tests/testthat/test-correctness_multilevel.R
In mmeierer/REndo: Fitting Linear Models with Endogenous Regressors using Latent Instrumental Variables
# Required data ----------------------------------------------------------------------------------------------------------------------
data("dataMultilevelIV")
all.L3.models <- c("REF", "FE_L2", "FE_L3", "GMM_L2", "GMM_L3")
all.L2.models <- c("REF", "FE_L2", "GMM_L2")
# Formula transformations ------------------------------------------------------------------------------------------------------------
context("Correctness - multilevelIV - Formula transformations")
test_that("Transformations are correct for L2", {
skip_on_cran()
expect_silent(correct.res <- multilevelIV(formula = y ~ X11 + X12 + X13 + X14 + X15 + X21 + X22 + X23 + X24 +
X31 + X32 + X33 + (1+X11 | SID) | endo(X15),
data = dataMultilevelIV, verbose = FALSE))
# Can handle transformations in LHS
data.altered <- dataMultilevelIV
data.altered$y <- exp(data.altered$y)
expect_silent(res.trans.lhs <- multilevelIV(formula = log(y) ~ X11 + X12 + X13 + X14 + X15 + X21 + X22 + X23 + X24 +
X31 + X32 + X33 + (1+X11 | SID) | endo(X15),
data = data.altered, verbose = FALSE))
expect_equal(coef(res.trans.lhs), coef(correct.res))
for(m in all.L2.models)
expect_equal(coef(summary(res.trans.lhs, model = m)), coef(summary(correct.res, model = m)))
# Can handle transformations in exo
data.altered <- dataMultilevelIV
data.altered$X23 <- exp(data.altered$X23)
expect_silent(res.trans.exo <- multilevelIV(formula = y ~ X11 + X12 + X13 + X14 + X15 + X21 + X22 + log(X23) + X24 +
X31 + X32 + X33 + (1+X11 | SID) | endo(X15),
data = data.altered, verbose = FALSE))
expect_equal(coef(res.trans.exo), coef(correct.res), check.attributes = FALSE)
for(m in all.L2.models)
expect_equal(coef(summary(res.trans.exo, model = m)), coef(summary(correct.res, model = m)), check.attributes = FALSE)
# Can handle transformations in endo
data.altered <- dataMultilevelIV
data.altered$X15 <- exp(data.altered$X15)
expect_silent(res.trans.endo <- multilevelIV(formula = y ~ X11 + X12 + X13 + X14 + log(X15) + X21 + X22 + X23 + X24 +
X31 + X32 + X33 + (1+X11 | SID) | endo(log(X15)),
data = data.altered, verbose = FALSE))
expect_equal(coef(res.trans.endo), coef(correct.res), check.attributes = FALSE)
for(m in all.L2.models)
expect_equal(coef(summary(res.trans.endo, model = m)), coef(summary(correct.res, model = m)), check.attributes = FALSE)
# Can handle transformations in slope
data.altered <- dataMultilevelIV
data.altered$X11 <- exp(data.altered$X11)
expect_silent(res.trans.slope <- multilevelIV(formula = y ~ log(X11) + X12 + X13 + X14 + X15 + X21 + X22 + X23 + X24 +
X31 + X32 + X33 + (1+log(X11) | SID) | endo(X15),
data = data.altered, verbose = FALSE))
expect_equal(coef(res.trans.slope), coef(correct.res), check.attributes = FALSE)
for(m in all.L2.models)
expect_equal(coef(summary(res.trans.slope, model = m)), coef(summary(correct.res, model = m)), check.attributes = FALSE)
})
test_that("Transformations are correct for L3", {
skip_on_cran()
expect_silent(correct.res <- multilevelIV(formula = y ~ X11 + X12 + X13 + X14 + X15 + X21 + X22 + X23 + X24 +
X31 + X32 + X33 + (1+X11 | CID) + (1 | SID) | endo(X15),
data = dataMultilevelIV, verbose = FALSE))
# Can handle transformations in LHS
data.altered <- dataMultilevelIV
data.altered$y <- (data.altered$y+99)/2
