Package 'networktree' reference manual

Title:	Recursive Partitioning of Network Models
Description:	Network trees recursively partition the data with respect to covariates. Two network tree algorithms are available: model-based trees based on a multivariate normal model and nonparametric trees based on covariance structures. After partitioning, correlation-based networks (psychometric networks) can be fit on the partitioned data. For details see Jones, Mair, Simon, & Zeileis (2020) <doi:10.1007/s11336-020-09731-4>.
Authors:	Payton Jones [aut, cre] , Thorsten Simon [aut] , Achim Zeileis [aut]
Maintainer:	Payton Jones <[email protected]>
License:	GPL-2 \| GPL-3
Version:	1.0.1.9000
Built:	2025-01-22 03:57:27 UTC
Source:	https://github.com/paytonjjones/networktree

comparetree

Description

Quickly compares two partitions of a networktree object

Usage

comparetree(
  tree,
  id1 = 2L,
  id2 = 3L,
  transform = "detect",
  highlights = 5,
  plot = FALSE,
  plot.type = c("compare", "subtract"),
  layout = "constrained",
  ...
)
comparetree(
  tree,
  id1 = 2L,
  id2 = 3L,
  transform = "detect",
  highlights = 5,
  plot = FALSE,
  plot.type = c("compare", "subtract"),
  layout = "constrained",
  ...
)

Arguments

`tree`	a networktree object
`id1`	the first partition
`id2`	the second partition
`transform`	should stored correlation matrices be transformed to partial correlations or graphical lasso? Can be set to "cor", "pcor", or "glasso". Defaults to automatic detection
`highlights`	the number of comparisons to highlight
`plot`	plot a comparison of the two partitions?
`plot.type`	"compare" or "subtract". "compare" plots the two networks side by side. "subtract" subtracts network 2 from network 1, and plots a network where edge weights indicate the difference
`layout`	layout for the plots. The default "constrained" uses a FR layout from the full dataset
`...`	additional arguments passed to qgraph

Examples

set.seed(1)
d <- data.frame(trend = 1:200, foo = runif(200, -1, 1))
d <- cbind(d, rbind(
  mvtnorm::rmvnorm(100, mean = c(0, 0, 0),
          sigma = matrix(c(1, 0.5, 0.5, 0.5, 1, 0.5, 0.5, 0.5, 1), ncol = 3)),
  mvtnorm::rmvnorm(100, mean = c(0, 0, 0),
          sigma = matrix(c(1, 0, 0.5, 0, 1, 0.5, 0.5, 0.5, 1), ncol = 3))
))
colnames(d)[3:5] <- paste0("y", 1:3)

## Generate a networktree
tree1 <- networktree(nodevars=d[,3:5], splitvars=d[,1:2])

## Print out the structure
tree1

## Compare any two partitions
comparetree(tree1, id1=2, id2=3, highlights=3)

## Add a comparison plot
comparetree(tree1, id1=2, id2=3, plot=TRUE)

set.seed(1)
d <- data.frame(trend = 1:200, foo = runif(200, -1, 1))
d <- cbind(d, rbind(
  mvtnorm::rmvnorm(100, mean = c(0, 0, 0),
          sigma = matrix(c(1, 0.5, 0.5, 0.5, 1, 0.5, 0.5, 0.5, 1), ncol = 3)),
  mvtnorm::rmvnorm(100, mean = c(0, 0, 0),
          sigma = matrix(c(1, 0, 0.5, 0, 1, 0.5, 0.5, 0.5, 1), ncol = 3))
))
colnames(d)[3:5] <- paste0("y", 1:3)

## Generate a networktree
tree1 <- networktree(nodevars=d[,3:5], splitvars=d[,1:2])

## Print out the structure
tree1

## Compare any two partitions
comparetree(tree1, id1=2, id2=3, highlights=3)

## Add a comparison plot
comparetree(tree1, id1=2, id2=3, plot=TRUE)

cortrafo

Description

An influence function that transforms the response variables (y1, y2, y3...) into scores relevant to the correlations, means, and/or variances.

For example, in the case of correlations the variables are transformed into a matrix of (n^2-n)/2 columns (e.g., the number of total correlations), and i rows, where i is the # of observations of y1, where the mean of each vector is equal to the correlation between y1 and y2, y1 and y3, etc.

Used internally in when method="ctree".

Usage

cortrafo(data, weights, control, model, ...)
cortrafo(data, weights, control, model, ...)

