Back to the main Siena page

SIENA news


Overview of availability of recent versions of packages

The newest "official" version of RSiena (1.2-12) is available from CRAN (May 13, 2018). This has, however, an error in function sienaGroupCreate for non-centered actor covariates; see below. This error is corrected since version 1.2-14 (December 5, 2018). The most recent version can be downloaded from the downloads page and also from R-Forge.




May 21, 2019:

Main changes in RSiena and RSienaTest visible to users are the following.

  • New effects outAct.c, inAct.c, outPop.c, inPop.c, degPlus.c. These are centered versions, which might have better convergence properties than their namesakes without the .c.
  • Effects antiInIso and antiInIso2 implemented also for non-directed networks.
  • Effects outTrunc, outTrunc2, outSqInv, isolateNet, and degPlus got endowment effects.
  • For inPopIntn, outPopIntn, inActIntn, and outActIntn, added 'centered' to the statistic name.
  • plot.sienaGOF has a new parameter fontsize; the dots ... will also accept parameters cex, cex.main, cex.lab, cex.axis, as used more generally for plotting in R.
  • In case of simulation as indicated by x$simOnly=TRUE, the object produced by siena07 now also contains the standard errors of the mean, called estMeans.sem (see the help page for siena07).
  • On various places diagnostic output messages were relabeled to 'message' or 'warning', which implies that they can be turned off by commands such as suppressMessages or suppressWarnings.
Main changes in RSienaTest:
  • For sienaBayes:
    • prevAns now is allowed to have a different specification of random effects than the parameter given as effects;
    • it is not allowed to use prevAns and prevBayes objects simultaneously;
    • some further small algorithm changes (see CHANGELOG).



May 10, 2019: Publication of a paper by Nynke Niezink and co-authors explaining the extension of the Stochastic Actor-oriented Model to continuous behavior variables.

Nynke M. D. Niezink, Tom A. B. Snijders, and Marijtje van Duijn (2019). No longer discrete: Modeling the dynamics of social networks and continuous behavior.
Sociological Methodology, in press.
DOI: http://dx.doi.org/10.1177/0081175019842263

Abstract

The dynamics of individual behavior are related to the dynamics of the social structures in which individuals are embedded. This implies that in order to study social mechanisms such as social selection or peer influence, we need to model the evolution of social networks and the attributes of network actors as interdependent processes. The stochastic actor-oriented model is a statistical approach to study network-attribute coevolution based on longitudinal data. In its standard specification, the coevolving actor attributes are assumed to be measured on an ordinal categorical scale. Continuous variables first need to be discretized to fit into such a modeling framework. This article presents an extension of the stochastic actor-oriented model that does away with this restriction by using a stochastic differential equation to model the evolution of a continuous attribute. We propose a measure for explained variance and give an interpretation of parameter sizes. The proposed method is illustrated by a study of the relationship between friendship, alcohol consumption, and self-esteem among adolescents.

The corresponding extension of RSiena will be made available later this year.



April 25 21, 2019: Publication of the paper about the four/five-parameter model for effects of actor covariates on networks, which earlier was at the ArXiV.

  • Tom A.B. Snijders and Alessandro Lomi (2019). Beyond Homophily: Incorporating Actor Variables in Statistical Network Models. Network Science, 7, 1-19.

    Abstract

    We consider the specification of effects of numerical actor attributes in statistical models for directed social networks. A fundamental mechanism is homophily or assortativity, where actors have a higher likelihood to be tied with others having similar values of the variable under study. But there are other mechanisms that may also play a role in how the attribute values of two actors influence the likelihood of a tie. We discuss three additional mechanisms: aspiration to send ties to others having high values; conformity in the sense of sending more ties to others whose values are close to what may be considered the `social norm'; and sociability, where those having higher values will tend to send more ties generally. These mechanisms may operate jointly, and then their effects will be confounded. We present a specification representing these effects simultaneously by a four-parameter quadratic function of the values of sender and receiver. Greater flexibility can be obtained by a five-parameter extension. We argue that empirical researchers often overlook the possibility that homophily may be confounded with these other mechanisms, and that for actor attributes that have important effects on directed networks, these specifications may provide an improvement. An illustration is given of the dependence of advice ties on academic grades in a network of MBA students, analyzed by the Stochastic Actor-oriented Model.

    Associated with this paper is the new script SelectionTables.r which is available from the scripts page.



  • February 26, 2019:

    Main changes in RSiena and RSienaTest visible to users are the following.

    • New effects maxAAlt, minXAlt (thanks to collaboration with Per Block, Marion Hoffman, Isabel Raabe, and Kieran Mepham) and outIsolate.
    • Effect name of isolate effect (not its shortName) changed to in-isolate; this is more clear.
    • New parameter thetaValues in siena07, allowing simulations in Phase 3 with varying parameters according to matrix input.
    • Wald.RSiena, Multipar.RSiena and score.Test also produce one-sided test statistic if df=1.
    • objects produced by Wald.RSiena, Multipar.RSiena and score.Test now have class sienaTest. This class also has a print method.
    • sienaRI: earlier check for bipartite dependent variables; if so, stop.
    • Argument verbose added to includeEffects and setEffect. (It existed already for includeInteraction.)
  • Main changes in RSienaTest:
    • For sienaBayes:
      • Allow arbitrary number of interactions.
      • check that prevAns is a sienaFit object;
        totally skip initial multigroup estimation if prevAns is given with ((prevAns$n3 >= 500) & (!usePrevOnly)).
    • simpleBayesTest and multipleBayesTest corrected; they were wrong for models with user-defined interactions.
    • glueBayes extended (to allow good use for models with fixed parameters)
      and its check for prior rates relaxed (to allow priorRatesFromData=2 leading to slightly different prior rates.)
    • New default prevBayes$nwarm >= 1 for newProposalFromPrev in sienaBayes.
    • siena.table for sienaBayes results extended with between-groups s.d.
    • Argument verbose added to extract.posteriorMeans.



  • January 28, 2019:



    December 27, 2018:

    • A paper about co-evolution of two networks: friendship and studying together, and their effects on achievement, combining a variety of statistical methods and with a beautiful visualization:

      Christoph Stadtfeld, András Vörös, Timon Elmer, Zsófia Boda, and Isabel J. Raabe (2018). Integration in emerging social networks explains academic failure and success. Proceedings of the National Academy of Sciences, in press.
      DOI: https://doi.org/10.1073/pnas.1811388115

      Abstract (partial).
      Academic success of students has been explained with a variety of individual and socioeconomic factors. Social networks that informally emerge within student communities can have an additional effect on their achievement. We investigate how social networks emerge between previously unacquainted students and how integration in these networks explains academic success. Our study measures multiple important dimensions of social ties between students: their positive interactions, friendships, and studying relations. By using statistical models for dynamic network data, we are able to investigate the processes of social network formation in the cohort. We find that friendship ties informally evolve into studying relationships over the academic year. Studying together with others, in turn, has a strong impact on students' success at the examination.



    December 12, 2018:

    • A paper for visualizing the ministeps in the Stochastic Actor-oriented Model:

      jimi Adams and David Schaefer (2018). Visualizing Stochastic Actor-based Model Microsteps. Socius, "Sociological Research for a Dynamic World", Sage Publications.
      DOI: https://doi.org/10.1177/2378023118816545
      This visualization provides a dynamic representation of the microsteps involved in modeling network and behavior change with a stochastic actor-based model.
      See the script for executing this at the Siena scripts page.



    December 6, 2018:

    Main changes in RSiena and RSienaTest visible to users are the following.

    • In the help page of sienaAlgorithmCreate, and in the manual, the possibility is mentioned that the multiplication factor (parameter mult) can be specified as a vector. This allows the possibility for ML estimation to specify the multiplication factor separately for periods × waves.
  • Main changes in RSienaTest:
    • For sienaBayes:
      • the possibility now exists to specify prior variances for non-varying parameters, which can be meaningful especially for effects of group-level variables;
      • a new, hopefully better, initial value is used, defined as a precision weighted mean of the multi-group MoM estimate and the prior mean;
      • number of subphases for initial global parameter estimation by MoM decreased to 2;
      • prewarming phase introduced before the first improveMH.
    • extract.sienaBayes corrected.



  • October 30, 2018:

    There was an error in sienaGroupCreate for non-centered actor covariates. This now is corrected in version 1.2-13.

    Main changes in RSiena and RSienaTest visible to users are the following.

    • Correct error in sienaGroupCreate for non-centered actor covariates.
    • Correct error in print01Report that occurred for changing dyadic covariates given as lists of sparse matrices.
    • Also get simulated dependent behavior variables for siena07(..., returnDeps=TRUE, ...) for ML estimation (see help page).
    • siena.table corrected for data sets with several dependent variables.
    • For siena08: new parameters which and useBound in plot.sienaMeta; new parameter reportEstimates, allowing to reduce the output produced.
    • updateSpecification: also update randomEffects column.
    • More explanation of sparse matrix input in the help page for sienaDependent.
  • Main changes in RSienaTest:
    • For sienaBayes:
      • new parameters proposalFromPrev, which allows taking proposal distributions from prevBayes object;
      • incidentalBasicRates resuscitated;
      • allow fixed rate parameters, with priorRatesFromData=-1;
      • changed initial values of scale factors for proposal distributions;
      • new parameters target and usePrevOnly;
      • in case prevBayes is used, parameters nrunMHBatches, nSampVarying, nSampConst, and nSampRates in the function call of sienaBayes supersede those in the prevBayes object;
      • correct bug for three-effect interactions;
      • copy parameters modelType, behModelType, MaxDegree, Offset, initML, from parameter algo to the algorithms created within sienaBayes;
      • more extensive checking of smallest eigenvalue of covariance matrix;
      • for priorRatesFromData = 1 or 2, the resulting prior Covariance matrix for priorKappa != 1 was incorrect; this was corrected;
      • reported timing changed to elapsed system time.



  • September, 2018: review of studies using actor-oriented models for studying peer influence on anti-social behaviors.

  • Jelle J. Sijtsema and Siegwart M. Lindenberg (2018). Peer influence in the development of adolescent antisocial behavior: Advances from dynamic social network studies. Developmental Review, in press.
    DOI: https://doi.org/10.1016/j.dr.2018.08.002

    Abstract (second half only)
    The current article presents a review of recent empirical studies that have used dynamic social network analyses to study peer influence effects for different forms of antisocial behavior (i.e., aggression, delinquency, externalizing behavior, weapon carrying) as these forms may be differently affected by peer influence. Studies that lump different kinds of antisocial behavior together as 'externalizing behavior' show mixed results with regard to peer influence. With regard to the development of delinquency and weapon carrying, peer influences were observed in studies that had six month to one-year measurement intervals, but not in those with shorter intervals. With regard to direct forms of aggression, peer influence was only observed in certain contexts and depended on individual antisocial traits.
    What is recommended for further advance in the field of peer influence is to avoid container variables of antisocial behavior (such as 'externalizing behavior'), to pay close attention to the role of status and belonging needs, and to focus more strongly on a detailed examination of the sequential order of peer selection and influence processes and their moderation by individual and contextual conditions.



