Other clubs

---

We also built increasingly complex models for the remaining 4 clubs. We started by modeling network structure and dynamics in kudos tie formation, using a rather simple model specification. If this did not produce adequate GOF, we assessed how the model may be extended, by a theory-guided exploration. We performed a forward model-selection procedure and use score-type tests to see which effect could improve the model. We use sienaGOF to estimate how the fit (based on Mahalanobis distance) would increase had we included the fixed-and-tested effects (hence, circumventing the need for re-estimating the model with the new effect). This resulted in slightly different model-specification for kudos-tie formation dynamics for each club.

If adequate GOF was produced, we started modeling running activity dynamics, and investigated how well our model was able to reproduce the distribution of running behavior values of our panel data.

Down below, we create the final effect-objects for each club. Note: for subsequent meta-analysis, we needed to ensure that the model specification for all our club-networks is identical. In some clubs, some effects were rather important; in clubs where this was not the case, we fixed these to 0.

---

Getting started

clean up

rm (list = ls( ))

read in data

load("clubdata_rsiena_freq.Rdata")

necessary packages

• RSiena

library(RSiena)

Effect objects

We make, for each club, 6 model specifications:

• Model 1: base model + indegree effect on running
• Model 2: Model 1 + average alter effect
• Model 3: Model 1 + average attraction higher
• Model 4: Model 1 + average attraction lower
• Model 5: Model 1 + average attraction higher + lower (main model used in the manuscript)
• Model 6: Model 1 + average similarity effect

We save these model specifications in a list for each club.

Club 2

myeff <- getEffects(clubdata_rsiena_freq[[2]])

# base model
myeff <- includeEffects(myeff, gwespFF, name = "kudonet")
myeff <- includeEffects(myeff, outActSqrt, inPopSqrt, name = "kudonet")
myeff <- includeEffects(myeff, outIso, name = "kudonet")
myeff <- setEffect(myeff, higher, name = "kudonet", interaction1 = "freq_run")
myeff <- includeEffects(myeff, egoX, altX, sameX, name = "kudonet", interaction1 = "gender")
myeff <- includeInteraction(myeff, recip, gwespFF, parameter = 69, name = "kudonet")
myeff <- includeEffects(myeff, reciAct, name = "kudonet")
myeff <- setEffect(myeff, outPopSqrt, name = "kudonet", fix = TRUE, test = FALSE, initialValue = 0)
myeff <- includeEffects(myeff, effFrom, name = "freq_run", interaction1 = "freq_other")
myeff <- includeEffects(myeff, effFrom, name = "freq_run", interaction1 = "gender")

# models 1-6
myeff1 <- includeEffects(myeff, indeg, name = "freq_run", interaction1 = "kudonet")  # model 1: indegree
myeff2 <- includeEffects(myeff1, avAlt, name = "freq_run", interaction1 = "kudonet")  # model 2: avAlt
myeff3 <- includeEffects(myeff1, avAttHigher, name = "freq_run", interaction1 = "kudonet")  # model 3: avAttHigher
myeff4 <- includeEffects(myeff1, avAttLower, name = "freq_run", interaction1 = "kudonet")  # model 4: avAttLower
myeff5 <- includeEffects(myeff3, avAttLower, name = "freq_run", interaction1 = "kudonet")  # model 5: avAttHigher+Lower
myeff6 <- includeEffects(myeff1, avSim, name = "freq_run", interaction1 = "kudonet")  # model 6: avSim

Club 3

myeff <- getEffects(clubdata_rsiena_freq[[3]])

# base model
myeff <- includeEffects(myeff, gwespFF, name = "kudonet")
myeff <- includeEffects(myeff, outActSqrt, inPopSqrt, name = "kudonet")
myeff <- includeEffects(myeff, outIso, name = "kudonet")
myeff <- setEffect(myeff, higher, name = "kudonet", interaction1 = "freq_run")
myeff <- includeEffects(myeff, egoX, altX, sameX, name = "kudonet", interaction1 = "gender")
myeff <- includeInteraction(myeff, recip, gwespFF, parameter = 69, name = "kudonet")
myeff <- includeEffects(myeff, outPopSqrt, name = "kudonet")
myeff <- setEffect(myeff, reciAct, name = "kudonet", fix = TRUE, test = FALSE, initialValue = 0)
myeff <- includeEffects(myeff, effFrom, name = "freq_run", interaction1 = "freq_other")
myeff <- includeEffects(myeff, effFrom, name = "freq_run", interaction1 = "gender")