expect_silent(res.trans.lhs <- multilevelIV(formula = I((y*2)-99) ~ X11 + X12 + X13 + X14 + X15 + X21 + X22 + X23 + X24 +
X31 + X32 + X33 + (1+X11 | CID) + (1 | SID) | endo(X15),
data = data.altered, verbose = FALSE))
expect_equal(coef(res.trans.lhs), coef(correct.res), check.attributes = FALSE)
for(m in all.L3.models)
expect_equal(coef(summary(res.trans.lhs, model = m)), coef(summary(correct.res, model = m)), check.attributes = FALSE)
# Can handle transformations in exo
data.altered <- dataMultilevelIV
data.altered$X23 <- (data.altered$X23+99)/2
expect_silent(res.trans.exo <- multilevelIV(formula = y ~ X11 + X12 + X13 + X14 + X15 + X21 + X22 + I((X23*2)-99) + X24 +
X31 + X32 + X33 + (1+X11 | CID) + (1 | SID) | endo(X15),
data = data.altered, verbose = FALSE))
expect_equal(coef(res.trans.exo), coef(correct.res), check.attributes = FALSE)
for(m in all.L3.models){
expect_equal(coef(summary(res.trans.exo, model = m)), coef(summary(correct.res, model = m)), check.attributes = FALSE)
}
# Can handle transformations in endo
data.altered <- dataMultilevelIV
data.altered$X15 <- (data.altered$X15+99)/2
expect_silent(res.trans.endo <- multilevelIV(formula = y ~ X11 + X12 + X13 + X14 + I((X15*2)-99) + X21 + X22 + X23 + X24 +
X31 + X32 + X33 + (1+X11 | CID) + (1 | SID) | endo(I((X15*2)-99)),
data = data.altered, verbose = FALSE))
expect_equal(coef(res.trans.endo), coef(correct.res), check.attributes = FALSE)
for(m in all.L3.models){
expect_equal(coef(summary(res.trans.endo, model = m)), coef(summary(correct.res, model = m)), check.attributes = FALSE)
}
# Can handle transformations in slope
data.altered <- dataMultilevelIV
data.altered$X11 <- (data.altered$X11+99)/2
expect_silent(res.trans.slope <- multilevelIV(formula = y ~ I((X11*2)-99) + X12 + X13 + X14 + X15 + X21 + X22 + X23 + X24 +
X31 + X32 + X33 + (1+I((X11*2)-99) | CID) + (1 | SID) | endo(X15),
data = data.altered, verbose = FALSE))
expect_equal(coef(res.trans.slope), coef(correct.res), check.attributes = FALSE)
for(m in all.L3.models)
expect_equal(coef(summary(res.trans.slope, model = m)), coef(summary(correct.res, model = m)), check.attributes = FALSE)
})
# Data sorting ------------------------------------------------------------------------------------------------
context("Correctness - multilevelIV - Data sorting")
# **correct if levels are given separately: (A|SID) + (B|SID) == (A+B|SID)
# ** Also check for W and V?
test_that("Unsorted data is correct L2", {
skip_on_cran()
# Correct = coefs + sorting of residuals / fitted
# Distinguishable non-standard rownames
rownames(dataMultilevelIV) <- as.character(seq(from=nrow(dataMultilevelIV)+100000, to=1+100000))
expect_silent(res.sorted <- multilevelIV(formula = y ~ X11 + X12 + X13 + X14 + X15 + X21 + X22 + X23 + X24 +
X31 + X32 + X33 + (1+X11 | SID) | endo(X15),
data = dataMultilevelIV, verbose = FALSE))
# Can handle transformations in LHS
data.altered <- dataMultilevelIV
data.altered <- data.altered[sample(x=nrow(dataMultilevelIV), size = nrow(dataMultilevelIV), replace = FALSE), ]
expect_silent(res.unsorted <- multilevelIV(formula = y ~ X11 + X12 + X13 + X14 + X15 + X21 + X22 + X23 + X24 +
X31 + X32 + X33 + (1+X11 | SID) | endo(X15),
data = data.altered, verbose = FALSE))
# Coefs + summary statistics the same
expect_equal(coef(res.unsorted), coef(res.sorted))
for(m in all.L2.models){
expect_equal(coef(summary(res.unsorted, model = m)), coef(summary(res.sorted, model = m)))