Arguments

`data`	a matrix or data
`weights`	not currently used
`control`	not currently used
`model`	can be any combination of c("correlation", "mean", "variance"). Scores are determined based on the specified characteristics
`...`	not currently used

Depression Anxiety and Stress Scale

Description

This dataset includes a randomly selected subsample of 5000 online participants who participated in a questionnaire available through the Open Source Psychometrics Project (https://openpsychometrics.org/), an organization that maintains an open website for the public to take psychometric tests for educational and entertainment purposes

Usage

dass
dass

Format

a dataframe. Columns represent questionnaire items and rows represent individuals

Details

The Depression Anxiety and Stress Scale (DASS) is a self-report instrument for measuring depression, anxiety, and tension or stress. Each of 42 items falls into one of the three corresponding subscales.

Labels for DASS items in this dataset are denoted by the prefix "dass" and the suffix "_D", "_A", or "_S", indicating the depression, anxiety, or stress subscale.

Also includes demographics such as country, education level, rearing environment (urban/suburban/rural), gender, English as a native language, age, religion, sexual orientation, race, voting status, marriage status, and number of children in one's family during childhood.

The full dataset is publicly available at https://openpsychometrics.org/_rawdata/DASS_data_21.02.19.zip and can be cited as:

OpenPsychometrics (2019). Depression Anxiety and Stress Scale Survey. Retrieved from https://openpsychometrics.org/_rawdata/DASS_data_21.02.19.zip

Examples

head(dass)


## Example networktree with DASS
data(dass)
## Select depression subscale
nodeVars <- colnames(dass)[(grep("_D", colnames(dass)))]
splitVars <- c("gender","orientation","race","married","engnat")
myTree<-networktree(dass[,nodeVars], dass[,splitVars])
myTree
plot(myTree)



head(dass)


## Example networktree with DASS
data(dass)
## Select depression subscale
nodeVars <- colnames(dass)[(grep("_D", colnames(dass)))]
splitVars <- c("gender","orientation","race","married","engnat")
myTree<-networktree(dass[,nodeVars], dass[,splitVars])
myTree
plot(myTree)

getnetwork

Description

Easily extract a network from one of the nodes in a networktree object

Usage

getnetwork(tree, id = 1L, transform = "detect", verbose = FALSE, ...)
getnetwork(tree, id = 1L, transform = "detect", verbose = FALSE, ...)

Arguments

`tree`	a networktree object
`id`	the node in the tree to extract. Use summary(tree) to see id numbers for each split
`transform`	should stored correlation matrices be transformed to partial correlations or graphical lasso? Can be set to "cor", "pcor", or "glasso". Defaults to automatic detection
`verbose`	should warnings and messages from transformation functions (qgraph) be printed?
`...`	arguments passed to qgraph (e.g., "tuning", "threshold")

Examples


set.seed(1)
d <- data.frame(trend = 1:200, foo = runif(200, -1, 1))
d <- cbind(d, rbind(
  mvtnorm::rmvnorm(100, mean = c(0, 0, 0),
          sigma = matrix(c(1, 0.5, 0.5, 0.5, 1, 0.5, 0.5, 0.5, 1), ncol = 3)),
  mvtnorm::rmvnorm(100, mean = c(0, 0, 0),
          sigma = matrix(c(1, 0, 0.5, 0, 1, 0.5, 0.5, 0.5, 1), ncol = 3))
))
colnames(d)[3:5] <- paste0("y", 1:3)

## Now use the function
tree1 <- networktree(nodevars=d[,3:5], splitvars=d[,1:2])

getnetwork(tree1, id=1)

set.seed(1)
d <- data.frame(trend = 1:200, foo = runif(200, -1, 1))
d <- cbind(d, rbind(
  mvtnorm::rmvnorm(100, mean = c(0, 0, 0),
          sigma = matrix(c(1, 0.5, 0.5, 0.5, 1, 0.5, 0.5, 0.5, 1), ncol = 3)),
  mvtnorm::rmvnorm(100, mean = c(0, 0, 0),
          sigma = matrix(c(1, 0, 0.5, 0, 1, 0.5, 0.5, 0.5, 1), ncol = 3))
))
colnames(d)[3:5] <- paste0("y", 1:3)

## Now use the function
tree1 <- networktree(nodevars=d[,3:5], splitvars=d[,1:2])

getnetwork(tree1, id=1)

Maximum Likelihood Estimation for Multivariate Normal Model

Description

Fit a multivariate normal model without covariates or covariance restrictions. In addition to the (straightforward) parameter estimates the fitted log-likelihood and corresponding score contributions are computed.

