Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

added multinomial, ordinal and firth logistic regression #179

Merged
merged 5 commits into from
Dec 10, 2022
Merged

added multinomial, ordinal and firth logistic regression #179

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
406a257
added multinomial, ordinal and firth logistic regression
fqixiang Nov 1, 2022
e9b4ac2
fixed unit test error
fqixiang Nov 1, 2022
5b78934
implemented suggested changes
fqixiang Nov 7, 2022
3f0d866
defined environment for model formulas
fqixiang Dec 9, 2022
1d6c846
fixed an if/else statement
fqixiang Dec 9, 2022
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
4 changes: 3 additions & 1 deletion DESCRIPTION
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -23,13 +23,15 @@ Imports:
jaspGraphs,
jaspTTests,
lmtest,
logistf,
MASS,
matrixStats,
mdscore,
ppcor,
purrr,
Rcpp,
statmod
statmod,
VGAM
Remotes:
AlexanderLyNL/bstats,
jasp-stats/jaspAnova,
Expand Down
187 changes: 142 additions & 45 deletions R/glmClassical.R
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -25,7 +25,6 @@ GeneralizedLinearModel <- function(jaspResults, dataset = NULL, options, ...) {
ready <- options[["dependent"]] != ""
}


if (ready) {
dataset <- .glmReadData(dataset, options)
.glmCheckDataErrors(dataset, options)
Expand All @@ -35,6 +34,7 @@ GeneralizedLinearModel <- function(jaspResults, dataset = NULL, options, ...) {
.glmModelSummaryTable(jaspResults, dataset, options, ready, position = 1)
.glmModelFitTable( jaspResults, dataset, options, ready, position = 2)
.glmEstimatesTable( jaspResults, dataset, options, ready, position = 3)
if (options$family == "other") return()

#diagnostic tables and plots
.glmDiagnostics(jaspResults, dataset, options, ready, position = 4)
Expand All @@ -58,7 +58,7 @@ GeneralizedLinearModel <- function(jaspResults, dataset = NULL, options, ...) {
dependentVar <- options[["dependent"]]
dependentVar <- dependentVar[dependentVar != ""]

if (options[["family"]] == "bernoulli") {
if (options[["family"]] %in% c("bernoulli", "other")) {
return(.readDataSetToEnd(columns.as.numeric = numericVars,
columns.as.factor = c(factorVars, dependentVar),
exclude.na.listwise = c(numericVars, factorVars, dependentVar)))
Expand Down Expand Up @@ -134,6 +134,21 @@ GeneralizedLinearModel <- function(jaspResults, dataset = NULL, options, ...) {
if (!is.numeric(dataset[, options[["dependent"]]]))
.quitAnalysis(gettextf("The Gaussian family requires the dependent variable to be a numerical variable."))
}

if ((options[["family"]] == "other") && (options[["otherGlmModel"]] == "multinomialLogistic")) {
if (length(levels(dataset[, options[["dependent"]]])) < 3)
.quitAnalysis(gettext("Multinomial logistic regression requires the dependent variable to be a factor with at least 3 levels."))
}

if (options[["otherGlmModel"]] == "ordinalLogistic") {
if (length(levels(dataset[, options[["dependent"]]])) < 3)
.quitAnalysis(gettext("Ordinal logistic regression requires the dependent variable to be a factor with at least 3 levels."))
}

if (options[["otherGlmModel"]] == "firthLogistic") {
if (length(levels(dataset[, options[["dependent"]]])) != 2)
.quitAnalysis(gettext("Firth logistic regression requires the dependent variable to be a factor with 2 levels."))
}
}