  • Already on the web in 2016: practical recommendations for data collection for longitudinal network studies in school settings.

  • Lars Leszczensky, Harald Beier, Hanno Kruse and Sebastian Pink (2016). Collecting network panel data in schools: Practical guidance based on the experiences of three german research projects.
    http://www.mzes.uni-mannheim.de/publications/wp/wp-166.pdf

    Abstract
    There is an increasing amount of literature on how to analyze longitudinal data of complete social networks. Guidance on how to collect such data in schools, however, is both scant and desperately needed, because longitudinal social network analysis has high data requirements. Aiming to provide guidance for future data collection of school-based networks, we share our experiences gained in three different projects collecting longitudinal social network data in different samples of German schools. This includes both one large-scale study that relied on a nationally representative sample of schools and two smaller studies that targeted more specific geographical areas. We discuss key decisions researchers have to make before data collection, the definition of school samples, and the selection of schools. We further offer advice on how to improve school participation and students' response, describe the fieldwork, and make recommendations for projects aiming to collect longitudinal social network data.



  • May 31, 2018: meta-analysis of results of actor-oriented models.

    Owen Gallupe,John McLevey, and Sarah Brown (2018). Selection and Influence: A Meta-Analysis of the Association Between Peer and Personal Offending.
    Journal of Quantitative Criminology, in press.
    DOI: https://doi.org/10.1007/s10940-018-9384-y.

    Abstract
    Whether people are affected by the criminal behavior of peers (the "influence" perspective) or simply prefer to associate with others who are similar in their offending (the "selection" perspective) is a long-standing criminological debate. The relatively recent development of stochastic actor-oriented models (SAOMs-also called SIENA models) for longitudinal social network data has allowed for the examination of selection and influence effects in more comprehensive ways than was previously possible. This article reports the results of a systematic review and meta-analysis of studies that use SAOMs to test for peer selection and influence effects.



    May 13, 2018: the new official version 1.2-12 of RSiena is available from CRAN.

    Except for a detail on a help page, it is identical to version 1.2-11.
    Thanks to efforts by Felix Schönenberger, it is now available on all R platforms, including Solaris.



    May 6, 2018: new version 1.2-11 of RSiena and RSienaTest available from the downloads page and the RSiena project at R-Forge.

    Main changes in RSiena and RSienaTest visible to users are the following. For the definition of the effects, consult the manual.

    • New effects gwdspFF and gwdspFB.
    • Effects simEgoInDist2 and simEgoInDist2W for two-mode networks were corrected. Perhaps simEgoDist2 and simEgoDist2W were broken in a previous version; if so, that was now corrected.
    • Added sqrt version for reciAct, obtained for parameter = 2.
    • Allowed the value parameter=NULL for setEffect and includeInteraction, meaning that no change is made for the internal effect parameter. For setEffect this is the new default, implying that when starting the default values from allEffects.csv are used, just like in includeEffects (where no parameter can be given).
    • New auxiliary functions for sienaGOF: Triad Census and mixedTriadCensus (contribution by Christoph Stadtfeld).
    • triad census from igraph added to help page of sienaGOF-auxiliary.
    • improved sienatable output for type="html": added rules=none, frame = void to general options of the html output file; changed - to ; added column for asterisks to have better alignment for estimates.
    • Stop cluster also for a user interrupt in siena07.
    • Extension of help page for siena07 by mentioning functions for accessing simulated networks for ML.
    • In RSienaTest:
      extract.sienaBayes: corrected error that occurred if called with extracted="all" but there are no varying, or no non-varying parameters.



    May 5, 2018: paper 'Statistical Power in Longitudinal Network Studies' by Christoph Stadtfeld, Tom A. B. Snijders, Christian Steglich, and Marijtje van Duijn. Sociological Methods and Research, 2018.
    DOI: https://doi.org/10.1177/0049124118769113.

    Abstract Longitudinal social network studies can easily suffer from insufficient statistical power. Studies that simultaneously investigate change of network ties and change of nodal attributes (selection and influence studies) are particularly at risk because the number of nodal observations is typically much lower than the number of observed tie variables. This article presents a simulation-based procedure to evaluate statistical power of longitudinal social network studies in which stochastic actor-oriented models are to be applied. Two detailed case studies illustrate how statistical power is strongly affected by network size, number of data collection waves, effect sizes, missing data, and participant turnover. These issues should thus be explored in the design phase of longitudinal social network studies.



    March 24, 2018: new version 1.2-10 of RSiena and RSienaTest available from the downloads page and the RSiena project at R-Forge.

    Versions 1.2-9 and 1.2-10 are almost identical.
    Main changes in RSiena and RSienaTest visible to users are the following. For the effects, consult the manual.

    • Added effects avExposure, totExposure, infectDeg, susceptAvCovar, infectCovar for symmetric networks.
    • New effects degAbsDiffX, degPosDiffX, degNegDiffX, degAbsContrX, degPosContrX, degNegContrX.
    • Effects XWX, XWX1, and XWX2 enabled for bipartite networks.
    • Added parameter=2 for FFDeg, BBDeg, FBDeg, FRDeg, BRDeg; the purpose here is to decrease collinearity with the ('single') degree.
    • Corrected altInDist2, totInDist2.
    • Export of last simulated state from ML simulations if returnDeps; meaningful for the imputed/simulated missing tie values.
    • Added endowment and creation effects for maxAlt and minAlt.
    • Added components requestedEffects, theta, and se to siena08 (theta and se are the ML estimates).
    • Error in summary of sienaMeta object corrected.
    • For non-directed networks, the initial model now contains only basic rates and degree effect.
    Changes in RSienaTest:
    • new function extract.posteriorMeans for sienaBayes results.
    • In sienaBayes, restrict check of maximum estimated parameter value after initialization to non-fixed effects.
    • Correct construction of groupwise effects object in sienaBayes so that this will work also when evaluation effects for density and/or reciprocity are not included; and when there are interaction effects for which the main effects are not included.
    • Corrections for print and summary of sienaBayes objects.
      In print.sienaBayesFit, include fixed parameters and give credibility intervals for rate parameters; include variance parameters; allow shorter ThinParameters; print objects returned through partialBayesResult.RData.
    • multipleBayesTest corrected (there was an error for testing 2 or more linear combinations simultaneously) and adapted for cases with fixed parameters; adapted help file text.



    March 21, 2018: Paper at ArXiV about model specification for numerical actor covariates.

  • Tom A.B. Snijders and Alessandro Lomi (2018). Beyond Homophily: Incorporating Actor Variables in Actor-oriented Network Models. arXiv:1803.07172v1.
    Web: https://arxiv.org/abs/1803.07172.

    The paper later was published, see entry for April 25, 2019.



  • February 24, 2018: first publication using sienaBayes.

  • Zsófia Boda (2018). Social Influence on Observed Race. Sociological Science, 5, 29-57.
    DOI: http://dx.doi.org/10.15195/v5.a3.

    Abstract

    This article introduces a novel theoretical approach for understanding racial fluidity, emphasizing the social embeddedness of racial classifications. We propose that social ties affect racial perceptions through within-group micromechanisms, resulting in discrepancies between racial self-identifications and race as classified by others. We demonstrate this empirically on data from 12 Hungarian high school classes with one minority group (the Roma) using stochastic actor-oriented models for the analysis of social network panel data. We find strong evidence for social influence: individuals tend to accept their peers' judgement about another student's racial category; opinions of friends have a larger effect than those of nonfriends. Perceived social position also matters: those well-accepted among majority-race peers are likely to be classified as majority students themselves. We argue that similar analyses in other social contexts shall lead to a better understanding of race and interracial processes.



  • February 24, 2018: error in distance-2 effects.

    The current versions of RSiena and RSienaTest at R-Forge have errors for the distance-2 selection effects such as altDist2 and altInDist2. The current version of RSiena at the Siena downloads page does not have this error.
    If you wonder about whether the version you are using has this error, the answer will be clear, because estimation with these effects will run into errors and not finish seemingly adequately.



    February 16, 2018: more help for selection and influence tables.

    In the manual, the section about selection and influence tables was updated.
    New scripts SelectionTables.r and InfluenceTables.r are available from the scripts page.



    September 11, 2017: new version 1.2-4 of RSiena and RSienaTest available from the downloads page and the RSiena project at R-Forge.

    Main changes in RSiena and RSienaTest visible to users:

    • setEffect and includeEffects work again with the option include = FALSE.



    September 9, 2017: error in version 1.2-3 of RSiena and RSienaTest

    The option include=FALSE does not work in functions setEffect and includeEffects of RSiena and RSienaTest, version 1.2-3. This has been repaired in R-Forge version 1.2-4. The version on CRAN, however, still is 1.2-3 and has this error.
    You can work around this by not using this option, i.e., just creating an effects object and including effects, not excluding them. This works unless you wish to exclude effects that are included by default, such as recip for directed and transTriads for non-directed networks. Another way to work around this is by finding out the row numbers of the effects you wish to exclude in the total effects object (for which you can employ something like [print(myeff,includeOnly=FALSE)]), and setting the include column in those rows to FALSE.
    Still another way is to download the two files given here (in most browsers you can download by right-clicking and then saving them on your computer; do keep the names intact; Internet Explorer is lousy and tends to mess them up):

    Suppose these two files are in your working directory, with the names they have here. Then after having attached RSiena or RSienaTest, in R give the command

    source("correction_excludeEffects.r")

    and this should do the trick.



    September 8, 2017: new versions 1.2-3 of RSiena at CRAN;
    new version 1.2-3 of RSienaTest is available from the downloads page and the RSiena project at R-Forge.

    For the moment, the versions of RSiena at CRAN and R-Forge are identical. The creation of the new CRAN version should have made the package more stable.
    The only change in functionality between versions 1.2-3 and 1.2-1 is that the totAlt effect now has been made available also for non-directed networks.



    September 4, 2017: new versions 1.2-1 of RSiena and 1.2-2 of RSienaTest are available from the downloads page and the RSiena project at R-Forge.

    Short overview of main changes:

    1. The version number now was moved up to 1.2.x because there will be a new CRAN version, and all the changes in the past years justify going from 1.1.x to 1.2.x.
    2. You now need to specify MaxDegree, modelType, behModelType and Offset in a different way for sienaAlgorithmCreate, and algorithm objects have to be created anew!!
    3. There is a new function score.Test which allows more easily getting results of score tests.
    4. The use of modelType and the recent addition behModelType was corrected. For networkModelType = 3 (initiative and confirmation model for non-directed networks), an offset is added to the confirmation model. See the manual for further explanation.
    5. There are improvements of sienaBayes, which should avoid errors occurring for some specifications.

    A more complete listing of all changes is as follows.

