurrence along ties and an alcohol-consumption association, with goodness-of-fit checks.[2] Mapped back: pupils → actors; friendships → fixed ties; smoking → 0/1 outcome; connected smokers → a counted configuration.[2] The reported association is not proof that a friend caused smoking.[2]

Online music preference. The same paper fits jazz preference on Deezer friendship networks in Croatia, Hungary and Romania.[2] Its tie-outcome estimate is positive and significant in Hungary and Romania but not Croatia; terms were adjusted for convergence in two networks.[2] Mapped back: users → actors; online friendships → fixed ties; liking jazz → 0/1 outcome; configuration statistics → inputs to the conditional law.[2] Different estimates show why a fitted effect should not be copied from one network to another.

Relationships to Other Abstractions

Local relationship map for Autologistic Actor Attribute ModelParents appear above the current abstraction, mutual partners to the right, and children below. Node labels state whether each abstraction is prime or domain-specific; colors identify relation types.Autologistic ActorAttribute ModelDOMAINDomain-specific abstraction: Social networks — presupposesSocial networksDOMAINDomain-specific abstraction: Statistical Model — is a kind ofStatisticalModelDOMAIN

Current abstraction Autologistic Actor Attribute Model Domain-specific

Parents (2) — outgoing typed relations

  • Autologistic Actor Attribute Model is a kind of Statistical Model Domain-specific

    A conditional exponential-family statistical model for binary actor outcomes on a fixed social network.

  • Autologistic Actor Attribute Model presupposes Social networks Domain-specific

    The model requires a social network of actors and ties as its fixed structural carrier.

Hierarchy paths (5) — routes to 4 parentless roots

Neighborhood in Abstraction Space

Autologistic Actor Attribute Model sits in a sparse region of the domain-specific corpus (82nd percentile for distinctiveness): few abstractions share its structure, so a faithful description tends to retrieve it precisely.

Family — Unclustered & Miscellaneous (2835 abstractions)

Nearest neighbors

Computed from structural-signature embeddings · 2026-10-31

Not to Be Confused With

  • A tie-formation model: that predicts edges; an ALAAM conditions on the observed ties.[3][2]

  • A software package or estimator: MPNet and ALAAMEE are implementations; they do not define the model family.[2]

  • Causal proof of social influence: model-conditional co-occurrence does not establish transmission from a single cross-section.[2]

References

[1] Andrew Parker, Francesca Pallotti and Alessandro Lomi, “New Network Models for the Analysis of Social Contagion in Organizations: An Introduction to Autologistic Actor Attribute Models,” Organizational Research Methods 25, no. 3 (2022): 513–540, publisher PDF, especially pp. 517–518, equation 1 and the nested comparison, and pp. 522–530 for the advice-network illustration. registry ↩a ↩b ↩c ↩d ↩e ↩f ↩g ↩h ↩i ↩j ↩k ↩l ↩m

[2] Alex Stivala, Peng Wang and Alessandro Lomi, “ALAAMEE: Open-source software for fitting autologistic actor attribute models,” PLOS Complex Systems 1, no. 4 (2024): e0000021, publisher PDF, especially pp. 2–3, equation 1; pp. 18–20, Figs. 6–8; and pp. 20–23, Table 4. registry ↩a ↩b ↩c ↩d ↩e ↩f ↩g ↩h ↩i ↩j ↩k ↩l ↩m ↩n ↩o ↩p ↩q ↩r ↩s ↩t ↩u ↩v ↩w ↩x ↩y ↩z ↩27 ↩28 ↩29 ↩30

[3] Galina Daraganova and Garry Robins, “Autologistic Actor Attribute Models,” in Exponential Random Graph Models for Social Networks: Theory, Methods, and Applications (Cambridge University Press, online 2013), chapter 9, pp. 102–114, publisher chapter PDF. The accessible chapter summary and opening section support fixed ties, binary outcomes and the cross-sectional model orientation; other chapter claims require direct page inspection. registry ↩a ↩b ↩c ↩d ↩e ↩f ↩g