GitHub release (latest by date) GitHub Release Date GitHub issues GitHub All Releases R-CMD-check Codecov test coverage CII Best Practices


The {goldfish} package offers a collection of tools designed for applying statistical models to dynamic network data. It primarily focus on models for relational event data, namely, sequences of interactions between actors or entities within a network, enriched by fine-grained time-stamps information. Relational event data emerge in various domains, such as automatically collected data about interactions in communication and social media research, social science studies using social sensors, and archival network studies that provide in-depth details regarding the timing or sequence of relational actions between nodes.

Currently, the package includes the following models:


For detailed documentation on each model, including usage examples, users are encouraged to consult the package’s vignettes and help files:

Table of Contents


You can install {goldfish} directly from CRAN:


To install the development version from GitHub, use the remotes package:

Or by downloading and install the latest binary releases for all major OSes – Windows, Mac, and Linux – can be found here.

Installing OpenMP on Mac OSX

In some cases, you may get an error that does not allow installation of {goldfish} from source on Mac OSX versions, including under R 4.0.0. The error may relate to compiling the parts of {goldfish} that are written in C++, or whether OpenMP (for parallelisation) can be found.

Many installation woes can be solved by directing R to use Homebrew installed gcc. An updated setting up instructions thanks to @timonelmer are available here.

More details can be found here (Thank you @Knieps for identifying this.). Other links that may be helpful include:

Please share feedback on which of these work and we will update the installation guide accordingly.


Below is a quick-start guide to using the {goldfish} package. The dataset used in this example is an abbreviated version of the MIT Social Evolution data (?Social_Evolution).

The main data objects required for the analysis are the node set(s) defineNodes() and network(s) defineNetwork(). The node set object contains labels and attributes of the actors in the network. In contrast, a network object contains the information of past relational events between actors. By default, defineNetwork() constructs an empty matrix, its dimensions defined by the length of the nodeset(s). Data frames containing event data that modify these data objects can be linked to them using the linkEvents() method.


callNetwork <- defineNetwork(nodes = actors, directed = TRUE) |> # 1
  linkEvents(changeEvent = calls, nodes = actors) # 2

The events data frame, which indicates the time-varying attributes in the node set, contains the following columns:

The events data frame that details the relational events between actors contains the following columns:

Define dependent events

The final step in defining the data objects is to identify the dependent events. Here we would like to model as the dependent variable the calls between individuals. We specify the event data frame and the node set.

callsDependent <- defineDependentEvents(
  events = calls, nodes = actors,
  defaultNetwork = callNetwork

Model specification and estimation

We specify our model using the standard R formula format like:

goldfish_dependent ~ effects(process_state_element)

We can see which effects are currently available and how to specify them here:


Now to estimate this model, we use the ?estimate function.

mod00Rate <- estimate(
  callsDependent ~ indeg + outdeg,
  model = "DyNAM", subModel = "rate"

#> Call:
#> estimate(x = callsDependent ~ indeg + outdeg, model = "DyNAM", 
#>     subModel = "rate")
#> Coefficients:
#>        Estimate Std. Error z-value  Pr(>|z|)    
#> indeg  0.551445   0.066344  8.3119 < 2.2e-16 ***
#> outdeg 0.263784   0.028386  9.2927 < 2.2e-16 ***
#> ---
#> Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
#>   Converged with max abs. score of 2e-05 
#>   Log-Likelihood: -1750.9
#>   AIC:  3505.8 
#>   AICc: 3505.9 
#>   BIC:  3514 
#>   model: "DyNAM" subModel: "rate"

mod00Choice <- estimate(
  callsDependent ~ inertia + recip + trans,
  model = "DyNAM", subModel = "choice"
#> Call:
#> estimate(x = callsDependent ~ inertia + recip + trans, model = "DyNAM", 
#>     subModel = "choice")
#> Coefficients:
#>         Estimate Std. Error z-value  Pr(>|z|)    
#> inertia  5.19690    0.17397 29.8725 < 2.2e-16 ***
#> recip    1.39802    0.17300  8.0812 6.661e-16 ***
#> trans   -0.23036    0.21554 -1.0687    0.2852    
#> ---
#> Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
#>   Converged with max abs. score of 7e-05 
#>   Log-Likelihood: -696.72
#>   AIC:  1399.4 
#>   AICc: 1399.5 
#>   BIC:  1411.7 
#>   model: "DyNAM" subModel: "choice"


This project is a joint collaboration between the Social Networks Lab at ETH Zürich and the Graduate Institute Geneva, and incorporates and supports several sub-projects.


Butts, C. T. 2008. “A Relational Event Framework for Social Action.” Sociological Methodology 38 (1): 155–200.

Hoffman, Marion, Per Block, Timon Elmer, and Christoph Stadtfeld. 2020. “A Model for the Dynamics of Face-to-Face Interactions in Social Groups.” Network Science 8 (S1): S4–25.

Stadtfeld, C., and P. Block. 2017. “Interactions, Actors, and Time: Dynamic Network Actor Models for Relational Events.” Sociological Science 4 (14): 318–52.

Stadtfeld, C., J. Hollway, and P. Block. 2017. “Dynamic Network Actor Models: Investigating Coordination Ties Through Time.” Sociological Methodology 47 (1): 1–40.