# models 1-6
myeff1 <- includeEffects(myeff, indeg, name = "freq_run", interaction1 = "kudonet")  # model 1: indegree
myeff2 <- includeEffects(myeff1, avAlt, name = "freq_run", interaction1 = "kudonet")  # model 2: avAlt
myeff3 <- includeEffects(myeff1, avAttHigher, name = "freq_run", interaction1 = "kudonet")  # model 3: avAttHigher
myeff4 <- includeEffects(myeff1, avAttLower, name = "freq_run", interaction1 = "kudonet")  # model 4: avAttLower
myeff5 <- includeEffects(myeff3, avAttLower, name = "freq_run", interaction1 = "kudonet")  # model 5: avAttHigher+Lower
myeff6 <- includeEffects(myeff1, avSim, name = "freq_run", interaction1 = "kudonet")  # model 6: avSim

Club 4

myeff <- getEffects(clubdata_rsiena_freq[[4]])

# base model
myeff <- includeEffects(myeff, gwespFF, name = "kudonet")
myeff <- includeEffects(myeff, outActSqrt, inPopSqrt, name = "kudonet")
myeff <- includeEffects(myeff, outIso, name = "kudonet")
myeff <- setEffect(myeff, higher, name = "kudonet", interaction1 = "freq_run")
myeff <- setEffect(myeff, egoX, name = "kudonet", interaction1 = "gender", fix = TRUE, test = FALSE,
    initialValue = 0)
myeff <- setEffect(myeff, altX, name = "kudonet", interaction1 = "gender", fix = TRUE, test = FALSE,
    initialValue = 0)
myeff <- setEffect(myeff, sameX, name = "kudonet", interaction1 = "gender", fix = TRUE, test = FALSE,
    initialValue = 0)
myeff <- setEffect(myeff, outPopSqrt, name = "kudonet", fix = TRUE, test = FALSE, initialValue = 0)
myeff <- setEffect(myeff, reciAct, name = "kudonet", fix = TRUE, test = FALSE, initialValue = 0)
(myeff <- includeInteraction(myeff, recip, gwespFF, parameter = 69, name = "kudonet"))
(eff1 <- myeff[myeff$include, ]$effect1[24])
(eff2 <- myeff[myeff$include, ]$effect2[24])
myeff <- setEffect(myeff, unspInt, fix = TRUE, test = FALSE, effect1 = eff1, effect2 = eff2)
myeff <- includeEffects(myeff, effFrom, name = "freq_run", interaction1 = "freq_other")
myeff <- setEffect(myeff, effFrom, name = "freq_run", interaction1 = "gender", fix = T, test = F, initialValue = 0)

# models 1-6
myeff1 <- includeEffects(myeff, indeg, name = "freq_run", interaction1 = "kudonet")  # model 1: indegree
myeff2 <- includeEffects(myeff1, avAlt, name = "freq_run", interaction1 = "kudonet")  # model 2: avAlt
myeff3 <- includeEffects(myeff1, avAttHigher, name = "freq_run", interaction1 = "kudonet")  # model 3: avAttHigher
myeff4 <- includeEffects(myeff1, avAttLower, name = "freq_run", interaction1 = "kudonet")  # model 4: avAttLower
myeff5 <- includeEffects(myeff3, avAttLower, name = "freq_run", interaction1 = "kudonet")  # model 5: avAttHigher+Lower
myeff6 <- includeEffects(myeff1, avSim, name = "freq_run", interaction1 = "kudonet")  # model 6: avSim

Club 5

myeff <- getEffects(clubdata_rsiena_freq[[5]])