# Sorting of fitted / residuals same as input
expect_equal(names(fitted(res.sorted, model = m)), rownames(dataMultilevelIV))
expect_equal(names(resid(res.sorted, model = m)), rownames(dataMultilevelIV))
expect_equal(names(fitted(res.unsorted, model = m)), rownames(data.altered))
expect_equal(names(resid(res.unsorted, model = m)), rownames(data.altered))
}
})
test_that("Unsorted data is correct L3", {
skip_on_cran()
# Correct = coefs + sorting of residuals / fitted
# Distinguishable non-standard rownames
rownames(dataMultilevelIV) <- as.character(seq(from=nrow(dataMultilevelIV)+100000, to=1+100000))
expect_silent(res.sorted <- multilevelIV(formula = y ~ X11 + X12 + X13 + X14 + X15 + X21 + X22 + X23 + X24 +
X31 + X32 + X33 + (1|CID)+(1 | SID) | endo(X15),
data = dataMultilevelIV, verbose = FALSE))
# Can handle transformations in LHS
data.altered <- dataMultilevelIV
data.altered <- data.altered[sample.int(n=nrow(dataMultilevelIV), replace = FALSE), ]
expect_silent(res.unsorted <- multilevelIV(formula = y ~ X11 + X12 + X13 + X14 + X15 + X21 + X22 + X23 + X24 +
X31 + X32 + X33 + (1|CID)+(1 | SID) | endo(X15),
data = data.altered, verbose = FALSE))
# Coefs + summary statistics the same
expect_equal(coef(res.unsorted), coef(res.sorted))
for(m in all.L3.models){
expect_equal(coef(summary(res.unsorted, model = m)), coef(summary(res.sorted, model = m)))
# Sorting of fitted / residuals same as input
expect_equal(names(fitted(res.sorted, model = m)), rownames(dataMultilevelIV))
expect_equal(names(resid(res.sorted, model = m)), rownames(dataMultilevelIV))
expect_equal(names(fitted(res.unsorted, model = m)), rownames(data.altered))
expect_equal(names(resid(res.unsorted, model = m)), rownames(data.altered))
}
})
# Reproduce results ------------------------------------------------------------------------------
context("Correctness - multilevelIV - Reproduce results")
test_that("REF is same as lmer()", {
# Have to use the same optimizer controls for lmer as used in multilevelIV
skip_on_cran()
# L2
expect_silent(res.ml2 <- multilevelIV(formula = y ~ X11 + X12 + X13 + X14 + X15 + X21 + X22 + X23 + X24 +
X31 + X32 + X33 + (1+X11 | SID) | endo(X15),
data = dataMultilevelIV, verbose = FALSE))
expect_silent(res.lmer2 <- lmer(formula = y ~ X11 + X12 + X13 + X14 + X15 + X21 + X22 + X23 + X24 +
X31 + X32 + X33 + (1+X11 | SID),
data = dataMultilevelIV,
control = lmerControl(optimizer = "Nelder_Mead",
optCtrl = list(maxfun=100000))))
expect_equal(coef(res.ml2)[, "REF"], coef(summary(res.lmer2))[, "Estimate"])
expect_silent(res.ml3 <- multilevelIV(formula = y ~ X11 + X12 + X13 + X14 + X15 + X21 + X22 + X23 + X24 +
X31 + X32 + X33 + (1|CID)+(1+X11 | SID) | endo(X15),
data = dataMultilevelIV, verbose = FALSE))
expect_silent(res.lmer3 <- lmer(formula = y ~ X11 + X12 + X13 + X14 + X15 + X21 + X22 + X23 + X24 +
X31 + X32 + X33 +(1|CID)+ (1+X11 | SID),
data = dataMultilevelIV,
control = lmerControl(optimizer = "Nelder_Mead",
optCtrl = list(maxfun=100000))))
expect_equal(coef(res.ml3)[, "REF"], coef(summary(res.lmer3))[, "Estimate"])
})
test_that("Retrieve generated data params", {
skip_on_cran()
expect_silent(res.ml3 <- multilevelIV(formula = y ~ X11 + X12 + X13 + X14 + X15 + X21 + X22 + X23 + X24 +
X31 + X32 + X33 + (1|CID)+(1| SID) | endo(X15),
data = dataMultilevelIV, verbose = FALSE))
correct.coefs <- c("(Intercept)"=64,
X11 = 3, X12=9, X13=-2, X14 = 2,
X15 = -1, #endo
X21 = -1.5, X22 = -4, X23 = -3, X24 = 6,
X31 = 0.5, X32 = 0.1, X33 = -0.5)