    Main changes in RSiena and RSienaTest visible to users:

    • Score test and sienaGOF corrected for the case that some parameters are fixed and not tested.
    • New function score.Test. This allows, for a sienaFit object in which some effects were tested with test=TRUE, to get the results of the score-type test for some or all of the parameters tested. When some effects are tested with test=TRUE, the results of the score test are also presented when printing the estimation result.
    • Argument varName added to updateTheta. This allows the update to be restricted to one or more of the dependent variables.
    • Operation of option 'absorb' (behModelType = 2) corrected.
    • In sienaAlgorithmCreate, for parameters Offset, MaxDegree, modelType, and behModelType, it is now required that these are named vectors with the names of the dependent variables, or NULL. The use of non-named vectors for these, in the case of using multiple processes may have led to errors in earlier versions. Note that the default for MaxDegree now is not MaxDegree=0 but MaxDegree=NULL.
    • New effects sameXInPopIntn, sameXOutPopIntn, sameXInActIntn, sameXOutActIntn.
    • For networkModelType = 3 (initiative and confirmation model), an offset is added to the confirmation model; this is taken from Offset in the algorithm object.
    • For effects inPopX, outPopX, inActX, outActX, sameXInPop, sameXOutAct, diffXInPop, diffXOutAct, homXInPop, homXOutAct, altXOutAct, diffXOutAct, changed interaction2 to '' (was '1', erroneously) and default parameter to 1.
    • Additional parameter dropRates in print.sienaEffects (useful for sienaBayes with many groups).
    • In print.sienaEffects, omit last remark about random effects if only one line is printed.
    • Corrected use of modeltype.
    • Added an example for multiple processes in the help page for sienaGOF, and took out verbose reporting from sienaGOF in case of multiple processes.
    • Change in print.sienaAlgorithm for modelType.
    Changes in RSiena :
    • New parameter OffSet in sienaAlgorithmCreate.
    • Registration of native routines (requirement for CRAN).
    Changes in RSienaTest:
    • Parameter UniversalOffset of sienaAlgorithmCreate renamed to Offset.
    • Added posterior variances to print.sienaBayesFit.
    • Correction of error in sienaBayes that could lead to errors in results of estimations with data sets for multiple dependent variables (networks and/or behavior) with model specifications that contain an interaction effect for a dependent variable that is not the last of the dependent variables.
    • Various changes in sienaBayes for less memory use and avoiding crashes in some situations.



    June, 2017: RSiena wins the INSNA William D Richards Award.

    This is a "lifetime achievement award", granted at the 2017 International Sunbelt Social Networks Conference (Beijing) to Tom Snijders and Christian Steglich, for "publically available social network analysis software without which it would be impossible to study social networks".



    May 12, 2017: new versions 1.1-307 of RSiena and RSienaTest are available from the downloads page and the RSiena project at R-Forge.

    Main changes in RSiena and RSienaTest visible to users:

    • New function updateSpecification to include in an effects object a set of effects that are already included in another effects object. This can be useful, e.g., for multivariate data sets, for copying the specification for one dependent variable to another dependent variable.
    • New effects inPopX, outPopX, inActX, outActX, sameWXClosure, degPlus, absDiffX, avAltPop, totAltPop, egoPlusAltX, egoPlusAltSqX, egoRThresholdX, egoLThresholdX, altRThresholdX, altLThresholdX.
    • outOutAss dropped for symmetric networks; only outInAss remains.
    • egoX effect has interactionType='ego' also for symmetric networks.
    • The exclusion of effects if the variance of a covariate is 0, or a covariate has only two values, is dropped (these effects have no meaning, but their exclusion was a potential nuisance for meta-analyses).
    • Description of gwespFB and gwespBF corrected.
    • includeInteraction has an additional parameter random.
    • siena08 now also accepts a list for ...
    • Argument behModelType added to sienaAlgorithmCreate. For behModelType=2, the 'absorbing option' is chosen in the model for behavioral dependent variables; see Section 5.8 of the manual.
    • Indication of effect parameters dropped in names for altInDist2 and totInDist2 (they have no effect parameters).
    • ModelType now is specific to the dependent variable: given as a named integer vector in sienaAlgorithmCreate.
    • sienaAlgorithmCreate, siena07: new option lessMem, reducing storage in siena07 by leaving out z$ssc and z$sf2 from the object produced;
      note that these are used by sienaTimeTest and sienaGOF, so running those functions will be impossible for sienaFit object obtained with lessMem=TRUE.
    • extended information in print.sienaAlgorithm.
    • Modified check for singular covariance matrix after Phase 3.
    • Warning if includeEffects is used with parameter random.
    • Warning for impossible or zero changes if maximum likelihood (see Section 6.9, and issue below for March 21, 2017).
    • Some parts dropped from the object produced by siena07 to reduce memory use.

    Changes in RSienaTest:

    • sienaBayes: various changes to save memory (thanks to Ruth Ripley); improved reporting of groups with no changes;
      warning for impossible changes, see Section 11.3.1 of the manual;
      if priorRatesFromData=2, change to different robust covariance matrix estimator when this is necessary (i.e., for small number of groups);
      in print.summary, also report nImproveMH;
      a few lines added to help file.
    • print.sienaEffects gives dimensions of priorMu and priorSigma if includeRandoms.



    March 21, 2017: Bug for maximum likelihood and sienaBayes estimation due to structural values
    This is a bug for versions up to and including 1.1-305.

    For some time it has been known that maximum likelihood estimation using siena07 may have problems with structural zeros; also, it has been found that sienaBayes sometimes hangs after the initialization phase.
    This seems to be because likelihood-based estimation by RSiena does not allow the following data configurations:

    • tie variables in two consecutive waves changing from structural zero (code 10) to 1;
    • tie variables in two consecutive waves changing from structural one (code 11) to 0;
    • tie variables in three consecutive waves changing from structural zero (code 10) to NA to 1;
    • tie variables in three consecutive waves changing from structural one (code 11) to NA to 0;
    • and, for more than three consecutive waves, similar patterns with more NAs in between.
    If data sets with such configurations are given to RSiena, it will hang at the moment that it tries to create the chain for likelihood simulations.

    To solve such issues, the following remedies may be applied.

    • If you are using siena07, then convert the data to a multi-group data set (a procedure somewhat similar to what is proposed by de la Haye, Embree, Punkay, Espelage, Tucker and Green in Social Networks, 2017; but without changes to the actor set).
    • If you are using sienaBayes, make minimal changes to the data set so as to avoid the configurations mentioned above. For example, replace sequences 10-1 by NA-1, and 10-NA-1 by NA-NA-1.
      An example script to make such replacements is changeForbiddenChanges.R
    This solution for siena07 is perfect because it does not change the data. For sienaBayes it is less than perfect, but acceptable if the fraction of such changes is small. The solution for sienaBayes makes sense because, if the observations for dyad (i,j) are a structural zero for Wave 1 and a tie for Wave 2, then the tie must have started some time before Wave 2; so the structural zero cannot have persisted all the way from Wave 1 right up to Wave 2.

    In version 1.1-306 of RSiena/Test, the functions siena07 and sienaBayes include checks for these configurations, so that they lead to decent error messages instead of R hanging indecently.



    October 10, 2016: new versions 1.1-302 of RSiena and RSienaTest are available from the downloads page and the RSiena project at R-Forge.

    Main changes in RSiena and RSienaTest visible to users:

    • New effects: sameXCycle4, homCovNetNet, contrastCovNetNet, covNetNetIn, homCovNetNetIn, contrastCovNetNetIn, inPopIntnX, inActIntnX, outPopIntnX, outActIntnX.
    • Changes permitting the 4-cycles effects for larger and denser networks.
    • Dropped cl.XWX effect from two-mode - one-mode coevolution (did not belong).
    • egoSqX is an ego effect.
    • Added cycle4 for one-mode networks.
    • Added outAct, outInAss for symmetric networks.
    • sienaRI: Structural zeros and ones are excluded from the calculations; option getChangeStatistics was added; row names were given to matrices that have rows corresponding to effects; the function now runs for models with only 1 parameter, and does not run, but also does not crash for a bipartite dependent variable.
      In the manual, a section about sienaRI was added (Section 13.5).
    • Warning if includeEffects is used with parameter parameter (which it doesn't have, and signals confusion with getEffect).
    • Small additions to print.sienaAlgorithm.
    • Clearer output for MaxDegree in print.sienaFit.
    • Correction of how effect parameter for outInAss for 2-mode networks is reported.


    August 18, 2016: new versions 1.1-296 of RSiena and RSienaTest are available from the downloads page and the RSiena project at R-Forge.

    Note: for the Mac version of RSienaTest, use the version at the downloads page rather than at R-Forge.

    Changes in RSiena and RSienaTest:

    • Corrected cycle4 effect for parameter=2 (i.e., sqrt version). (The results were not wrong but the new estimator for this effect is more efficient.)
    • Error corrected in effects avWAlt and totWAlt.
    • Warning if includeInteraction is used for more interactions than available given parameters nintn and behNintn.
    • Additional auxiliary function CliqueCensus presented in help page sienaGOF-auxiliary.
    • Deleted session parameter from print01Report.


    Incompatibilities

    It is not guaranteed that objects created by earlier versions of RSiena/Test can still be manipulated by more recent versions. Sometimes they will, but there also may be clashes. Therefore when starting to work with a new version of RSiena/Test it is advisable to create data objects and effects objects anew. If this leads to serious problems that cannot be solved by updating your objects, you could bring this up in the Siena/Stocnet yahoo discussion group.
    Incompatibilities of functions that may lead to the necessity to adapt scripts are listed below.

    • Since version 1.1-282, parameter priorRatesFromData in sienaBayes has values 0-1-2, with 0 = former FALSE, 1 = former TRUE, 2 = (new and default) robust estimation of prior for rate parameters from estimates at the end of initialization phase.
    • Since version 1.1-279, an effects object is no longer used as argument for print01Report() (this was superfluous).
    • Since version 1.1-279, effect AltsAvAlt was renamed to avXAlt (in line with other naming conventions).

    Bugs

    Minor bugs

    • There is an error in the description of the gwespFB and gwespBF effects in RSiena/Test. The error is made in the effectsDocumentation and also when printing results of a fit including these effects, and also in the manual p. 113. The direction of the arrows is reversed. The description of the mixed effects on p. 133 is correct.
      This will be corrected in the next version (1.1-306).
    • Under R 3.4.0, running siena07 and sienaBayes will lead to the following warning messages:

      In x$truncation * fra/maxRatio :
      Recycling array of length 1 in vector-array arithmetic is deprecated. Use c() or as.vector() instead.

      This can safely be ignored (it's just a warning, due to stricter and safer rules in R 3.4.0). In the next version of RSiena/Test (1.1-306) this warning will have disappeared.