# Model Summary Table
Expand All @@ -149,7 +164,7 @@ GeneralizedLinearModel <- function(jaspResults, dataset = NULL, options, ...) {
modelSummary <- createJaspTable(gettextf("Model Summary - %s", options[['dependent']]))
}

dependList <- c("dependent", "family", "link", "modelTerms", "interceptTerm", "weights", "offset")
dependList <- c("dependent", "family", "link", "modelTerms", "interceptTerm", "weights", "offset", "otherGlmModel")
modelSummary$dependOn(dependList)
modelSummary$position <- position
modelSummary$showSpecifiedColumnsOnly <- TRUE
Expand All @@ -174,14 +189,20 @@ GeneralizedLinearModel <- function(jaspResults, dataset = NULL, options, ...) {
glmModels <- .glmComputeModel(jaspResults, dataset, options)
hasNuisance <- .hasNuisance(options)
if (hasNuisance) {
terms <- rownames(summary(glmModels[["nullModel"]])[["coefficients"]])
terms <- jaspBase::gsubInteractionSymbol(terms[terms!="(Intercept)"])
message <- gettextf("Null model contains nuisance parameters: %s",
paste(terms, collapse = ", "))
if ((options[["family"]] == "other") & (options[["otherGlmModel"]] %in% c("multinomialLogistic", "ordinalLogistic")))
termsNullModel <- rownames(VGAM::coef(VGAM::summaryvglm(glmModels[["nullModel"]])))
else if (options[["otherGlmModel"]] == "firthLogistic")
termsNullModel <- names(coef((glmModels[["nullModel"]])))
else
termsNullModel <- rownames(coef(summary(glmModels[["nullModel"]])))
nuisanceTerms <- jaspBase::gsubInteractionSymbol(termsNullModel[!grepl("(Intercept)", termsNullModel, fixed = TRUE)])
message <- gettextf("Null model contains nuisance parameters: %s.",
paste(nuisanceTerms, collapse = ", "))
jaspResults[["modelSummary"]]$addFootnote(message)
}

#log-likelihood ratio test to compare nested models (null vs full)
if (options[["family"]] %in% c("bernoulli", "binomial", "poisson")) {
if (options[["family"]] %in% c("bernoulli", "binomial", "poisson", "other")) {
testType <- "Chisq"
pvalName <- "Pr(>Chi)"
}
Expand All @@ -191,24 +212,68 @@ GeneralizedLinearModel <- function(jaspResults, dataset = NULL, options, ...) {
pvalName <- "Pr(>F)"
}

anovaRes <- anova(glmModels[["nullModel"]], glmModels[["fullModel"]],
test = testType)
if (options[["family"]] == "other") {
if (options[["otherGlmModel"]] == "firthLogistic") {
devNullModel <- glmModels[["nullModel"]][["loglik"]]["full"]*-2
aicNullModel <- devNullModel + 2*glmModels[["nullModel"]][["df"]]
bicNullModel <- devNullModel + log(nobs(glmModels[["nullModel"]]))*glmModels[["nullModel"]][["df"]]
dofNullModel <- glmModels[["nullModel"]][["n"]] - glmModels[["nullModel"]][["df"]]

devFullModel <- glmModels[["fullModel"]][["loglik"]]["full"]*-2
aicFullModel <- devFullModel + 2*glmModels[["fullModel"]][["df"]]
bicFullModel <- devFullModel + log(nobs(glmModels[["fullModel"]]))*glmModels[["fullModel"]][["df"]]
dofFullModel <- glmModels[["fullModel"]][["n"]] - glmModels[["fullModel"]][["df"]]

chiValue <- devNullModel - devFullModel
pValue <- 1 - pchisq(chiValue, glmModels[["fullModel"]][["df"]])
}
else {
devNullModel <- VGAM::deviance(glmModels[["nullModel"]])
aicNullModel <- VGAM::AIC(glmModels[["nullModel"]])
bicNullModel <- VGAM::BIC(glmModels[["nullModel"]])
dofNullModel <- VGAM::df.residual(glmModels[["nullModel"]])

devFullModel <- VGAM::deviance(glmModels[["fullModel"]])
aicFullModel <- VGAM::AIC(glmModels[["fullModel"]])
bicFullModel <- VGAM::BIC(glmModels[["fullModel"]])
dofFullModel <- VGAM::df.residual(glmModels[["fullModel"]])