expect_equal(coef(res.ml3)[, "REF"], correct.coefs, tolerance = 0.1)
})
test_that("Reproduce results by Kim and Frees 2007", {
skip_on_cran()
kf.formula <- TLI ~ GRADE_3 + RETAINED + SWITCHSC + S_FREELU +
FEMALE + BLACK + HISPANIC + OTHER+ C_COHORT+
T_EXPERI + CLASS_SI+ P_MINORI + (1 + GRADE_3|NEWCHILD) | endo(CLASS_SI)
# Is in tests/testthat folder
df.data.kf <- read.csv("dallas2485.csv", header=TRUE)
expect_silent(res.kf <- multilevelIV(formula = kf.formula, data = df.data.kf, verbose = FALSE))
correct.coefs <- cbind(REF = c("(Intercept)"=69.78, GRADE_3=3.375, RETAINED=9.205, SWITCHSC=-0.365, S_FREELU=-0.227, FEMALE=-1.234, BLACK=-4.745,
HISPANIC=-3.608, OTHER=6.526, C_COHORT=1.497, T_EXPERI=-0.116, CLASS_SI=0.157, P_MINORI=0.069))
# Compare coefs
expect_equal(coef(res.kf)[, "REF"], correct.coefs[, "REF"], tolerance=0.1)
})
# Predict ----------------------------------------------------------------------------------------
context("Correctness - multilevelIV - Predict")
expect_silent(res.m2 <- multilevelIV(y ~ X11 + X12 + X13 + X14 + X15 + X21 + X22 + X23 + X24 + X31 +
X32 + X33 + (1|SID) | endo(X15),
data = dataMultilevelIV, verbose = FALSE))
expect_silent(res.m3 <- multilevelIV(y ~ X11 + X12 + X13 + X14 + X15 + X21 + X22 + X23 + X24 + X31 +
X32 + X33 + (1| CID) + (1|SID) | endo(X15),
data = dataMultilevelIV, verbose = FALSE))
test_that("No newdata results in fitted values", {
for(m in all.L2.models)
expect_equal(predict(res.m2, model=m), fitted(res.m2, model=m))
for(m in all.L3.models)
expect_equal(predict(res.m3, model=m), fitted(res.m3, model=m))
})
test_that("Same prediction data as for fitting results in fitted values", {
for(m in all.L2.models)
expect_equal(predict(res.m2, newdata=dataMultilevelIV,model=m), fitted(res.m2, model=m))
for(m in all.L3.models)
expect_equal(predict(res.m3, newdata=dataMultilevelIV, model=m), fitted(res.m3, model=m))
})
test_that("Correct structure of predictions", {
for(m in all.L2.models){
expect_silent(pred.2 <- predict(res.m2, newdata=dataMultilevelIV, model=m))
expect_true(is.numeric(pred.2))
expect_true(length(pred.2) == nrow(dataMultilevelIV))
expect_true(all(names(pred.2) == names(fitted(res.m2, model=m))))
expect_true(all(names(pred.2) == rownames(dataMultilevelIV)))
}
for(m in all.L3.models){
expect_silent(pred.3 <- predict(res.m3, newdata=dataMultilevelIV, model=m))
expect_true(is.numeric(pred.3))
expect_true(length(pred.3) == nrow(dataMultilevelIV))
expect_true(all(names(pred.3) == names(fitted(res.m3, model=m))))
expect_true(all(names(pred.3) == rownames(dataMultilevelIV)))
}
})
test_that("Correct when using transformations in the formula", {
skip_on_cran()
# do endogenous and non-endogenous at the same time
expect_silent(res.m2 <- multilevelIV(y ~ X11 + X12 + X13 + X14 + I((X15+14)/4) + X21 + X22 + X23 +
X24 + log(X31) + X32 + X33 + (1|SID) | endo(I((X15+14)/4)),
data = dataMultilevelIV, verbose = FALSE))
for(m in all.L2.models)
expect_equal(predict(res.m2, newdata=dataMultilevelIV,model=m), fitted(res.m2, model=m))
expect_silent(res.m3 <- multilevelIV(y ~ X11 + X12 + X13 + X14 + I((X15+14)/4) + X21 + X22 + X23 +
X24 + log(X31) + X32 + X33 + (1| CID) + (1|SID) | endo(I((X15+14)/4)),
data = dataMultilevelIV, verbose = FALSE))
for(m in all.L3.models)
expect_equal(predict(res.m3, newdata=dataMultilevelIV, model=m), fitted(res.m3, model=m))
})
mmeierer/REndo documentation built on Sept. 11, 2023, 11:51 a.m.