    • There is an error in sienaGOF() in the case that co-evolution of one-mode and two-mode networks is analysed; the bug occurs only if the networks were ordered in sienaDataCreate in such a way, that a two-mode network comes before a one-mode network. Then sienaGOF will give incorrect results but most likely will run into an error.
      The solution is simple: order the networks in sienaDataCreate so that the one-mode networks come first, followed by the two-mode networks, followed by the behavioral variables.
    • The implementation of endowment and creation effects does not agree with changing structural zeros or ones. Therefore, if there is changing composition in a data set and the model should contain endowment and/or creation effects, the changing composition should be reflected by a changing composition ("compositionChange" object) and not by structural zeros.
      On the other hand, for goodness of fit testing by sienaGOF, data with a "compositionChange" object are not treated properly; there the structural zeros are necessary for treating changing composition. But if you use NA codes for the tie variables of absent actors, and these are indicated as such by a "compositionChange" object, then sienaGOF will work properly.
    • The function sienaRIDynamics seems still / again to be incorrect (runs into a crash).
    • The effect outRateLog (dependence of log rate on log(outdegree+1)) works properly only for non-conditional estimation (set cond = FALSE in sienaAlgorithmCreate).

    For past bugs, see below.


    May 28, 2016: new versions 1.1-294 of RSiena and RSienaTest are available from the downloads page and the RSiena project at R-Forge.

    Changes in RSiena and RSienaTest:

    • The manual was taken out of the installation; it still is in the source code distribution as RSienatest\doc\RSiena_Manual.tex. It still is publicly available from this website as http://www.stats.ox.ac.uk/~snijders/siena/RSiena_Manual.pdf.
    • New effects totAltEgoX, totAltAltX, egoSqX, diffX, diffSqX, egoDiffX, avAltW, totAltW, avSimW, totSimW, jumpFrom, jumpSharedIn, mixedInXW, mixedInWX, avWalt, totWAlt.
    • inActIntn also implemented for two-mode dependent networks.
    • Endowment and creation effects were added for inAct, inActSqrt, outAct, outActSqrt.

    Changes in RSiena:

    • siena01Gui() and sienaDataCreateFromSession() dropped.
    Changes in RSienaTest:
    • Change in endowment effect estimation for avAlt effect.
    • New effects simmelian, simmelianAltX, avSimmelianAlt, totSimmelianAlt.


    February 22, 2016. Convergence issues; and AdSUM-2016

    Additional advice for how to achieve good convergence (in cases where this is difficult, e.g., complicated models and data sets) is given in Section 6.2 of the manual.
    This - and more - was discussed at the third Advanced Siena Users' Meeting (AdSUM-2016) held in Zürich February 19-20, of which the slides are available here.



    February 4, 2016: new versions 1.1-290 of RSiena and 1.1-291 of RSienaTest are available from the RSiena project at R-Forge and from the downloads page.

    Changes in RSiena and RSienaTest:

    • New effects FBDeg, FRDeg, BRDeg (RFDeg was mentioned earlier but was not implemented; its place is now taken by FRDeg), gwespFFMix, gwespBBMix, gwespBFMix, gwespFBMix, gwespRRMix, gwespMix.
    • Corrected the omission of the check for positive derivative matrix at the end of phase 1 for effects with fixed parameters.
    • Added parameters fix, test, parameter to includeInteraction(). This implies that the roundabout way of setting internal effect parameters for interaction effects, still explained at the help page of setEffect(), now is not necessary any more.
    • More helpful error message for incorrect nodesets in sienaDataCreate(); extended help pages for sienaDataCreate() and sienaDependent().
    • Correction to allow estimation for a one-dimensional parameter.
    • fromBayes bug corrected.

    Changes in RSienaTest:

    • sienacpp() added (programmed by Felix Schönenberger); this includes Generalized Method of Moments estimation (Amati - Snijders); in the manual (Section 6.7) and in Siena algorithms (Section 3.3), some explanations are given.
    • sienaBayes(): option initML added; check for zero distances; corrected function improveMH() in initialization.
    • print.multipleBayesTest(): option descriptives added.
    • glueBayes(): added p1 and p2 to created object.



    January 19, 2016. New book: Multilevel Network Analysis for the Social Sciences

    This book is announced here. It is mostly about cross-sectional network data. In Chapter 2, The Multiple Flavours of Multilevel Issues for Networks by Tom Snijders, there is a section about the Stochastic Actor-Oriented Models for Multilevel Networks.
    Note that almost at the same time, a special issue appeared of the journal Social Networks about the same theme.



    September 12, 2015. Specification of transitivity for actor-oriented models

    For modeling transitivity in actor-oriented models, the gwesp effect often is better than the 'traditional' transTrip and transTies effects. 'Better' in the sense that often it will give a better fit (or, as good a fit as using the other two together) and better convergence. The gwesp effects (there are several: gwespFF, gwespBB, etc.) are extensively explained in the manual. gwespFF is similar to transTrip, and gwespBB is similar to cycle3. Then there are gwespFB, gwespBF, and gwespRR, which are other ways to improve the fit of the triad census in cases where this is desirable.

    It probably is usually not a big difference, but one might use gwespFF as a default instead of transTrip or transTies, unless one does not want to go through the hassle of the extra explanation.

    Note that the gwesps are elementary effects. The difference between elementary and evaluation effects also is explained in the manual. A practical consequence is that they can be used in interactions (their interactionType is dyadic). Therefore, the gwesp-related effect that has the similar role as transRecTrip (cf. Per Block, 2015, "Reciprocity, transitivity, and the mysterious three-cycle", Social Networks, 40:163-173) is the interaction between gwespFF and reciprocity, and can be defined by includeInteraction(). Since version 1.1-290, the internal effect parameter of such an interaction can be set straightforwardly by includeInteraction(). See the help page for this function (and possible also the help page for setEffect()).



    September 10, 2015: new version (1.1-289) of RSiena and RSienaTest is available from the RSiena project at R-Forge and from the downloads page.

    Changes in RSiena and RSienaTest:

    • New defaults for siena07() set in sienaAlgorithmCreate(): doubleAveraging=0, diagonalize=0.2 (for MoM).
      Note: this will influence convergence properties (it should improve convergence).
    • Improved one-step approximations to expected Mahalanobis distances in sienaGOF() (using control variates for score function).
    • Permit 3-way interactions with one ego and two dyadic effects (this was erroneously not allowed).
    • New effects Jin, Jout, JinMix, JoutMix, altXOutAct, doubleInPop, doubleOutAct.
    • print01Report() now reports in-degrees also for two-mode networks.
    • Better error handling for sienaTimeTest and scoreTest.
    • inOutAss is dyadic.
    • Corrected effectName and functionName of inPopIntn, outPopIntn, inActIntn, and outActIntn ('in' and 'out' were missing).
    • Check for positive derivative matrix at the end of phase 1 (non-positive estimated derivatives lead to repeating a prolonged phase 1) omitted for effects with fixed parameters.
    Changes in RSiena:
    • New effects homXOutAct, FFDeg, BBDeg, RFDeg, diffXTransTrip (ported from RSienaTest).
    • sameXInPop and diffXInPop also added for two-mode networks; but they are not dyadic!
    • In names of behavior effects and statistics dropped the (redundant) parts 'behavior' and 'beh.' (as done earlier in RSienaTest).
    Changes in RSienaTest:
    • New function extract.sienaBayes().
    • sienaBayes(): options diagonalize=0.2, doubleAveraging=0 for estimation of initial models in initialization phase; save initial results in case of divergence during initialization phase; check for large initial estimates done only for non-rate parameters.


    July 21, 2015: new version (1.1-288) of RSiena and RSienaTest is available from the RSiena project at R-Forge and from the downloads page.

    Changes in RSiena and RSienaTest:

    • plot.sienaRI(): new parameter actors; proportions with piechart improved (hopefully); effect of parameter radius changed.
    • siena.table does no more produce the double minus sign in html output.
    Changes in RSiena:
    • Correction of error for two-mode networks in sienaGOF().
    Changes in RSienaTest:
    • New effects homXOutAct, FFDeg, BBDeg, RFDeg, diffXTransTrip.
    • sameXInPop and diffXInPop also added for two-mode networks; but they are not dyadic!
    • In names of behavior effects and statistics dropped the (redundant) parts behavior and beh..
    • sienaBayes: new parameter nSampRates; correction in use of prevBayes; more efficient calculation of multivariate normal density.
    • Small changes in HowToCommit.tex.


    Some recent publications (many others are at the applications page)

    A paper that presents a method for meta-analysis, which is (presumably) better than siena08(). It uses R package mvmeta; note that another package for meta analysis that may be used is metafor. These packages can include explanatory variables at the network level. Also see the RSiena manual, Section 11.2:

  • Weihua (Edward) An (2015). Multilevel meta network analysis with application to studying network dynamics of network interventions. Social Networks, 43, 48-56.
    DOI: http://dx.doi.org/10.1016/j.socnet.2015.03.006.

    Abstract This paper introduce new methods for multilevel meta network analysis. The new methods can combine results from multiple network models, assess the effects of predictors at network or higher levels and account for both within- and cross-network correlations of the parameters in the network models. They are demonstrated by applying them to network dynamics of a smoking prevention intervention implemented in 76 classes of six middle schools in China.

    Remark: on p. 48 of this paper, it is stated that 'Previous meta network analysis, maybe except the Fisher's method for combining independent p-values (Snijders and Bosker, 2012; Ripley et al., 2014), mostly assumes that the estimated parameters for a particular variable in the network models are generated by a common effect'. This is incorrect. In the method of Snijders and Baerveldt (2003) implemented in RSiena function siena08(), the model is a random effects model (following Cochran, 1954), but without the assumption of a normal distribution. The statement is correct that Fisher's method does not make the assumption of a common effect.
    The fixed effects meta-analysis model is available in RSiena as the multi-group option. In siena08() also a ML estimate is implemented under normality; see the manual.
    However, the REML method used in An's paper, and available in mvmeta and metafor, is usually slightly better than the ML method; again, see the manual.


    Two papers about co-evolution of two networks, one positive and one negative:

  • J. Ashwin Rambaran, Jan Kornelis Dijkstra, Anke Munniksma, and Antonius H.N. Cillessen (2015). The development of adolescents' friendships and antipathies: A longitudinal multivariate network test of balance theory. Social Networks, 43, 162-176.
    DOI: http://dx.doi.org/10.1016/j.socnet.2015.05.003.

    Abstract This paper examines the interplay between friendship (best friend) and antipathy (dislike) relationships among adolescents. Based on (structural) balance theory, it was expected that friendships would be formedor maintained when two adolescents disliked the same person (shared enemy hypothesis), that friends would tend to agree on whom they disliked (friends' agreement hypothesis), that adolescents would tend to dislike the friends of those they disliked (reinforced animosity hypothesis), and that they would become or stay friends with dislikes of dislikes (enemy's enemy hypothesis). Support was found for the first three hypotheses, and partially for the fourth.