# base model
myeff <- includeEffects(myeff, gwespFF, name = "kudonet")
myeff <- includeEffects(myeff, outActSqrt, inPopSqrt, name = "kudonet")
myeff <- includeEffects(myeff, outIso, name = "kudonet")
myeff <- setEffect(myeff, higher, name = "kudonet", interaction1 = "freq_run")
myeff <- includeEffects(myeff, egoX, altX, sameX, name = "kudonet", interaction1 = "gender")
myeff <- includeInteraction(myeff, recip, gwespFF, parameter = 69, name = "kudonet")
myeff <- setEffect(myeff, outPopSqrt, name = "kudonet", fix = TRUE, test = FALSE, initialValue = 0)
myeff <- setEffect(myeff, reciAct, name = "kudonet", fix = TRUE, test = FALSE, initialValue = 0)
myeff <- includeEffects(myeff, effFrom, name = "freq_run", interaction1 = "freq_other")
myeff <- includeEffects(myeff, effFrom, name = "freq_run", interaction1 = "gender")

# models 1-6
myeff1 <- includeEffects(myeff, indeg, name = "freq_run", interaction1 = "kudonet")  # model 1: indegree
myeff2 <- includeEffects(myeff1, avAlt, name = "freq_run", interaction1 = "kudonet")  # model 2: avAlt
myeff3 <- includeEffects(myeff1, avAttHigher, name = "freq_run", interaction1 = "kudonet")  # model 3: avAttHigher
myeff4 <- includeEffects(myeff1, avAttLower, name = "freq_run", interaction1 = "kudonet")  # model 4: avAttLower
myeff5 <- includeEffects(myeff3, avAttLower, name = "freq_run", interaction1 = "kudonet")  # model 5: avAttHigher+Lower
myeff6 <- includeEffects(myeff1, avSim, name = "freq_run", interaction1 = "kudonet")  # model 6: avSim

---

Estimation

With the following script, we estimate, for each club i (2-5), all models j (1-6). We rerun the models until adequate convergence is reached. We store the sienaFit objects in a list, which we save later on.

for (i in 2:c) { # for every club

  # we take the rsiena object
  mydata <- clubdata_rsiena_freq[[i]]
  
  # and the list containing myeff objects
  load(file=paste("test", "/", "myeff", "/", "myeff_club", i, ".RData", sep = "")) 
  
  # we make a list for storing the RSiena fit objects
  sienaFit <- list()

  # for club i we run models j in 1:m
  for (j in 1:m) {
 
    # we estimate the model
    try <- 1
    print(paste("Estimating model ", j, " for club ", i, sep=""))
    sienaFit[[j]] <- siena07(myalgorithm, data = mydata, effects = myeff[[j]], returnDeps=TRUE,
                             useCluster=TRUE, nbrNodes=10, initC=TRUE, batch=TRUE) # store it in the list
    
    # re-run until we reach adequate convergence 
    while (TRUE){
      if(sienaFit[[j]]$tconv.max >= .25){
        try <- try + 1
        if (try>30) { # with at max 30 runs.
          print(paste("Now it lasted to long!") 
          break      
        }
        print(paste("Model did not converge adequately (", sienaFit[[j]]$tconv.max, "); ", "Repeat the estimation (", "try ", try, ")", sep = ""))
        sienaFit[[j]] <- siena07( myalgorithm, data = mydata, effects = myeff[[j]], prevAns= sienaFit[[j]], returnDeps=TRUE, useCluster=TRUE, nbrNodes=10, initC=TRUE, batch=TRUE)
      }else{
        print(paste("Reached overall maximum convergence ratio of: ", sienaFit[[j]]$tconv.max, sep = ""))
        print("")
        break
      }
    }
    
  }
  # and save the list with RSiena fit objects
  save(sienaFit, file=paste("test", "/", "sienaFit", "/", "sienaFit_club", i, ".RData", sep = ""))
  print(paste("All models are estimated for club ", i, ". Model results are stored in sienaFit_club", i, ".RData", sep=""))
  print("")
  ifelse(i<c, print(paste("Continuing with club ", i+1, sep="")), print("Estimation finished!"))
  
}

sienaFit_clubL <- list()

for (i in 1:5) {
  temp.space <- new.env()
  bar <- load(paste("test/sienaFit/sienaFit_club", i, ".RData", sep=""), temp.space)
  sienaFit_clubL[[i]] <- get(bar, temp.space)
  rm(temp.space)
}

lapply(sienaFit_clubL, '[[', 5)
map(sienaFit_clubL, 6)

---

Copyright © 2021 Rob Franken