R Package Documentation
Browse R Packages
We want your feedback!
Note that we can't provide technical support on individual packages. You should contact the package authors for that.
# Required data ----------------------------------------------------------------------------------------------------------------------
data("dataMultilevelIV")
all.L3.models <- c("REF", "FE_L2", "FE_L3", "GMM_L2", "GMM_L3")
all.L2.models <- c("REF", "FE_L2", "GMM_L2")
# Formula transformations ------------------------------------------------------------------------------------------------------------
context("Correctness - multilevelIV - Formula transformations")
test_that("Transformations are correct for L2", {
skip_on_cran()
expect_silent(correct.res <- multilevelIV(formula = y ~ X11 + X12 + X13 + X14 + X15 + X21 + X22 + X23 + X24 +
X31 + X32 + X33 + (1+X11 | SID) | endo(X15),
data = dataMultilevelIV, verbose = FALSE))
# Can handle transformations in LHS
data.altered <- dataMultilevelIV
data.altered$y <- exp(data.altered$y)
expect_silent(res.trans.lhs <- multilevelIV(formula = log(y) ~ X11 + X12 + X13 + X14 + X15 + X21 + X22 + X23 + X24 +
X31 + X32 + X33 + (1+X11 | SID) | endo(X15),
data = data.altered, verbose = FALSE))
expect_equal(coef(res.trans.lhs), coef(correct.res))
for(m in all.L2.models)
expect_equal(coef(summary(res.trans.lhs, model = m)), coef(summary(correct.res, model = m)))
# Can handle transformations in exo
data.altered <- dataMultilevelIV
data.altered$X23 <- exp(data.altered$X23)
expect_silent(res.trans.exo <- multilevelIV(formula = y ~ X11 + X12 + X13 + X14 + X15 + X21 + X22 + log(X23) + X24 +
X31 + X32 + X33 + (1+X11 | SID) | endo(X15),
data = data.altered, verbose = FALSE))
expect_equal(coef(res.trans.exo), coef(correct.res), check.attributes = FALSE)
for(m in all.L2.models)
expect_equal(coef(summary(res.trans.exo, model = m)), coef(summary(correct.res, model = m)), check.attributes = FALSE)
# Can handle transformations in endo
data.altered <- dataMultilevelIV
data.altered$X15 <- exp(data.altered$X15)
expect_silent(res.trans.endo <- multilevelIV(formula = y ~ X11 + X12 + X13 + X14 + log(X15) + X21 + X22 + X23 + X24 +
X31 + X32 + X33 + (1+X11 | SID) | endo(log(X15)),
data = data.altered, verbose = FALSE))
expect_equal(coef(res.trans.endo), coef(correct.res), check.attributes = FALSE)
for(m in all.L2.models)
expect_equal(coef(summary(res.trans.endo, model = m)), coef(summary(correct.res, model = m)), check.attributes = FALSE)
# Can handle transformations in slope
data.altered <- dataMultilevelIV
data.altered$X11 <- exp(data.altered$X11)
expect_silent(res.trans.slope <- multilevelIV(formula = y ~ log(X11) + X12 + X13 + X14 + X15 + X21 + X22 + X23 + X24 +
X31 + X32 + X33 + (1+log(X11) | SID) | endo(X15),
data = data.altered, verbose = FALSE))
expect_equal(coef(res.trans.slope), coef(correct.res), check.attributes = FALSE)
for(m in all.L2.models)
expect_equal(coef(summary(res.trans.slope, model = m)), coef(summary(correct.res, model = m)), check.attributes = FALSE)
})
test_that("Transformations are correct for L3", {
skip_on_cran()
expect_silent(correct.res <- multilevelIV(formula = y ~ X11 + X12 + X13 + X14 + X15 + X21 + X22 + X23 + X24 +
X31 + X32 + X33 + (1+X11 | CID) + (1 | SID) | endo(X15),
data = dataMultilevelIV, verbose = FALSE))
# Can handle transformations in LHS
data.altered <- dataMultilevelIV
data.altered$y <- (data.altered$y+99)/2
expect_silent(res.trans.lhs <- multilevelIV(formula = I((y*2)-99) ~ X11 + X12 + X13 + X14 + X15 + X21 + X22 + X23 + X24 +