  • Gijs Huitsing, Tom A.B. Snijders, Marijtje A.J. van Duijn, and René Veenstra (2014). Victims, bullies, and their defenders: a longitudinal study of the coevolution of positive and negative networks. Development and Psychopathology, 26, 645-659.
    DOI: http://dx.doi.org/10.1017/S0954579414000297

    Abstract The (co)evolution of bullying/victimization and defending was examined within three elementary schools across three time points within a year. Most bullies and defenders were in the same grade as the victims, although a substantial number of bullies and defenders were in other grades (most often one grade higher). In line with goal-framing theory, multiplex network analyses provided evidence for the social support hypothesis (victims with the same bullies defended each other over time) as well as the retaliation hypothesis (defenders run the risk of becoming victimized by the bullies of the victims they defend). In addition, the analysis revealed that bullies with the same victims defended each other over time and that defenders of bullies initiated harassment of those bullies' victims.


    Co-evolution of a one-mode and a two-mode network, with differentiation between creation and maintenance of ties:

  • Christoph Stadtfeld, Daniele Mascia, Francesca Pallotti, and Alessandro Lomi (2015). Assimilation and differentiation: A multilevel perspective on organizational and network change. Social Networks, in press.
    DOI: http://dx.doi.org/10.1016/j.socnet.2015.04.010.

    Abstract This paper analyses how concurrent multilevel processes of (internal) organizational and (external) network change affect one another over time. New effects are presented that afford specification and identification of two apparently conflicting micro-relational mechanisms that jointly affect decisions to modify the portfolio of internal organizational activities. The first mechanism, assimilation, makes network partners more similar by facilitating the replication and diffusion of experience. The second mechanism, functional differentiation, operates to maintain and amplify differences between network partners by preventing or limiting internal organizational change. Data are analyzed that were collected on a regional community of hospital organizations connected by collaborative patient transfer relations observed over a period of seven years. It is found that processes of social influence conveyed by network ties may lead both to similarity and differences among connected organizations.


    Network dynamics with actors grouped in pairs:

  • Christoph Stadtfeld and Alex (Sandy) Pentland (2015). Partnership Ties Shape Friendship Networks: A Dynamic Social Network Study. Social Forces 94, 453-477.
    DOI: http://dx.doi.org/10.1093/sf/sov079.

    Abstract Partnership ties shape friendship networks through different social forces. First, partnership ties drive clustering in friendship networks: individuals who are in a partnership tend to have common friends and befriend other couples. Second, partnership ties influence the level of homophily in these emerging friendship clusters. Partners tend to be similar in a number of attributes (homogamy). If one partner selects friends based on preferences for homophily, then the other partner may befriend the same person regardless of whether they also have homophilic preferences. Thus, two homophilic ties emerge based on a single partner's preferences. This amplification of homophily can be observed in many attributes (e.g., ethnicity, religion, age). Gender homophily, however, may be de-amplified, as the gender of partners differs in heterosexual partnerships. In our study, we follow dynamic friendship formation among 126 individuals and their cohabiting partners in a university-related graduate housing community over a period of nine months.



  • June/July, 2015: Advanced Siena users' workshop slides of Sunbelt 2015

    At the literature tab of the Siena pages, section Presentations (teaching material), there are available the slides of the Advanced Siena users' workshops of Sunbelt 2014 and 2015.



    May 22, 2015: Stronger convergence criterion necessary

    It turns out that the usual convergence criterion, tmax <= 0.10 where tmax is the maximum absolute value of the t-ratios for convergence, is not a satisfactory measure of convergence of the Robbins-Monro estimation algorithm used for RSiena. Earlier, simulation results had been obtained that provided support for this criterion, but it now has been found that especially when using non-centered covariates this criterion may be inadequate; and if models with non-centered covariates lead to potential problems for this convergence criterion, then it cannot be excluded that other models also do.
    A stronger convergence criterion has been found that does give an adequate signal of convergence (the past shows that you never can be sure, but to our current knowledge there are convincing signs that this is adequate). This is the overall maximum convergence ratio, described since some time in the RSiena manual, and given with sienaFit objects as the element tconv.max. This 'overall maximum convergence ratio' is defined as the maximum convergence ratio for linear combinations of the estimation statistics, while tmax is the maximum absolute convergence ratio for only the set of statistics (i.e., coordinates) directly used for the estimation. Section 3.2 of Siena_algorithms.pdf gives definitions and further explanations, and arguments for the use of the criterion tconv.max <= 0.25. Some examples were found, for models including non-centered covariates, with important differences between results for estimations where only tmax <= 0.10, and estimations where also tconv.max <= 0.25. This shows that the earlier convergence criterion is not reliable. (@*&#!!) The stronger criterion can be satisfied by repeating, if necessary, estimation with successive use of the prevAns parameter, as before, and as explained in the manual. To improve convergence, some additional options for siena07() were developed. These are given as parameters to sienaAlgorithmCreate(). The main parameters for this purpose are doubleAveraging, diagonalize (this option was corrected and improved), and n2start. Their use is explained in Section 6.1.3 of the new edition of the RSiena users' manual.

    Thus, the new rule for convergence is that tmax <= 0.10 (as before) and tconv.max <= 0.25 (new); the latter condition is the more important one. This now is mentioned in the RSiena manual.

    The conclusion is that RSiena users are advised to reconsider their estimation results and continue the estimation until this criterion is satisfied. There is a possibility of differences in the results; it is not expected that there will be many important differences, but the possibility does exist.



    May 22, 2015: new version (1.1-286) of RSiena and RSienaTest is available from the RSiena project at R-Forge.

    Changes in RSiena and RSienaTest:

    • The new convergence criterion tconv.max <= 0.25 is advertised in the manual.
    • Component tmax added to sienaFit objects and tconv.max mentioned in print.sienaFit().
    • sienaAlgorithmCreate() has new arguments n2start, truncation, doubleAveraging, standardizeVar; these allow improved possibilities for convergence, as indicated in Section 6.1.3 of the manual.
    • Diagonalization (option diagonalize) corrected.
    • When there are missings in constant or changing monadic covariates, and centered=FALSE for their creation by coCovar() or varCovar(), the mean will be imputed (used to be 0, which was an error). For changing covariates, this is the global mean.
    • In coCovar() and varCovar() there is a new argument imputationValues, which are used (if given) for imputation of missing values. Like all missings, they are not used for the calculation of the target statistics in the Method of Moments.
      If one has a reasonable way of imputing the missing values, this will give more adequate simulations than the imputation of everything by the global mean.
    • New effects: outOutActIntn, toDist2, from.w.ind.
    • In the target statistic for the higher effect, contributions for value(ego)=value(alter) are now set appropriately at 0.5 (was 0).
    • Decent error message when there are (almost) all NA in the dependent behavioural variable.
    • The centering within effects for similarity variables at distance 2 now is done by the same similarity means as for the simX effect.



    Some past bugs

    • The four-cycle effects (cycle4, cycle4ND, sharedPop) sometimes had convergence difficulties caused by numerical overflow of the int data type in C++; this was the case if the number of nodes and the degrees etc. were such that the number of three-paths constituting the four-cycles could become too large. This was repaired in version 1.1-302.
      By the way, numerical overflow also will be a limitation for applying RSiena to very large networks for some other effects.
    • For two-mode networks there was an error in sienaGOF() (semi-diagonal of generated adjacency matrices set to 0); this was repaired in RSiena version 1.1-288 and in RSienaTest version 1.1-286 of June 02, 2015.
    • Siena and RSienaTest until and including version 1.1-284 dealt incorrectly with missing data in covariates when in coCovar() and varCovar() the option "center=FALSE" is used (this is not the default option). When covariates are not centered and contain missing values, these missings were imputed by the value 0. In most cases, this is not reasonable. This means that results for non-centered covariates with missings will be wrong.
    • sienaBayes until and including version 1.1-283 contained an error for sampling the constant parameters.
      This was corrected in version 1.1-284.
    • If there is an exact multicollinearity, a warning was given 'Noninvertible estimated covariance matrix', but nevertheless standard errors were reported (which should not be used).
      This was corrected in version 1.1-284.
    • print01Report gave an error message for a sienaGroup object where the component objects have constant dyadic covariates.
      This was corrected in version 1.1-284.
    • Effects cl.XWX and cl.XWX2 were not implemented correctly in versions RSiena and RSienaTest up to and including version 1.1-281. In these versions they were identical to cl.XWX1. A correction was made in version 1.1-282. (This error was not present in effects XWX and XWX2.)
    • The component se of sienaFit objects produced by siena07(), which is there since version 1.1-271, was incorrect until and including version 1.1-277: it contained the squared standard errors instead of the standard errors themselves. This was fixed in version 1.1-278.
      This only affected the se component, the standard errors given in the usual output and print of siena07() were not affected.



    Slides of presentations in 2014 on 'Advanced Siena' issues

    • The slides of the presentation by Tom Snijders about recent developments in Siena, presented at the meeting of the European Collaborative Research Project Social Influence in Dynamic Networks (Paris, December 15-18, 2014) are available here.

    • The slides of the Advanced Siena user's workshop at Sunbelt 2014 are available here. Pages 10-13 are new (they were not included in the workshop).



    Some recent publications (many others are at the applications page)

    First a position paper about coevolution that contains no application of Siena, but mentions it as a tool.

    • Stefan Tasselli, Martin Kilduff, and Jochen Menges. Microfoundations of Organizational Social Networks: A Review and an Agenda for Future Research (2015). Journal of Management, 41, 1361-1387.
      DOI: http://dx.doi.org/10.1177/0149206315573996.

      Abstract
      This paper focuses on an emergent debate about the microfoundations of organizational social networks. We consider three theoretical positions: an individual agency perspective suggesting that people, through their individual characteristics and cognitions, shape networks; a network patterning perspective suggesting that networks, through their structural configuration, form people; and a coevolution perspective suggesting that people, in their idiosyncrasies, and networks, in their differentiated structures, coevolve. We conclude that individual attitudes, behaviors, and outcomes cannot be fully understood without considering the structuring of organizational contexts in which people are embedded, and that social network structuring and change in organizations cannot be fully understood without considering the psychology of purposive individuals. To guide future research, we identify key questions from each of the three theoretical perspectives and, particularly, encourage more research on how individual actions and network structure coevolve in a dynamic process of reciprocal influence.

    Then some applications that may be of general interest.

    • Per Block (2015). Reciprocity, transitivity, and the mysterious three-cycle. Social Networks, 40, 163-173.
      DOI: http://dx.doi.org/doi:10.1016/j.socnet.2014.10.005.