Usage

mvnfit(
  y,
  x = NULL,
  start = NULL,
  weights = NULL,
  offset = NULL,
  model = c("correlation", "mean", "variance"),
  ...,
  estfun = FALSE,
  object = FALSE
)
mvnfit(
  y,
  x = NULL,
  start = NULL,
  weights = NULL,
  offset = NULL,
  model = c("correlation", "mean", "variance"),
  ...,
  estfun = FALSE,
  object = FALSE
)

Arguments

`y`	A matrix or data.frame where each row corresponds to a k-dim observation.
`x`	Not used yet
`start`	Not used yet
`weights`	Not used yet
`offset`	Not used yet
`model`	Vector of characters. Specifies which estimated parameters are returned.
`...`	Not used yet
`estfun`	Logical. Should the matrix of score contributions (aka estimating functions) be returned?
`object`	Not used yet

Details

Used internally in when method="mob"

networktree: Partitioning of network models

Description

Computes a tree model with networks at the end of branches. Can use model-based recursive partitioning or conditional inference.

Wraps the mob() and ctree() functions from the partykit package.

Usage

networktree(...)

## Default S3 method:
networktree(
  nodevars,
  splitvars,
  method = c("mob", "ctree"),
  model = "correlation",
  transform = c("cor", "pcor", "glasso"),
  na.action = na.omit,
  weights = NULL,
  ...
)

## S3 method for class 'formula'
networktree(
  formula,
  data,
  transform = c("cor", "pcor", "glasso"),
  method = c("mob", "ctree"),
  na.action = na.omit,
  model = "correlation",
  ...
)
networktree(...)

## Default S3 method:
networktree(
  nodevars,
  splitvars,
  method = c("mob", "ctree"),
  model = "correlation",
  transform = c("cor", "pcor", "glasso"),
  na.action = na.omit,
  weights = NULL,
  ...
)

## S3 method for class 'formula'
networktree(
  formula,
  data,
  transform = c("cor", "pcor", "glasso"),
  method = c("mob", "ctree"),
  na.action = na.omit,
  model = "correlation",
  ...
)

Arguments

`...`	additional arguments passed to `mob_control` (mob) or `ctree_control` (ctree)
`nodevars`	the variables with which to compute the network. Can be vector, matrix, or dataframe
`splitvars`	the variables with which to test split the network. Can be vector, matrix, or dataframe
`method`	"mob" or "ctree"
`model`	can be any combination of c("correlation", "mean", "variance") splits are determined based on the specified characteristics
`transform`	should stored correlation matrices be transformed to partial correlations or a graphical lasso for plotting? Can be set to "cor" (default), "pcor", or "glasso"
`na.action`	a function which indicates what should happen when the data contain missing values (`NA`s).
`weights`	weights
`formula`	A symbolic description of the model to be fit. This should either be of type `y1 + y2 + y3 ~ x1 + x2` with node vectors `y1`, `y2`, and `y3` or `y ~ x1 + x2` with a matrix response y. `x1` and `x2` are used as partitioning variables.
`data`	a data frame containing the variables in the model

References

Jones, P.J., Mair, P., Simon, T., Zeileis, A. (2020). Network trees: A method for recursively partitioning covariance structures. Psychometrika, 85(4), 926-945. https://doi.org/10.1007/s11336-020-09731-4

Examples


set.seed(1)
d <- data.frame(trend = 1:200, foo = runif(200, -1, 1))
d <- cbind(d, rbind(
  mvtnorm::rmvnorm(100, mean = c(0, 0, 0),
          sigma = matrix(c(1, 0.5, 0.5, 0.5, 1, 0.5, 0.5, 0.5, 1), ncol = 3)),
  mvtnorm::rmvnorm(100, mean = c(0, 0, 0),
          sigma = matrix(c(1, 0, 0.5, 0, 1, 0.5, 0.5, 0.5, 1), ncol = 3))
))
colnames(d)[3:5] <- paste0("y", 1:3)