X31 + X32 + X33 + (1+X11 | CID) + (1 | SID) | endo(X15),
data = data.altered, verbose = FALSE))
expect_equal(coef(res.trans.lhs), coef(correct.res), check.attributes = FALSE)
for(m in all.L3.models)
expect_equal(coef(summary(res.trans.lhs, model = m)), coef(summary(correct.res, model = m)), check.attributes = FALSE)
# Can handle transformations in exo
data.altered <- dataMultilevelIV
data.altered$X23 <- (data.altered$X23+99)/2
expect_silent(res.trans.exo <- multilevelIV(formula = y ~ X11 + X12 + X13 + X14 + X15 + X21 + X22 + I((X23*2)-99) + X24 +
X31 + X32 + X33 + (1+X11 | CID) + (1 | SID) | endo(X15),
data = data.altered, verbose = FALSE))
expect_equal(coef(res.trans.exo), coef(correct.res), check.attributes = FALSE)
for(m in all.L3.models){
expect_equal(coef(summary(res.trans.exo, model = m)), coef(summary(correct.res, model = m)), check.attributes = FALSE)
}
# Can handle transformations in endo
data.altered <- dataMultilevelIV
data.altered$X15 <- (data.altered$X15+99)/2
expect_silent(res.trans.endo <- multilevelIV(formula = y ~ X11 + X12 + X13 + X14 + I((X15*2)-99) + X21 + X22 + X23 + X24 +
X31 + X32 + X33 + (1+X11 | CID) + (1 | SID) | endo(I((X15*2)-99)),
data = data.altered, verbose = FALSE))
expect_equal(coef(res.trans.endo), coef(correct.res), check.attributes = FALSE)
for(m in all.L3.models){
expect_equal(coef(summary(res.trans.endo, model = m)), coef(summary(correct.res, model = m)), check.attributes = FALSE)
}
# Can handle transformations in slope
data.altered <- dataMultilevelIV
data.altered$X11 <- (data.altered$X11+99)/2
expect_silent(res.trans.slope <- multilevelIV(formula = y ~ I((X11*2)-99) + X12 + X13 + X14 + X15 + X21 + X22 + X23 + X24 +
X31 + X32 + X33 + (1+I((X11*2)-99) | CID) + (1 | SID) | endo(X15),
data = data.altered, verbose = FALSE))
expect_equal(coef(res.trans.slope), coef(correct.res), check.attributes = FALSE)
for(m in all.L3.models)
expect_equal(coef(summary(res.trans.slope, model = m)), coef(summary(correct.res, model = m)), check.attributes = FALSE)
})
# Data sorting ------------------------------------------------------------------------------------------------
context("Correctness - multilevelIV - Data sorting")
# **correct if levels are given separately: (A|SID) + (B|SID) == (A+B|SID)
# ** Also check for W and V?
test_that("Unsorted data is correct L2", {
skip_on_cran()
# Correct = coefs + sorting of residuals / fitted
# Distinguishable non-standard rownames
rownames(dataMultilevelIV) <- as.character(seq(from=nrow(dataMultilevelIV)+100000, to=1+100000))
expect_silent(res.sorted <- multilevelIV(formula = y ~ X11 + X12 + X13 + X14 + X15 + X21 + X22 + X23 + X24 +
X31 + X32 + X33 + (1+X11 | SID) | endo(X15),
data = dataMultilevelIV, verbose = FALSE))
# Can handle transformations in LHS
data.altered <- dataMultilevelIV
data.altered <- data.altered[sample(x=nrow(dataMultilevelIV), size = nrow(dataMultilevelIV), replace = FALSE), ]
expect_silent(res.unsorted <- multilevelIV(formula = y ~ X11 + X12 + X13 + X14 + X15 + X21 + X22 + X23 + X24 +
X31 + X32 + X33 + (1+X11 | SID) | endo(X15),
data = data.altered, verbose = FALSE))
# Coefs + summary statistics the same
expect_equal(coef(res.unsorted), coef(res.sorted))
for(m in all.L2.models){
expect_equal(coef(summary(res.unsorted, model = m)), coef(summary(res.sorted, model = m)))
# Sorting of fitted / residuals same as input
expect_equal(names(fitted(res.sorted, model = m)), rownames(dataMultilevelIV))