anovaRes <- VGAM::anova.vglm(glmModels[["nullModel"]], glmModels[["fullModel"]], type = 1)
chiValue <- devNullModel - devFullModel
pValue <- anovaRes[["Pr(>Chi)"]][[2]]
}
} else {
devNullModel <- glmModels[["nullModel"]][["deviance"]]
aicNullModel <- glmModels[["nullModel"]][["aic"]]
bicNullModel <- BIC(glmModels[["nullModel"]])
dofNullModel <- glmModels[["nullModel"]][["df.residual"]]

devFullModel <- glmModels[["fullModel"]][["deviance"]]
aicFullModel <- glmModels[["fullModel"]][["aic"]]
bicFullModel <- BIC(glmModels[["fullModel"]])
dofFullModel <- glmModels[["fullModel"]][["df.residual"]]

anovaRes <- anova(glmModels[["nullModel"]], glmModels[["fullModel"]],
test = testType)
chiValue <- anovaRes$Deviance[[2]]
pValue <- anovaRes[[pvalName]][[2]]
}

rows <- list(
list(mod = "H\u2080",
dev = glmModels[["nullModel"]][["deviance"]],
aic = glmModels[["nullModel"]][["aic"]],
bic = BIC(glmModels[["nullModel"]]),
dof = glmModels[["nullModel"]][["df.residual"]],
dev = devNullModel,
aic = aicNullModel,
bic = bicNullModel,
dof = dofNullModel,
chi = "",
pvl = ""),
list(mod = "H\u2081",
dev = glmModels[["fullModel"]][["deviance"]],
aic = glmModels[["fullModel"]][["aic"]],
bic = BIC(glmModels[["fullModel"]]),
dof = glmModels[["fullModel"]][["df.residual"]],
chi = anovaRes$Deviance[[2]],
pvl = anovaRes[[pvalName]][[2]])
dev = devFullModel,
aic = aicFullModel,
bic = bicFullModel,
dof = dofFullModel,
chi = chiValue,
pvl = pValue)
)
} else {
rows <- list(
Expand Down Expand Up @@ -253,25 +318,40 @@ GeneralizedLinearModel <- function(jaspResults, dataset = NULL, options, ...) {
glmModels <- .glmComputeModel(jaspResults, dataset, options)
modelObj <- glmModels[["fullModel"]]

if (options[["otherGlmModel"]] == "firthLogistic") {
dev <- modelObj[["loglik"]]["full"]*-2
pearsonResid <- (modelObj[['y']] - modelObj[["predict"]])/sqrt(modelObj[["predict"]]*(1-modelObj[["predict"]]))
pearson <- sum(pearsonResid^2)
dof <- modelObj[["n"]] - modelObj[["df"]]

} else {

if (options[["family"]] == "other")
pearson <- sum(VGAM::residuals(modelObj, type = "pearson")^2)
else
pearson <- sum(residuals(modelObj, type = "pearson")^2)
dev <- deviance(modelObj)
dof <- df.residual(modelObj)
}

if (options[["devianceGoodnessOfFit"]]) {
jaspResults[["modelFitTable"]]$addRows(
list(gofType = "Deviance",
gof = modelObj$deviance,
dof = modelObj$df.residual,
pval = pchisq(modelObj$deviance,
df=modelObj$df.residual,
gof = dev,
dof = dof,
pval = pchisq(dev,
df=dof,
lower.tail=FALSE))
)
}

if (options[["pearsonGoodnessOfFit"]]) {
pearson <- sum(modelObj$weights * modelObj$residuals^2)
jaspResults[["modelFitTable"]]$addRows(
list(gofType = "Pearson",
gof = pearson,
dof = modelObj$df.residual,
dof = dof,
pval = pchisq(pearson,
df=modelObj$df.residual,
df=dof,
lower.tail=FALSE))
)
}
Expand All @@ -288,11 +368,14 @@ GeneralizedLinearModel <- function(jaspResults, dataset = NULL, options, ...) {
estimatesTable$position <- position
estimatesTable$showSpecifiedColumnsOnly <- TRUE

if (options[["family"]] %in% c("bernoulli", "binomial", "poisson"))
if (options[["family"]] %in% c("bernoulli", "binomial", "poisson", "other"))
testStat <- "z"
else
testStat <- "t"

if (options["otherGlmModel"] == "firthLogistic")
testStat <- "\u03A7\u00B2"