      Abstract
      Reciprocity and transitivity are the two most important structural mechanisms underlying friendshipnetwork evolution. While on their own they are understood in great detail, the relation between them israrely studied systematically. Are friendships outside of social groups more or less likely to be reciprocated than friendships embedded in a group? Using a theoretical framework that focusses on the situations inwhich friends interact and the social structures that stabilise one-sided friendships, I propose that thetendency towards reciprocation of friendships within transitive groups is usually lower than outside of transitive groups. In a meta-analysis of two datasets including 29 friendship networks using stochastic actor-oriented models, the interaction between reciprocity and transitivity is analysed. Supporting the theoretical reasoning, the interaction is consistently negative. Second, the tendency against forming three-cycles in friendship networks, which was consistently found in previous studies, is shownto be spurious and a result of neglecting to control for the tendency against reciprocation in transitive groups. The tendency against three-cycles is commonly seen as an indicator that unreciprocated friendships indicate local hierarchy differences between individuals; this proposition has to be re-evaluated in light of the findings of this study. Future studies that analyse the evolution of friendship networks should consider modelling reciprocation in transitive triplets and potentially omit modelling three-cycles.

    • Per Block and Thomas Grund (2014). Multidimensional homophily in friendship networks. Network Science, 2, 189-212.
      DOI: http://dx.doi.org/10.1017/nws.2014.17.

      Abstract
      Homophily -the tendency for individuals to associate with similar others- is one of the most persistent findings in social network analysis. Its importance is established along the lines of a multitude of sociologically relevant dimensions, e.g. sex, ethnicity and social class. Existing research, however, mostly focuses on one dimension at a time. But people are inherently multidimensional, have many attributes and are members of multiple groups. In this article, we explore such multidimensionality further in the context of network dynamics. Are friendship ties increasingly likely to emerge and persist when individuals have an increasing number of attributes in common? We analyze eleven friendship networks of adolescents, draw on stochastic actor-oriented network models and focus on the interaction of established homophily effects. Our results indicate that main effects for homophily on various dimensions are positive. At the same time, the interaction of these homophily effects is negative. There seems to be a diminishing effect for having more than one attribute in common. We conclude that studies of homophily and friendship formation need to address such multidimensionality further.

    • Alessandro Lomi and Christoph Stadtfeld (2014). Social Networks and Social Settings: Developing a Coevolutionary View. KZfSS Kölner Zeitschrift für Soziologie und Sozialpsychologie, 66, 395-415.
      DOI: http://dx.doi.org/10.1007/s11577-014-0271-8.

      Abstract
      One way to think about social context is as a sample of alters. To understand individual action, therefore, it matters greatly where these alters may be coming from, and how they are connected. According to one vision, connections among alters induce local dependencies-emergent rules of social interaction that generate endogenously the observed network structure of social settings. Social selection is the decision of interest in this perspective. According to a second vision, social settings are collections of social foci-physical or symbolic locales where actors meet. Because alters are more likely to be drawn from focused sets, shared social foci are frequently considered as the main generators of network ties, and hence of setting structure. Affiliation to social foci is the decision of central interest in this second view. In this paper we show how stochastic actor-oriented models (SAOMs) originally derived for studying the dynamics of multiple networks may be adopted to represent and examine these interconnected systems of decisions (selection and affiliation) within a unified analytical framework. We illustrate the empirical value of the model in the context of a longitudinal sample of adolescent participating in the Glasgow Teenage Friends and Lifestyle Study. Social selection decisions are examined in the context of networks of friendship relations. The analysis treats musical genres as the main social foci of interest.

    • Steven A. Haas and David R. Schaefer (2014). With a little help from my friends? Asymmetrical social influence on adolescent smoking initiation and cessation. Journal of Health and Social Behavior, 55, 126-143.
      DOI: http://dx.doi.org/10.1177/0022146514532817.

      Abstract
      This study investigates whether peer influence on smoking among adolescents is asymmetrical. We hypothesize that several features of smoking lead peers to have a stronger effect on smoking initiation than cessation. Using data from the National Longitudinal Study of Adolescent Health we estimate a dynamic network model that includes separate effects for increases versus decreases in smoking, while also controlling for endogenous network change. We find that the impact of peer influence is stronger for the initiation of smoking than smoking cessation. Adolescents rarely initiate smoking without peer influence but will cease smoking while their friends continue smoking. We discuss the implications of these results for theories of peer influence and health policy.



    April 2, 2015: new version (1.1-284) of RSiena and RSienaTest is available from the RSiena project at R-Forge.

    Changes in RSiena and RSienaTest:

    • New effects: simEgoDist2, simEgoInDist2, simEgoDist2W, simEgoInDist2W, sameXOutAct, diffXInPop, diffXOutAct. The centering for the similarity measure in effects such as simEgoDist2 and simDist2 is not yet sorted out (but this affects only the outdegree parameter).
    • For creating new effects, it is relevant that new effect groups covarABNetNetObjective, covarANetNetObjective, and covarBNetNetObjective were introduced. See SienaSpec.tex/pdf, section 4.9: covarNetNet.
    • Bug corrected that occurred in print01Report() for a sienaGroup object where the component objects have constant dyadic covariates.
    • When a statistic is not plotted in plot.sienaGOF() because its variance is 0, a note about this is printed to the screen.
    • Minimum and maximum of plotted region in plot.sienaGOF() is calculated without taking into account non-plotted statistics.
    • Bug in sienaTimeTest() corrected that occurred for includeTimeDummy when timeDummy was greater than or equal to 10.
    • In case of collinear parameter estimates, standard errors are reported as NA.
    • plot.sienaGOF(): arguments main and ylab now are not any more arguments of this function proper, but if used they are understood as part of the ... argument (so using main and ylab as arguments now should work properly). (Thanks to David Kavaler.)

    Changes in RSienaTest:

    • sienaBayes: error corrected for sampling constant parameters (this error led to incorrect results); correction in initialization of truncation rate parameters based on prior.

    December 12, 2014: new version (1.1-282) of RSiena and RSienaTest is available from the RSiena project at R-Forge.

    Changes in RSiena and RSienaTest:

    • Effects cl.XWX and cl2.XWX corrected (thanks to Christoph Stadtfeld).
    • interactionType of gwesp.. effects was made dyadic (this permits interactions).
    • New effects reciPop, reciAct, in3Plus, maxAlt, minAlt, transTrip1, transTrip2.
    • Effect antiInIsolate2 got alias in2Plus.
    • inPop is a dyadic effect (except for non-directed networks) (this permits interactions).
    • egoX added as effect for non-directed networks (can be important for representing effects of group-level covariates in multi-group analyses).
    • Components IActors and expectedI (not relative, but 'raw' importance of effects) added to sienaRI() and print.sienaRI().
    • The check for MaxDegree when running siena07() now works properly also for sienaGroup objects.
    • Manual introduces the term elementary effects.

    Changes in RSienaTest:

    • sienaBayes: the stop caused by singularity of the precision matrix after the multi-group estimation now is circumvented; still a warning is printed to the screen. (This permits the use of multiple elementary effects that have the same target statistic).
    • sienaBayes: option priorRatesFromData changed to values 0-1-2, with 0 = former FALSE, 1 = former TRUE, 2 = robust estimation of prior for rate parameters from estimates at the end of initialization phase.



    November 19, 2014: new version (1.1-280) of RSienaTest is available from the RSiena project at R-Forge.

    Version 1.1-280 is almost the same as 1.1-279.

    Changes in RSienaTest:

    • Small changes in help pages for sienaGOF and for sienaCompositionChange.
    • New parameters nSampVarying and nSampConst in sienaBayes().



    November 13, 2014: new version (1.1-279) of RSiena is available from the RSiena project at R-Forge, and of RSienaTest from the downloads page.

    The main changes of version 1.1-279 compared to 1.1-278 are the following.

    There are two changes in RSiena and RSienaTest that will lead to incompatibilities with existing scripts, so that you will need to modify these. All scripts at the website were modified accordingly. Note, however, that sensible error messages will be produced (hopefully) when scripts according to the old implementations are used:

    • Effect AltsAvAlt renamed to avXAlt (this seems more in line with other naming conventions).
    • Effects object no longer used as argument for print01Report() (this was quite superfluous really).
    Changes in RSiena and RSienaTest:
    • A lot of new effects were added: sameXInPop, transRecTrip2, totAlt, avInAlt, totInAlt, totRecAlt, totXAlt, avXInAlt, totXInAlt, avAltDist2, totAltDist2, avTAltDist2, totAAltDist2, avXAltDist, totXAltDist2, avTXAltDist2, totAXAltDist2, avInAltDist2, totInAltDist2, avTInAltDist2, totAInAltDist2, avXInAltDist2, totXInAltDist2, avTXInAltDist, totAXInAltDist2, XWX1, XWX2, cl.XWX1, cl.XWX2. See the manual!
    • Endowment and creation effects were added for gwesp... effects.
    • Some meaningless effects for two-mode networks were dropped.
    • For non-invertible covariance matrices occurring at the end of siena07(), a diagnostic now is given for the linear combination that causes trouble.
    • The function igraphNetworkExtraction() in the help page for sienaGOF-auxiliary was corrected (the earlier version dropped isolated nodes from simulated networks).
    • In the help page for sienaGOF-auxiliary.Rd, the example of constraint is replaced by the example of eigenvector centrality (because constraint is undefined in igraph for isolated nodes, leading to computational problems).
    • The diagonal of observed networks is set to 0 in sparseMatrixExtraction(); this diagonal was liable to lead to problems in assessing goodness of fit for data sets where the original data has a non-zero diagonal.
    • The function sienaRIDynamics() (Indlekofer and Brandes, Network Science, 2013) was restored, after corrections.
    • The "file" parameter of sienaRI() was dropped (it implied platform dependence).
    • The section in manual about user-defined interaction effects was updated.
    • Parameter showAll was added to descriptives.sienaGOF().
    • Some minor checks were implemented, bugs corrected, and error messages improved (see the manual or the CHANGELOG file in the package for details).
    • The file Siena_algorithms4.tex, which contains an overview of much of the coding used in RSiena, was renamed Siena_algorithms.tex and the associated pdf file now is available also at the Siena website (literature tab).
    • p-values for goodness of fit in sienaGOF() are rounded to 3 decimal places, because more is definitely unreliable.
    • The file effects.pdf was dropped from the distribution (it can be created by effectsDocumentation()).
    Changes in RSienaTest:
    • sienaBayes(): new parameters nImproveMH and priorRatesFromData; these give the possibility to truncate initial rate parameters depending on the prior (see manual).
    • glueBayes() was corrected so that it can be applied sequentially.
    • multipleBayesTest() now allows a matrix parameter to test linear combinations.
    • Improved plot.multipleBayesTest (showing truncation at 0).


    July 25, 2014: new version (1.1-278) of RSiena and RSienaTest is available from the downloads page.

    The main changes of version 1.1-278 compared to 1.1-276 are the following.