expect_equal(names(resid(res.sorted, model = m)), rownames(dataMultilevelIV))
expect_equal(names(fitted(res.unsorted, model = m)), rownames(data.altered))
expect_equal(names(resid(res.unsorted, model = m)), rownames(data.altered))
}
})
test_that("Unsorted data is correct L3", {
skip_on_cran()
# Correct = coefs + sorting of residuals / fitted
# Distinguishable non-standard rownames
rownames(dataMultilevelIV) <- as.character(seq(from=nrow(dataMultilevelIV)+100000, to=1+100000))
expect_silent(res.sorted <- multilevelIV(formula = y ~ X11 + X12 + X13 + X14 + X15 + X21 + X22 + X23 + X24 +
X31 + X32 + X33 + (1|CID)+(1 | SID) | endo(X15),
data = dataMultilevelIV, verbose = FALSE))
# Can handle transformations in LHS
data.altered <- dataMultilevelIV
data.altered <- data.altered[sample.int(n=nrow(dataMultilevelIV), replace = FALSE), ]
expect_silent(res.unsorted <- multilevelIV(formula = y ~ X11 + X12 + X13 + X14 + X15 + X21 + X22 + X23 + X24 +
X31 + X32 + X33 + (1|CID)+(1 | SID) | endo(X15),
data = data.altered, verbose = FALSE))
# Coefs + summary statistics the same
expect_equal(coef(res.unsorted), coef(res.sorted))
for(m in all.L3.models){
expect_equal(coef(summary(res.unsorted, model = m)), coef(summary(res.sorted, model = m)))
# Sorting of fitted / residuals same as input
expect_equal(names(fitted(res.sorted, model = m)), rownames(dataMultilevelIV))
expect_equal(names(resid(res.sorted, model = m)), rownames(dataMultilevelIV))
expect_equal(names(fitted(res.unsorted, model = m)), rownames(data.altered))
expect_equal(names(resid(res.unsorted, model = m)), rownames(data.altered))
}
})
# Reproduce results ------------------------------------------------------------------------------
context("Correctness - multilevelIV - Reproduce results")
test_that("REF is same as lmer()", {
# Have to use the same optimizer controls for lmer as used in multilevelIV
skip_on_cran()
# L2
expect_silent(res.ml2 <- multilevelIV(formula = y ~ X11 + X12 + X13 + X14 + X15 + X21 + X22 + X23 + X24 +
X31 + X32 + X33 + (1+X11 | SID) | endo(X15),
data = dataMultilevelIV, verbose = FALSE))
expect_silent(res.lmer2 <- lmer(formula = y ~ X11 + X12 + X13 + X14 + X15 + X21 + X22 + X23 + X24 +
X31 + X32 + X33 + (1+X11 | SID),
data = dataMultilevelIV,
control = lmerControl(optimizer = "Nelder_Mead",
optCtrl = list(maxfun=100000))))
expect_equal(coef(res.ml2)[, "REF"], coef(summary(res.lmer2))[, "Estimate"])
expect_silent(res.ml3 <- multilevelIV(formula = y ~ X11 + X12 + X13 + X14 + X15 + X21 + X22 + X23 + X24 +
X31 + X32 + X33 + (1|CID)+(1+X11 | SID) | endo(X15),
data = dataMultilevelIV, verbose = FALSE))
expect_silent(res.lmer3 <- lmer(formula = y ~ X11 + X12 + X13 + X14 + X15 + X21 + X22 + X23 + X24 +
X31 + X32 + X33 +(1|CID)+ (1+X11 | SID),
data = dataMultilevelIV,
control = lmerControl(optimizer = "Nelder_Mead",
optCtrl = list(maxfun=100000))))
expect_equal(coef(res.ml3)[, "REF"], coef(summary(res.lmer3))[, "Estimate"])
})
test_that("Retrieve generated data params", {
skip_on_cran()
expect_silent(res.ml3 <- multilevelIV(formula = y ~ X11 + X12 + X13 + X14 + X15 + X21 + X22 + X23 + X24 +
X31 + X32 + X33 + (1|CID)+(1| SID) | endo(X15),
data = dataMultilevelIV, verbose = FALSE))
correct.coefs <- c("(Intercept)"=64,
X11 = 3, X12=9, X13=-2, X14 = 2,
X15 = -1, #endo
X21 = -1.5, X22 = -4, X23 = -3, X24 = 6,
X31 = 0.5, X32 = 0.1, X33 = -0.5)
expect_equal(coef(res.ml3)[, "REF"], correct.coefs, tolerance = 0.1)
})
test_that("Reproduce results by Kim and Frees 2007", {
skip_on_cran()