estimatesTable$addColumnInfo(name = "param", title = "", type = "string")
estimatesTable$addColumnInfo(name = "est", title = gettext("Estimate"), type = "number")
estimatesTable$addColumnInfo(name = "se", title = gettext("Standard Error"), type = "number")
Expand All @@ -317,11 +400,21 @@ GeneralizedLinearModel <- function(jaspResults, dataset = NULL, options, ...) {
return()
# compute glm models
glmModels <- .glmComputeModel(jaspResults, dataset, options)
modelSummary <- summary(glmModels[["fullModel"]])[["coefficients"]]
fullModel <- glmModels[["fullModel"]]

if ((options[["family"]] == "other") & (options[["otherGlmModel"]] %in% c("multinomialLogistic", "ordinalLogistic")))
modelSummary <- VGAM::coef(VGAM::summaryvglm(fullModel))

if (options[["otherGlmModel"]] == "firthLogistic")
modelSummary <- cbind(coef(fullModel), sqrt(diag(fullModel$var)), qchisq(1 - fullModel$prob, 1), fullModel$prob)

if (options[["family"]] != "other")
modelSummary <- coef(summary(fullModel))

rowNames <- rownames(modelSummary)

if (options[["coefficientCi"]]) {
coefCiSummary <- confint(glmModels[["fullModel"]], level = options[["coefficientCiLevel"]])
coefCiSummary <- confint(fullModel, level = options[["coefficientCiLevel"]])
if (length(rowNames) == 1) coefCiSummary <- matrix(coefCiSummary, ncol = 2)
} else {
coefCiSummary <- matrix(nrow = length(rowNames),
Expand All @@ -335,24 +428,28 @@ GeneralizedLinearModel <- function(jaspResults, dataset = NULL, options, ...) {
estimatesTableData <- cbind(paramDf, modelSummary, coefCiSummary)
jaspResults[["estimatesTable"]]$setData(estimatesTableData)

# for (i in seq_along(rowNames)) {
# jaspResults[["estimatesTable"]]$addRows(
# list(param = jaspBase::gsubInteractionSymbol(rowNames[i]),
# est = modelSummary[i, "Estimate"],
# se = modelSummary[i, "Std. Error"],
# testStat = modelSummary[i, 3],
# pval = modelSummary[i, 4],
# ciLow = coefCiSummary[i, 1],
# ciUpp = coefCiSummary[i, 2])
# )
# }

if (options[["family"]] == "bernoulli") {
dv <- as.character(glmModels[["nullModel"]][["terms"]])[2]
dvLevel <- levels(glmModels[["nullModel"]][["data"]][[dv]])[2]
if ((options[["family"]] == "bernoulli") | (options[["otherGlmModel"]] == "firthLogistic")) {
dv <- options$dependent
dvLevel <- levels(dataset[[dv]])[2]

jaspResults[["estimatesTable"]]$addFootnote(gettextf("%1$s level '%2$s' coded as class 1.", dv, dvLevel))
}

if ((options["family"] == "other") & options[["otherGlmModel"]] == "multinomialLogistic") {
dv <- options$dependent
dvReferenceLevel <- tail(levels(dataset[[dv]]), 1)
dvLevels <- paste(paste(seq(1, length(dv)), levels(dataset[[dv]]), sep = ":"), collapse = ", ")

jaspResults[["estimatesTable"]]$addFootnote(gettextf("%1$s levels: %2$s. '%3$s' is the reference level.", dv, dvLevels, dvReferenceLevel))
}

if (options[["otherGlmModel"]] == "ordinalLogistic") {
dv <- options$dependent
dvLevels <- paste(paste(seq(1, length(dv)), levels(dataset[[dv]]), sep = ":"), collapse = ", ")
linearPredictors <- paste(VGAM::summaryvglm(fullModel)@misc$predictors.names, collapse = ", ")

jaspResults[["estimatesTable"]]$addFootnote(gettextf("%1$s levels: %2$s. Linear predictors: %3$s.", dv, dvLevels, linearPredictors))
}
}

# Diagnostics container
Expand Down
Loading