| Title: | Recursive Partitioning of Network Models |
| Version: | 1.0.1 |
| Date: | 2021-2-4 |
| 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>. |
| Depends: | R (≥ 3.5.0) |
| License: | GPL-2 | GPL-3 |
| Encoding: | UTF-8 |
| LazyData: | true |
| Imports: | partykit, qgraph, stats, utils, Matrix, mvtnorm, Formula, grid, graphics, gridBase, reshape2 |
| RoxygenNote: | 7.1.1 |
| Suggests: | R.rsp, knitr, rmarkdown, fxregime, zoo |
| URL: | https://paytonjjones.github.io/networktree/ |
| BugReports: | https://github.com/paytonjjones/networktree/issues |
| NeedsCompilation: | no |
| Packaged: | 2021-02-04 14:10:22 UTC; payton-jones |
| Author: | Payton Jones 
[aut, cre],
Thorsten Simon
[aut],
Achim Zeileis
[aut] |
| Maintainer: | Payton Jones <paytonjjones@gmail.com> |
| Repository: | CRAN |
| Date/Publication: | 2021-02-04 15:30:06 UTC |
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",
...
)
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)
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, ...)
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
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)
getnetwork
Description
Easily extract a network from one of the nodes in a networktree object
Usage
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)
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
)
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",
...
)
Arguments
... |
additional arguments passed to |
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 ( |
weights |
weights |
formula |
A symbolic description of the model to be fit. This
should either be of type |
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)
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,
...
)
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 |
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 |
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"), ...)
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, ...)
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
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)
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
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)