    • s50s was added to the data set.
    • The se component of sienaFit objects was corrected (it should be the standard error, but was its square).
    • new effects totDist2, altInDist2, totInDist2, totDist2W, altInDist2W, and totInDist2W were added.
    • Some errors in print01Report() were corrected, and this function was further slightly improved, for descriptives for changing dyadic covariates and for cases where changes are only upward or downward.
    • In sienaBayes (available in RSienaTest only), the data-dependent choice of priorSigma for the rate parameters now is internally multiplied by priorKappa; z$nwarm is changed to 0 if prevBayes is used; and the plotit functionality was dropped.
    • sienaBayes(), glueBayes(), and print.sienaBayes() (all available in RSienaTest only) were adapted to allow inclusion of interaction effects without the corresponding main effects.
    • A parameter nwarm2 was added to glueBayes().

    In general you can see the changes of the two packages with respect to earlier versions in appendix B of the Siena users' manual and, more in detail, in the changeLog file in the source code.



    July 16, 2014: sienaBayes

    The function sienaBayes() in package RSienaTest version 1.1-276, developed by Johan Koskinen and Tom Snijders, now has developed far enough to be used. This is a function for multilevel Siena analyses, i.e., for analyzing network dynamics in multiple independent groups in which the same data structure, number of waves, and model specification hold, but (contrary to the existing multiGroup option) with parameters that may differ between groups. The R help page provides, as usual, the primary information. Section 11.3 of the manual is further important documentation.
    It is still a beta version, and we are in a phase of collective collection of information about how to use it. The wiki-type of interactive website that was mentioned here previously has been discontinued because it was too slow. Please use the regular Siena-Stocnet discussion list for interaction about sienaBayes.



    June 3, 2014: new version (1.1-276) of RSiena and RSienaTest is available from the downloads page.

    The main changes of version 1.1-276 compared to 1.1-274 are the following.

    • The function sienaBayes with some auxiliary functions (print, summary, bayesGlue) is now implemented in RSienaTest and released for use.
    • Function includeEffects() now includes parameters fix and test.



    April 15, 2014: Paper by Charlotte Greenan about diffusion of innovations

    The paper by Charlotte Greenan, Diffusion of innovations in dynamic networks, appeared on the website of Journal of the Royal Statistical Society: Series A (Statistics in Society).

    Abstract
    The evolution of a dynamic social network and the diffusion of an innovation are jointly modelled, dependent on one another, by using an extension of a stochastic actor-oriented model developed by Snijders, which is modified so that the adoption times follow a proportional hazards model. The asymptotic behaviour of the method-of-moments estimator is examined. The model is demonstrated on a data set involving the initiation of cannabis smoking among adolescents, and a simulation study is presented.

    Further information
    This paper models the co-evolution of networks and behavior for the case where the behavior is a binary variable that can only increase. The basic idea is that such a variable indicates a time to event. This is also called adoption of innovation, event history analysis, duration analysis, survival analysis, etc.
    The special aspect of this model is that effects are contained mainly in the rate function instead of the evaluation function. This yields the proportional hazards model for the time-to-event distribution. The implemented effects are described in the RSiena manual in Section 12.2.4: search for the average exposure effect (`avExposure') which is the first of these effects mentioned.

    For an application, see

  • John M. Light, Charlotte C. Greenan, Julie C. Rushby, Kimberly M. Nies, and Tom A.B. Snijders (2013). Onset to First Alcohol Use in Early Adolescence: A Network Diffusion Model. Journal of Research on Adolescence, 23, 487-499.
    DOI: http://dx.doi.org/10.1111/jora.12064.



  • April 26, 2014: new version (1.1-274) of RSiena and RSienaTest is available from R-Forge.

    A new version (1.1-274) of RSiena and RSienaTest can be obtained from the RSiena project at R-Forge.
    A version for Mac is available from the downloads page.

    The main changes of version 1.1-274 compared to 1.1-254 are the following.

    • There now is the new function sienaRI(), contributed by Natalie Indlekofer, to assess relative importance of effects according to
      Indlekofer, Natalie, and Brandes, Ulrik, (2013). Relative importance of effects in stochastic actor-oriented models. Network Science, Vol. 1, Issue 3, 278-304.
      DOI: http://dx.doi.org/10.1017/nws.2013.21.
    • Effect homWXClosure repaired.
    • Possibility (optional) not to center dyadic covariates.
    • Further work on sienaBayes() in RSienaTest.
    • Duplication of outInv and outSqInv effects for two-mode networks canceled.
    • Added component se (standard errors) to sienaFit objects.



    February 17, 2014: new version (1.1-254) of RSiena and RSienaTest is available from R-Forge.

    A new version (1.1-254) of RSiena and RSienaTest can be obtained from the RSiena project at R-Forge.

    The main changes of version 1.1-254 compared to 1.1-250 are the following.

    It should be noted that two changes were made that potentially have an influence on some results obtained.

    First, the effects gwespFF, gwespBB, gwespFB, gwespBF, gwespRR were modified (thanks to Nynke Niezink) to bring their parametrization in accordance with the literature. This means that the 'old' parameter alpha' was effectively replaced by alpha = -log(1-exp(alpha')); here alpha is the internal effect parameter divided by 100. For the default alpha = log(2) we have alpha' = alpha, so this means no difference.
    Second, in the help page for sienaGOF-auxiliary, geodesic distances were changed to non-directed. This makes more sense usually and was done to avoid runtime errors that occurred very rarely.

    • New effects cl.XWX, homXTransTrip, homWXClosure, and sharedPop were added.
    • Effect cycle4 extended to non-directed one-mode networks (for directed one-mode networks this is sharedPop).
    • Effects outRateLog and outTrunc2 were ported to RSiena from RSienaTest.
    • Effect jumpXTransTrip extended to non-directed networks.
    • gwesp.. effects modified (see above) and extended to non-directed networks.
    • Manual: added paragraph about how to import results from xtable() and siena.table() into MS-Word.
    • User-defined interactions for dependent behavior variables now also are handled more adequately in functions setEffect() and updateTheta(), and when using the prevAns parameter in siena07(). (This was done for user-defined interactions for networks already in version 1.1-250.)
    • sienaGOF: added the name of the sienaFit object as attribute sienaFitName to each of the sienaGofTest objects.
    • Correction in sparseNetworkExtraction() to avoid errors occurring when the extracted network has no edges.
    • In the help page for sienaGOF-auxiliary, geodesic distances changed to non-directed; this avoids a further error when the extracted network has no edges.
    • Correction: Effect to is not a dyadic effect.
    • Correction of an error in print.siena for data sets including other types than oneMode.
    • Changed bandwidth selector for violin plots in plot.sienaGOF to "nrd", to avoid long violins in cases where all simulations have the same outcome.
    • Further work on sienaBayes() in RSienaTest.



    December 4, 2013: new version (1.1-250) of RSiena and RSienaTest available from R-Forge.

    A new version (1.1-250) of RSiena and RSienaTest can be obtained from the RSiena project at R-Forge.

    The main changes of version 1.1-250 compared to 1.1-246 are:

    • Actor covariates do not need to be centered any more. This can be important for use with the effects avSimAltX, totSimAltX and avAltAltX, new in version 1.1-246. Centering is determined by the parameter 'centered' in functions coCovar() and varCovar(). Centering still is the default.
    • Functions Wald.RSiena() and Multipar.RSiena() were added, carrying out Wald-type test after estimation by siena07().
    • User-defined interactions are now handled more adequately in functions setEffect() and updateTheta(), and also when using the prevAns parameter in siena07().
    • A problem in siena07() sometimes (rarely) occurring, giving an error message mentioning 'cvalue', was corrected.
    • Divergent parameters in siena07() get NA for their rows and columns in the covariance matrix, instead of values like 999.
    • A number of changes in revision 244 were ported from RSienaTest to RSiena.



    October 31, 2013: new version (1.1-246) of RSiena and RSienaTest available from R-Forge.

    A new version (1.1-246) of RSiena and RSienaTest can be obtained from the RSiena project at R-Forge.

    The main changes of version 1.1-246 compared to 1.1-245 are:

    • New behavior objective function effects avSimAltX, totSimAltX and avAltAltX to differentiate sources of peer influence in directed networks.
    • Fix of a bug that occurred in the case of on average decreasing behavior variables.



    October 17, 2013: new version (1.1-245) of RSiena and RSienaTest available from R-Forge.

    A new version (1.1-245) of RSiena and RSienaTest can be obtained from the RSiena project at R-Forge.

    The main changes of version 1.1-245 compared to 1.1-243 are (for RSienaTest only, RSiena still is the same):

    • Available again for Mac (RSiena was already available for Mac), thanks to Mark Ortmann and James Hollway.
    • New structural rate effect outRateLog (dependence of rates on a power of (outdegree + 1)).
    • New effect outTrunc2 (duplicate of outTrunc, allows using this effect simultaneously with two effect parameters).
    • Start of the manual reorganized and partially rewritten (with help from Zsófia Boda and Andras Vörös); instructions for siena01Gui() (which is still supported but without a positive advice) now are in the separate document siena01gui.pdf.
    • Larger example for sienaCompositionChange.
    • In siena08(), also Bonferroni combination of the two Fisher combinations now is reported.
    • In phase2 of siena07(), rolled back change in truncation from version 1.1-227 to the earlier procedure.
    • Function descriptives.sienaGOF() was added. This gives the numerical values of the statistics reported graphically in plot.sienaGOF().
    • For ML estimation: added autocorrelations during phase 3 to summary.sienaFit().
    • Minor changes of output in various functions, and in error message for includeEffects. In case results of siena07() are artificial, they are now reported not as variances equal to 999 but as NA.



    September 17, 2013: new version (1.1-243) of RSiena and RSienaTest available from R-Forge.

    A new version (1.1-243) of RSiena and RSienaTest can be obtained from the RSiena project at R-Forge.

    The main changes of version 1.1-243 compared to 1.1-241 are:

    • Improved treatment of structural values in sienaGOF for the case that structural values differ between waves. If your data include changing structural values, please use this new version for sienaGOF.
    • Improved plotting of sienaGOF objects so that observed values outside of the range of simulated values don't run off the chart.
    • New effects: anti isolates, anti in-isolates, anti in-near-isolates. These can be helpful in better modeling isolates and the number of indegrees equal to 0 or 1. (Outdegrees 0 could already be modeled by the outTrunc effect with parameter 1.) This may especially be helpful for improving goodness of fit for modeling bipartite networks.
    • Correction to the Dolby option in siena07 for the case of more than 2 waves. The error did not lead to incorrect results, but to inferior convergence in the case of many waves (such as may occur in the multi-group option).
    • Improved printing of results of siena07 in the case that it is used for simulation without estimation (option simOnly).

    RSiena and RSienaTest are almost the same, except for function sienaBayes(), which is still experimental and undocumented, and included only in RSienaTest.



    August, 2013: Special issue of Journal of Research on Adolescence on Network and Behavior Dynamics in Adolescence.

    A special issue of Journal of Research on Adolescence was published about Network and Behavior Dynamics in Adolescence, almost all papers using RSiena. The editors are René Veenstra, Jan Kornelis Dijkstra, Christian Steglich, and Maarten H. W. Van Zalk.