## Now use the function
tree1 <- networktree(nodevars=d[,3:5], splitvars=d[,1:2])

## Formula interface
tree2 <- networktree(y1 + y2 + y3 ~ trend + foo, data=d)

## plot 
plot(tree2)
plot(tree2, terminal_panel = "box")
plot(tree2, terminal_panel = "matrix")


## Conditional version
tree3 <- networktree(nodevars=d[,3:5], splitvars=d[,1:2], 
                     method="ctree")

## Change control arguments
tree4 <- networktree(nodevars=d[,3:5], splitvars=d[,1:2],
                     alpha=0.01)

set.seed(1)
d <- data.frame(trend = 1:200, foo = runif(200, -1, 1))
d <- cbind(d, rbind(
  mvtnorm::rmvnorm(100, mean = c(0, 0, 0),
          sigma = matrix(c(1, 0.5, 0.5, 0.5, 1, 0.5, 0.5, 0.5, 1), ncol = 3)),
  mvtnorm::rmvnorm(100, mean = c(0, 0, 0),
          sigma = matrix(c(1, 0, 0.5, 0, 1, 0.5, 0.5, 0.5, 1), ncol = 3))
))
colnames(d)[3:5] <- paste0("y", 1:3)

## Now use the function
tree1 <- networktree(nodevars=d[,3:5], splitvars=d[,1:2])

## Formula interface
tree2 <- networktree(y1 + y2 + y3 ~ trend + foo, data=d)

## plot 
plot(tree2)
plot(tree2, terminal_panel = "box")
plot(tree2, terminal_panel = "matrix")


## Conditional version
tree3 <- networktree(nodevars=d[,3:5], splitvars=d[,1:2], 
                     method="ctree")

## Change control arguments
tree4 <- networktree(nodevars=d[,3:5], splitvars=d[,1:2],
                     alpha=0.01)

Plotting 'networktree' objects

Description

Wraps plot.party to plot a tree model with networks on the ends. Networks are plotted with qgraph, and additional arguments are passed there

Usage

## S3 method for class 'networktree'
plot(
  x,
  terminal_panel = NULL,
  transform = NULL,
  layout = "lock",
  sdbars = FALSE,
  tnex = 3,
  partyargs = list(),
  na.rm = TRUE,
  ...
)
## S3 method for class 'networktree'
plot(
  x,
  terminal_panel = NULL,
  transform = NULL,
  layout = "lock",
  sdbars = FALSE,
  tnex = 3,
  partyargs = list(),
  na.rm = TRUE,
  ...
)

Arguments

`x`	an object of type 'networktree'
`terminal_panel`	an optional panel function of the form function(node) plotting the terminal nodes. Alternatively, a panel generating function of class "grapcon_generator" that is called with arguments x and tp_args to set up a panel function. Or, a character choosing one of the implemented standard plots `"graph"`, `"box"`, `"matrix"` or `"bar"`. The default (`NULL`) chooses an appropriate panel function depending on the "model" argument.
`transform`	"cor", "pcor", or "glasso". If set to NULL, transform detected from x
`layout`	network layout, passed to qgraph. Default "lock" computes spring layout for the full sample and applies this to all graphs
`sdbars`	if using a barplot, should std deviation error bars be plotted?
`tnex`	terminal node extension (passed to plot.party). To make the terminal plots bigger, increase this value.
`partyargs`	additional arguments (list format) passed to `partykit::plot.party` plotting function that takes partitioned data as input
`na.rm`	should NA values be removed prior to calculating relevant parameters?
`...`	additional arguments passed to qgraph or barplot

Predict 'networktree' objects

Description

Wraps predict.party

Usage

## S3 method for class 'networktree'
predict(object, newdata = NULL, type = c("node", "parameter"), ...)
## S3 method for class 'networktree'
predict(object, newdata = NULL, type = c("node", "parameter"), ...)

Arguments

`object`	a fitted 'networktree'
`newdata`	An optional data frame in which to look for variables with which to predict. If omitted, the fitted values are used.
`type`	"node", or "parameter". Specifies whether to predict nodes (return value is a vector) or parameters (matrix).
`...`	not used

Printing 'networktree' objects

Description

Wraps print.modelparty to print a tree model with networks on the ends.

Usage

## S3 method for class 'networktree'
print(x, parameters = FALSE, FUN = NULL, ...)
## S3 method for class 'networktree'
print(x, parameters = FALSE, FUN = NULL, ...)

