Wald function

Wald and score tests for RSiena results

These functions test parameters in RSiena results estimated by siena07. Tests can be Wald-type (if the parameters were estimated) or score-type tests (if the parameters were fixed and tested).

Wald.RSiena(A, ans)

Multipar.RSiena(ans, ...)

score.Test(ans, test=ans$test)

testSame.RSiena(ans, e1, e2)

Arguments

• A: A k * p matrix, where p = ans$pp, the number of parameters in ans excluding the basic rate parameters used for conditional estimation.
• ans: An object of class sienaFit, resulting from a call to siena07.
• ...: One or more integer numbers between 1 and p, specifying the tested effects (numbered as in print(ans); if conditional estimation was used, numbered as the 'Other parameters').
• test: One or more integer numbers between 1 and p, or a logical vector of length p; these should specify the tested effects (numbered as described for the ...).
• e1,e2: Each an integer number between 1 and p, or a vector of such numbers; the hypothesis tested is that the parameters for effects with number/s e1 are equal to those in e2.

Details

Wald.RSiena produces a Wald-type test, applicable to estimated parameters. Multipar.RSiena and testSame.RSiena are special cases of Wald.RSiena. The hypothesis tested by Wald.RSiena

is $A theta = 0$, where $theta$ is the parameter estimated in the process leading to ans.

The hypothesis tested by Multipar.RSiena is that all parameters given in $...$ are 0.

The hypothesis tested by testSame.RSiena is that all parameters given in e1 are equal to those in e2.

score.Test produces a score-type test. The tested effects for score.Test should have been specified in includeEffects or setEffect with fix=TRUE, test=TRUE, i.e., they should not have been estimated. The hypothesis tested by score.Test is that the tested parameters have the value indicated in the effects object used for obtaining ans.

These tests should be carried out only when convergence is adequate (overall maximum convergence ratio less than 0.25 and all t-ratios for convergence less than 0.1 in absolute value).

These functions have their own print method, see print.sienaTest.

Returns

An object of class sienaTest, which is a list with elements: - chisquare:: The test statistic, assumed to have a chi-squared null distribution.

• df:: The degrees of freedom.

• pvalue:: The associated p-value.

• onesided:: For df=1, the onesided test statistic.

• efnames:: For Multipar.RSiena and score.Test, the names of the tested effects.

References

See the manual and https://www.stats.ox.ac.uk/~snijders/siena/

M. Schweinberger (2012). Statistical modeling of network panel data: Goodness-of-fit. British Journal of Statistical and Mathematical Psychology 65 , 263--281.

Author(s)

Tom Snijders

See Also

siena07, print.sienaTest

Examples

mynet <- sienaDependent(array(c(s501, s502), dim=c(50, 50, 2)))
mydata <- sienaDataCreate(mynet)
myeff <- getEffects(mydata)
myalgorithm <- sienaAlgorithmCreate(nsub=1, n3=40, seed=1777, projname=NULL)
# nsub=1 and n3=40 is used here for having a brief computation,
# not for practice.
myeff <- includeEffects(myeff, transTrip, transTies)
myeff <- includeEffects(myeff, outAct, outPop, fix=TRUE, test=TRUE)
(ans <- siena07(myalgorithm, data=mydata, effects=myeff, batch=TRUE))
A <- matrix(0, 2, 6)
A[1, 3] <- 1
A[2, 4] <- 1
wa <- Wald.RSiena(A, ans)
wa
# A shortcut for the above is:
Multipar.RSiena(ans, 3, 4)
# The following two are equivalent:
sct <- score.Test(ans, c(FALSE, FALSE, FALSE, FALSE, FALSE, TRUE))
sct <- score.Test(ans,6)
print(sct)
# Getting all 1-df score tests separately:
for (i in which(ans$test)){
   sct <- score.Test(ans,i)
   print(sct)}
# Testing that endowment and creation effects are identical:
myeff1 <- getEffects(mydata)
myeff1 <- includeEffects(myeff1, recip, include=FALSE)
myeff1 <- includeEffects(myeff1, recip, type='creation')
(myeff1 <- includeEffects(myeff1, recip, type='endow'))
(ans1 <- siena07(myalgorithm, data=mydata, effects=myeff1, batch=TRUE))
testSame.RSiena(ans1, 2, 3)