kf.formula <- TLI ~ GRADE_3 + RETAINED + SWITCHSC + S_FREELU +
FEMALE + BLACK + HISPANIC + OTHER+ C_COHORT+
T_EXPERI + CLASS_SI+ P_MINORI + (1 + GRADE_3|NEWCHILD) | endo(CLASS_SI)
# Is in tests/testthat folder
df.data.kf <- read.csv("dallas2485.csv", header=TRUE)
expect_silent(res.kf <- multilevelIV(formula = kf.formula, data = df.data.kf, verbose = FALSE))
correct.coefs <- cbind(REF = c("(Intercept)"=69.78, GRADE_3=3.375, RETAINED=9.205, SWITCHSC=-0.365, S_FREELU=-0.227, FEMALE=-1.234, BLACK=-4.745,
HISPANIC=-3.608, OTHER=6.526, C_COHORT=1.497, T_EXPERI=-0.116, CLASS_SI=0.157, P_MINORI=0.069))
# Compare coefs
expect_equal(coef(res.kf)[, "REF"], correct.coefs[, "REF"], tolerance=0.1)
})
# Predict ----------------------------------------------------------------------------------------
context("Correctness - multilevelIV - Predict")
expect_silent(res.m2 <- multilevelIV(y ~ X11 + X12 + X13 + X14 + X15 + X21 + X22 + X23 + X24 + X31 +
X32 + X33 + (1|SID) | endo(X15),
data = dataMultilevelIV, verbose = FALSE))
expect_silent(res.m3 <- multilevelIV(y ~ X11 + X12 + X13 + X14 + X15 + X21 + X22 + X23 + X24 + X31 +
X32 + X33 + (1| CID) + (1|SID) | endo(X15),
data = dataMultilevelIV, verbose = FALSE))
test_that("No newdata results in fitted values", {
for(m in all.L2.models)
expect_equal(predict(res.m2, model=m), fitted(res.m2, model=m))
for(m in all.L3.models)
expect_equal(predict(res.m3, model=m), fitted(res.m3, model=m))
})
test_that("Same prediction data as for fitting results in fitted values", {
for(m in all.L2.models)
expect_equal(predict(res.m2, newdata=dataMultilevelIV,model=m), fitted(res.m2, model=m))
for(m in all.L3.models)
expect_equal(predict(res.m3, newdata=dataMultilevelIV, model=m), fitted(res.m3, model=m))
})
test_that("Correct structure of predictions", {
for(m in all.L2.models){
expect_silent(pred.2 <- predict(res.m2, newdata=dataMultilevelIV, model=m))
expect_true(is.numeric(pred.2))
expect_true(length(pred.2) == nrow(dataMultilevelIV))
expect_true(all(names(pred.2) == names(fitted(res.m2, model=m))))
expect_true(all(names(pred.2) == rownames(dataMultilevelIV)))
}
for(m in all.L3.models){
expect_silent(pred.3 <- predict(res.m3, newdata=dataMultilevelIV, model=m))
expect_true(is.numeric(pred.3))
expect_true(length(pred.3) == nrow(dataMultilevelIV))
expect_true(all(names(pred.3) == names(fitted(res.m3, model=m))))
expect_true(all(names(pred.3) == rownames(dataMultilevelIV)))
}
})
test_that("Correct when using transformations in the formula", {
skip_on_cran()
# do endogenous and non-endogenous at the same time
expect_silent(res.m2 <- multilevelIV(y ~ X11 + X12 + X13 + X14 + I((X15+14)/4) + X21 + X22 + X23 +
X24 + log(X31) + X32 + X33 + (1|SID) | endo(I((X15+14)/4)),
data = dataMultilevelIV, verbose = FALSE))
for(m in all.L2.models)
expect_equal(predict(res.m2, newdata=dataMultilevelIV,model=m), fitted(res.m2, model=m))
expect_silent(res.m3 <- multilevelIV(y ~ X11 + X12 + X13 + X14 + I((X15+14)/4) + X21 + X22 + X23 +
X24 + log(X31) + X32 + X33 + (1| CID) + (1|SID) | endo(I((X15+14)/4)),
data = dataMultilevelIV, verbose = FALSE))
for(m in all.L3.models)
expect_equal(predict(res.m3, newdata=dataMultilevelIV, model=m), fitted(res.m3, model=m))
})
R Package Documentation
Browse R Packages
We want your feedback!
Note that we can't provide technical support on individual packages. You should contact the package authors for that.
Embedding an R snippet on your website
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.