BasicRSiena

In RSiena: Siena - Simulation Investigation for Empirical Network Analysis

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>"
)

An example of a basic sequence of commands for estimating a basic model by function siena07() of RSiena. With a lot of use of help pages; this can be skipped as you like. Note that lines starting with # are comment lines, not commands.

What is your current working directory?

getwd()

If you wish it to be different, change it by

# setwd()

In my case:

# setwd("C:\\Users\\tom.snijders\\Documents\\Siena\\s50_script")

Note the double backslashes used for R.

Define data sets

If you have internet access, you can download the data from the Siena website ("Data sets" tab) http://www.stats.ox.ac.uk/~snijders/siena/s50_data.zip and unzip it in your working directory. The data description is at http://www.stats.ox.ac.uk/~snijders/siena/s50_data.htm

Then you can read the data files by the commands (this can be replaced by using the internal data set, see below)

# friend.data.w1 <- as.matrix(read.table("s50-network1.dat"))
# friend.data.w2 <- as.matrix(read.table("s50-network2.dat"))
# friend.data.w3 <- as.matrix(read.table("s50-network3.dat"))
# drink <- as.matrix(read.table("s50-alcohol.dat"))
# smoke <- as.matrix(read.table("s50-smoke.dat"))

But without internet access, the data can be obtained from within RSiena (see below), because this is an internal data set.

library(RSiena)
# Now we use the internally available s50 data set.
# Look at its description:
?s50
# 3 waves, 50 actors
# Look at the start and end of the first wave matrix
head(s501)
tail(s501)
# and at the alcohol variable
s50a
# Now define the objects with the same names as above
# (this step is superfluous if you read the data already).
        friend.data.w1 <- s501
        friend.data.w2 <- s502
        friend.data.w3 <- s503
        drink <- s50a
        smoke <- s50s

Now the data must be given the specific roles of variables in an RSiena analysis.

Dependent variable

?sienaDependent
# First create a 50 * 50 * 3 array composed of the 3 adjacency matrices
friendshipData <- array( c( friend.data.w1, friend.data.w2, friend.data.w3 ),
           dim = c( 50, 50, 3 ) )
# and next give this the role of the dependent variable:
    friendship <- sienaDependent(friendshipData)

What did we construct?

friendship

We also must prepare the objects that will be the explanatory variables.

Actor covariates

We use smoking for wave 1 as a constant actor covariate:

smoke1 <- coCovar( smoke[ , 1 ] )
# A variable actor covariate is defined for drinking:
alcohol <- varCovar( drink )
# (This choice is purely for the purpose of illustration here.)

Put the variables together in the data set for analysis

?sienaDataCreate
mydata <- sienaDataCreate( friendship, smoke1, alcohol )
# Check what we have
mydata

You can get an outline of the data set with some basic descriptives from

print01Report( mydata, modelname="s50")
# For the model specification we need to create the effects object
myeff <- getEffects( mydata )
# All the effects that are available given the structure
# of this data set can be seen from
effectsDocumentation(myeff)
# For a precise description of all effects, see Chapter 12 in the RSiena manual.
# A basic specification of the structural effects:
?includeEffects
myeff <- includeEffects( myeff, transTrip, cycle3)
# and some covariate effects:
myeff <- includeEffects( myeff, egoX, altX, simX, interaction1 = "alcohol" )
myeff <- includeEffects( myeff, simX, interaction1 = "smoke1" )
myeff

Create object with algorithm settings Accept defaults but specify name for output file (which you may replace by any name you prefer)

?sienaAlgorithmCreate
myalgorithm <- sienaAlgorithmCreate( projname = 's50' )

Estimate parameters

?siena07
ans <- siena07( myalgorithm, data = mydata, effects = myeff)
ans

This gives results from a random starting point. To use a fixed starting point, use the "seed" parameter:

# myalgorithm <- sienaAlgorithmCreate( projname = 's50', seed=435123 )

For checking convergence, look at the 'Overall maximum convergence ratio' mentioned under the parameter estimates.

It can also be shown by requesting r ans$tconv.max

If this is less than 0.25, convergence is good. If convergence is inadequate, estimate once more, using the result obtained as the "previous answer" from which estimation continues:

ans <- siena07( myalgorithm, data = mydata, effects = myeff, prevAns=ans)
ans
# If convergence is good, you can look at the estimates.
# More extensive results
summary(ans)

Still more extensive results are given in the output file s50.out in the current directory.

Note that by putting an R command between parentheses (....), the result will also be printed to the screen. Next add the transitive reciprocated triplets effect, an interaction between transitive triplets and reciprocity,

(myeff <- includeEffects( myeff, transRecTrip))
(ans1 <- siena07( myalgorithm, data = mydata, effects = myeff, prevAns=ans))
# If necessary, repeat the estimation with the new result:
(ans1 <- siena07( myalgorithm, data = mydata, effects = myeff, prevAns=ans1))

This might still not have an overall maximum convergence ratio less than 0.25. If not, you could go on once more.

Inspect the file s50.txt in your working directory and understand the meaning of its contents.

To have a joint test of the three effects of alcohol:

?Multipar.RSiena
Multipar.RSiena(ans1, 7:9)

Focusing on alcohol similarity, the effect is significant; diluting the effects of alcohol by also considering ego and alter, the three effects simultaneously are not significant.

Assignment 1

1a.

Drop the effect of smoke1 similarity and estimate the model again. Do this by the function setEffects() using the <> parameter. Give the changed effects object and the new answer object new names, such as effects1 and ans1, to distinguish them.

1b.

Change the three effects of alcohol to the single effect of alcohol similarity, and estimate again.

Networks and behavior study

Now we redefine the role of alcohol drinking as a dependent behaviour variable.

# Once again, look at the help file
?sienaDependent
# now paying special attention to the <<type>> parameter.
drinking <- sienaDependent( drink, type = "behavior" )

Put the variables together in the data set for analysis

NBdata <- sienaDataCreate( friendship, smoke1, drinking )
NBdata
NBeff <- getEffects( NBdata )
effectsDocumentation(NBeff)
NBeff <- includeEffects( NBeff, transTrip, transRecTrip )
NBeff <- includeEffects( NBeff, egoX, egoSqX, altX, altSqX, diffSqX,
                         interaction1 = "drinking" )
NBeff <- includeEffects( NBeff, egoX, altX, simX, interaction1 = "smoke1" )
NBeff
# For including effects also for the dependent behaviour variable, see
?includeEffects
NBeff <- includeEffects( NBeff, avAlt, name="drinking",
                         interaction1 = "friendship" )
NBeff
# Define an algorithm with a new project name
myalgorithm1 <- sienaAlgorithmCreate( projname = 's50_NB' )

# Estimate again, using the second algorithm right from the start.
NBans <- siena07( myalgorithm1, data = NBdata, effects = NBeff)
# You may improve convergence (considering the overall maximum
# convergence ratio) by repeated estimation in the same way as above.

# Look at results
NBans
# Make a nicer listing of the results
siena.table(NBans, type="html", sig=TRUE)

This produces an html file; siena.table can also produce a LaTeX file.

Assignment 2

2a.

Replace the average alter effect by average similarity (avSim) or total similarity (totSim) and estimate the model again.

2b.

Add the effect of smoking on drinking and estimate again.

Assignment 3

Read Sections 13.3 and 13.4 of the Siena Manual, download scripts SelectionTables.r and InfluenceTables.r from the Siena website, and make plots of the selection table and influence table for drinking.

RSiena documentation built on Sept. 24, 2020, 3 p.m.