Arguments

`x`	an object of type 'networktree'
`parameters`	print parameters for each partition? See getnetwork function for extracting parameters conveniently
`FUN`	only evaluated if parameters=TRUE, passed to print.modelparty
`...`	additional arguments passed print.modelparty

Ten Item Personality Questionnaire

Description

This dataset includes 1899 online participants who participated in a questionnaire available through the Open Source Psychometrics Project (https://openpsychometrics.org/), an organization that maintains an open website for the public to take psychometric tests for educational and entertainment purposes

Usage

tipi
tipi

Format

a dataframe. Columns represent questionnaire items and rows represent individuals

Details

The Ten Item Personality Questionnaire (TIPI) is a brief inventory of the Big Five personality domains. Each personality domain is assessed with two items. One item measures the domain normally and the other item measures the domain in reverse (e.g., "reserved, quiet" for reverse extraversion).

Labels for TIPI items in this dataset correspond to the first letter of each Big Five personality domain (Extraversion, Neuroticism, Conscientiousness, Agreeableness, and Openness to experience), with the character “r” indicating items that measure the domain in reverse.

Also includes demographics such as education level, rearing environment (urban/suburban/rural), gender, English as a native language, age, religion, sexual orientation, race, voting status, marriage status, and number of children in one's family during childhood.

The dataset is publicly available at http://openpsychometrics.org/_rawdata/GCBS.zip and can be cited as:

OpenPsychometrics (2019). Generic Conspiracist Beliefs Scale Survey. Retrieved from http://openpsychometrics.org/_rawdata/GCBS.zip

Examples

head(tipi)


## Example networktree with TIPI
data(tipi)
nodeVars <- c("E","A_r","C","N","O","E_r","A","C_r","N_r","O_r")
splitVars <- c("gender","education","engnat")
myTree<-networktree(tipi[,nodeVars], tipi[,splitVars])
myTree
plot(myTree)



head(tipi)


## Example networktree with TIPI
data(tipi)
nodeVars <- c("E","A_r","C","N","O","E_r","A","C_r","N_r","O_r")
splitVars <- c("gender","education","engnat")
myTree<-networktree(tipi[,nodeVars], tipi[,splitVars])
myTree
plot(myTree)

Workaholism and Psychiatric Symptoms

Description

This dataset includes 16,426 workers who were assessed on symptoms of psychiatric disorders (ADHD, OCD, anxiety, depression) and workaholism.

Usage

workaholic
workaholic

Format

a dataframe. Columns represent symptoms and rows represent individuals

Details

Scales: Adult ADHD Self-Report Scale, Obsession-Compulsive Inventory-Revised, Hospital Anxiety and Depression Scale, and the Bergen Work Addiction Scale.

Also includes demographics such as age, gender, work status, position, sector, annual income.

The dataset is publicly available at https://doi.org/10.1371/journal.pone.0152978 and can be cited as:

Andreassen, C. S., Griffiths, M. D., Sinha, R., Hetland, J., & Pallesen, S. (2016). The relationships between workaholism and symptoms of psychiatric disorders: a large-scale cross-sectional study. PloS One, 11, e0152978.

Examples

head(workaholic)


## Example networktree with OCI-R scale
data(workaholic)
nodeVars <- paste("OCIR",1:18,sep="")
splitVars <- c("Workaholism_diagnosis","Gender")
myTree<-networktree(workaholic[,nodeVars], workaholic[,splitVars])
myTree
plot(myTree)



head(workaholic)


## Example networktree with OCI-R scale
data(workaholic)
nodeVars <- paste("OCIR",1:18,sep="")
splitVars <- c("Workaholism_diagnosis","Gender")
myTree<-networktree(workaholic[,nodeVars], workaholic[,splitVars])
myTree
plot(myTree)

Package 'networktree'

Help Index

comparetree

Description

Usage

Arguments

Examples

cortrafo

Description

Usage

Arguments

Depression Anxiety and Stress Scale

Description

Usage

Format

Details

Examples

getnetwork

Description

Usage

Arguments

Examples

Maximum Likelihood Estimation for Multivariate Normal Model

Description

Usage

Arguments

Details

networktree: Partitioning of network models

Description

Usage

Arguments

References

Examples

Plotting 'networktree' objects

Description

Usage

Arguments

Predict 'networktree' objects

Description

Usage

Arguments

Printing 'networktree' objects

Description

Usage

Arguments

Ten Item Personality Questionnaire

Description

Usage

Format

Details

Examples

Workaholism and Psychiatric Symptoms

Description

Usage

Format

Details

Examples