    August 24, 2013: new version (1.1-241) of RSiena and RSienaTest available from R-Forge.

    A new version (1.1-241) of RSiena and RSienaTest can be obtained from the RSiena project at R-Forge.

    The main changes of version 1.1-241 compared to 1.1-232 are:

    • A bug, occurring sometimes for bipartite networks with only upward changes, was corrected.
    • The errors that occurred when the covariance matrix was singular (because of perfect collinearity) now do not lead any more to an abort of siena07() but are trapped so the function continues to run and ends normally, with warning messages in the output.
    • A few degree-related effects that were available already for one-mode networks now are also available for bipartite networks.
    • For those who look at the values reported for the target statistics: the target statistic for the inPop effect was erroneously multiplied by n, and this is now redressed.
    • A new effect inIsolatePop was added, but this still is not quite OK, so please leave it aside until the next revision.

    RSiena and RSienaTest are almost the same, except for function sienaBayes(), which is still experimental and undocumented, and included only in RSienaTest.



    June 15, 2013: new version (1.1-231) of RSiena and RSienaTest available from R-Forge

    Bugs in sienaGOF repaired.

    • There were some bugs in the auxiliary functions for sienaGOF(), affecting the results for data sets with bipartite networks, structural zeros or ones, and/or missing data. (For missing data, the results were affected only slightly.) These bugs were corrected.
    • For data sets with composition change, Method of Moments estimation now is forced to be non-conditional.


    May 14, 2013: new version (1.1-230) of RSienaTest

    Error in MaxDegree solved; sienaGOF() still contains a bug for structural zeros.

    Version 1.1-230 repairs an error in the operation of MaxDegree (set in sienaAlgorithmCreate / sienaModelCreate, used in siena07). This error existed for some time, and it is unknown for how long this was the case. Nynke Niezink repaired it (thank you!). Everybody who has used the MaxDegree option is advised to rerun the models that used this option, because the outcomes are likely to be wrong. We apologize...


    April 2013: new version (1.1-227) of RSienaTest

    Version 1.1-227 of RSienaTest has been committed to R-Forge.
    Note that R also was updated a few weeks ago to R 3.0.0, so it makes sense to first update R and then RSienaTest.

    A lot of functions and options that were up to now available in RSiena have been ported to RSienaTest.

    • For the added effects see the new version of the manual, p. 174-175. (Some effects are totally new; also effects that were ported from RSienaTest to RSiena are mentioned.)
    The main further changes are:
    • sienaGOF() was changed and now has a basic functionality that does not require the use of other functions. Any dependent variable can be tested. There are out-of-the-box tests for IndegreeDistribution, OutdegreeDistribution, and BehaviorDistribution; and a network extractor function for the network/sna package. To achieve independence from other packages such as sna and igraph that do not belong to the core R distribution, some other functions are outside of RSiena/RSienaTest but given on the help page for "sienaGOF-auxiliary". A network extractor function for package igraph is given there, and auxiliary functions for testing the distribution of geodesic distances and the distribution of Burt's constraint measure. The former is suggested for general use, the latter only as an example and for specific use.

      There is now (as from May 2) a new version of the script sienaGOF_new.R that demonstrates the basic functionality of sienaGOF(). More is to follow.

    • sienaTimeTest() was changed and now also contains effect-wise tests, groupwise tests (for group objects), automatic exclusion of collinear effects, and has improved print and summary methods.
    • The function effectsDocumentation() now also can be used to give documentation for the currently used effects object, with shortName and other characteristics used to define the effects.
    • A bug in the starting values for two-mode networks was corrected.
    • A new option "Dolby" was added to sienaAlgorithmCreate() for variance reduction in siena07(). This will always (perhaps almost always) improve convergence. The default is Dolby=TRUE, which means that you don't have to do anything about it. Also a new option "diagonalize" was added, that is expected to improve convergence when the starting value for the parameter is relatively good already.
    • Many help pages were improved: sometimes to make them better understandable or complete, sometimes to give more appropriate examples.
    • The scripts Rscript02VariableFormat.R, Rscript03SienaRunModel.R, and Rscript04SienaBehaviour.R were updated. The same update was done for these scripts on the Siena website and in the manual. Because of the following item, this means these scripts are not downward compatible with older versions of RSiena and RSienaTest (just because these do not recognize the changed function names).
    • To get function names that are a better indication of the function's purpose, sienaModelCreate() is now called sienaAlgorithmCreate(), but the earlier name is still retained as an alias; the class name of the object created by this function is now called sienaAlgorithm. For the same reason, sienaNet() is now called sienaDependent(), but here also the earlier name is still retained as an alias; the class name of the object created by this function is now sienaDependent.

    RSiena: Siena in R

    Appendix B of the RSiena manual contains a list of changes compared to earlier versions.

    The RSiena users' manual is frequently updated. Likewise, the Siena scripts page is frequently updated with new or improved scripts. The front page of the manual and the header of the scripts report the date of publication.

    Recently added features in RSiena are the possibility to estimate dynamics of multiple interdependent networks, which may be one-mode or two-mode networks; and the possibility to estimate models for networks with valued ties. See the Siena scripts page for examples.

    For multiple networks, the elaboration and explanation are presented in:
    Tom A.B. Snijders, Alessandro Lomi, and Vanina Jasmine Torló (2013). A model for the multiplex dynamics of two-mode and one-mode networks, with an application to employment preference, friendship, and advice. Social Networks, 35, 265-276.
    DOI: http://dx.doi.org/10.1016/j.socnet.2012.05.005.
    For two-mode networks, the article mentioned above gives some explanation, but a more specific explanation and example are given in:

    • Guido Conaldi, Alessandro Lomi and Marco Tonellato (2012). Dynamic models of affiliation and the network structure of problem solving in an open source software project. Organizational Research Methods, 15, 385-412.
      DOI: http://dx.doi.org/10.1177/1094428111430541.
    • Alessandro Lomi, Guido Conaldi, Marco Tonellato (2012). Organized Anarchies and the Network Dynamics of Decision Opportunities in an Open Source Software Project,
      p. 363-397 in Alessandro Lomi, J. Richard Harrison (ed.), The Garbage Can Model of Organizational Choice: Looking Forward at Forty, Research in the Sociology of Organizations, Volume 36, Emerald Group Publishing.
      DOI: http://dx.doi.org/10.1108/S0733-558X(2012)0000036017.
    Important new options in RSiena since April 2010 are the sienaGOF and sienaTimeTest functions contributed by Josh Lospinoso, respectively for goodness of fit checking and for testing time homogeneity for data with 3 or more waves. See the manual for further explanation of these functions.



    Some recent publications (many others are at the applications page)

    • See the publications mentioned above.
    • Jacob E. Cheadle, Michael Stevens, Deadric T. Williams, and Bridget J. Goosby (2013). The differential contributions of teen drinking homophily to new and existing friendships: An empirical assessment of assortative and proximity selection mechanisms. Social Science Research, 42, 1297-1310.
      DOI: http://dx.doi.org/10.1016/j.ssresearch.2013.05.001.

      Abstract
      Alcohol use is pervasive in adolescence. Though most research is concerned with how friends influence drinking, alcohol is also important for connecting teens to one another. Prior studies have not distinguished between new friendship creation, and existing friendship durability, however. We argue that accounting for distinctions in creation-durability processes is critical for understanding the selection mechanisms drawing drinkers into homophilous friendships, and the social integration that results. In order to address these issues, we applied stochastic actor based models of network dynamics to National Longitudinal Study of Adolescent Health data. Adolescents only modestly prefer new friendships with others who drinker similarly, but greatly prefer friends who indirectly connect them to homophilous drinkers. These indirect homophilous drinker relationships are shorter lived, however, and suggest that drinking is a social focus that connects adolescents via proximity, rather than assortativity. These findings suggest that drinking leads to more situational and superficial social integration.

      Note
      This paper differentiates between creation and maintenance (endowment effect; called durability here) of ties and between homophily at distances 1 and 2. For both, important results are found.

    • Special issue Journal of Research on Adolescence, 23, Issue 3, pages 399-603:
      Network and Behavior Dynamics in Adolescence.

      Introduction:
      René Veenstra, Jan Kornelis Dijkstra, Christian Steglich and Maarten H. W. Van Zalk (2013). Network-Behavior Dynamics. Journal of Research on Adolescence, 23, 399-412.
      DOI: http://dx.doi.org/10.1111/jora.12070.

    • Mark Lubell, John Scholz, Ramiro Berardo, and Garry Robins (2012). Testing Policy Theory with Statistical Models of Networks. Policy Studies Journal 40, 351-374.
      DOI: http://dx.doi.org/10.1111/j.1541-0072.2012.00457.x.
    • Garry Robins, Jenny M. Lewis, and Peng Wang (2012). Statistical Network Analysis for Analyzing Policy Networks. Policy Studies Journal 40, 375-401.
      DOI: http://dx.doi.org/10.1111/j.1541-0072.2012.00458.x.
    • Tom A.B. Snijders. (2011). Statistical Models for Social Networks. Annual Review of Sociology, 37, 129-151.
      DOI: http://dx.doi.org/10.1146/annurev.soc.012809.102709.
    • Veenstra, R., and Steglich, C. (2012). Actor-based model for network and behavior dynamics: A tool to examine selection and influence processes. Chapter 34 (pp. 598-618) in B. Laursen, T. D. Little, and N. A. Card (Eds.), Handbook of developmental research methods. New York: Guilford Press.
    • Social Networks has a special issue in two parts, issue 32.1 (2010) and 34.3 (2012), devoted to Network Dynamics. This carries a tutorial and several applications of Siena.
    • Kevin Lewis, Marco Gonzalez, and Jason Kaufman (2012). Social selection and peer influence in an online social network. PNAS 109, 68-72.
      DOI: http://dx.doi.org/10.1073/pnas.1109739109.

      Abstract
      Disentangling the effects of selection and influence is one of social science's greatest unsolved puzzles: Do people befriend others who are similar to them, or do they become more similar to their friends over time? Recent advances in stochastic actor-based modeling, combined with self-reported data on a popular online social network site, allow us to address this question with a greater degree of precision than has heretofore been possible. Using data on the Facebook activity of a cohort of college students over 4 years, we find that students who share certain tastes in music and in movies, but not in books, are significantly likely to befriend one another. Meanwhile, we find little evidence for the diffusion of tastes among Facebook friends - except for tastes in classical/jazz music. These findings shed light on the mechanisms responsible for observed network homogeneity; provide a statistically rigorous assessment of the coevolution of cultural tastes and social relationships; and suggest important qualifications to our understanding of both homophily and contagion as generic social processes.
      DOI: http://dx.doi.org/10.1073/pnas.1109739109.

    Further publications are listed at the webpage with literature and the webpage with further applications.




    Back to the main Siena page