ResIN {ResIN}

ResIN

Description

Performs Response Item-Network (ResIN) analysis

Usage

ResIN(
  df,
  node_vars = NULL,
  cor_method = "auto",
  weights = NULL,
  method_wCorr = "Polychoric",
  poly_ncor = 2,
  remove_negative = TRUE,
  EBICglasso = FALSE,
  EBICglasso_arglist = NULL,
  node_covars = NULL,
  node_costats = NULL,
  network_stats = FALSE,
  cluster = FALSE,
  seed = 42
)

Arguments

df

A data-frame object containing the raw data.

node_vars

An optional character string detailing the attitude item columns to be selected for ResIN analysis (i.e. the subset of attitude variables in df).

cor_method

Which correlation method should be used? Defaults to "auto" which applies the cor_auto function from the qgraph package. Possible arguments are "auto", "pearson", "kendall", and "spearman".

weights

An optional continuous vector of survey weights. Should have the same length as number of observations in df. If weights are provided, weighted correlation matrix will be estimated with the weightedCorr function from the wCorr package.

method_wCorr

If weights are supplied, which method for weighted correlations should be used? Defaults to "Polychoric". See wCorr::weightedCorr for all correlation options.

poly_ncor

How many CPU cores should be used to estimate polychoric correlation matrix? Only used if cor_method = "polychoric".

remove_negative

Should all negative correlations be removed? Defaults to TRUE (highly recommended). Setting to FALSE makes it impossible to estimate a force-directed network layout. Function will use igraph::layout_nicely instead.

EBICglasso

Should a sparse, Gaussian-LASSO ResIN network be estimated? Defaults to FALSE. If set to TRUE, EBICglasso function from the qgraph packages performs regularization on (nearest positive-semi-definite) ResIN correlation matrix.

EBICglasso_arglist

An argument list feeding additional instructions to the EBICglasso function if EBICglasso is set to TRUE.

node_covars

An optional character string selecting quantitative covariates that can be used to enhance ResIN analysis. Typically, these covariates provide grouped summary statistics for item response nodes. (E.g.: What is the average age or income level of respondents who selected a particular item response?) Variable names specified here should match existing columns in df.

node_costats

If any node_covars are selected, what summary statistics should be estimated from them? Argument should be a character vector of the same length of node_covarsand call a base-R function. (E.g. "mean", "median", "sd"). The first element in node_costats specifies the summary statistic extracted from the first element in node_covars, and so on.

network_stats

Should common network structuration and centralization metrics be extracted? Calls qgraph::centrality_auto and DirectedClustering::ClustF to the ResIN graph object to extract network average betweenness, closeness, strength centrality (mean) and centralization scores (sd). Also estimates network expected influence, average path length, and global clustering coefficients.

cluster

Optional, should community detection be performed on item response network? Defaults to FALSE. If set to TRUE, performs "cluster_leading_eigen" function from the igraph package and stores results in node_frame.

seed

Random seed for force-directed algorithm.

Value

A list object containing the ResIN adjacency matrix (adj_matrix), a numeric vector detailing which item responses belong to which item (same_items), a ggplot-ready edge-list type dataframe (edgelist_frame), a node-level dataframe (node_frame), a vector with the optional graph structuration (graph_structuration) and centralization (graph_centralization) statistics, as well as the dummy-coded item-response dataframe (df_dummies).

Examples

## Load the 12-item simulated Likert-type ResIN toy dataset
data(lik_data)
library(ggplot2)

# Apply the ResIN function to toy Likert data:
output <- ResIN(lik_data, cor_method = "spearman", network_stats = TRUE, cluster = TRUE)

# Create a basic outcome plot with ggplot
output$edgelist_frame <- output$edgelist_frame[order(output$edgelist_frame$Strength,
                                                 decreasing = FALSE), ]
ResIN_plot <- ggplot2::ggplot(output$edgelist_frame)+
  geom_curve(data = output$edgelist_frame, aes(x = from.x, xend = to.x, y = from.y,
                                             yend = to.y, linewidth = weight,
                                             color = Strength), curvature = 0.2)+
  geom_point(aes(x = from.x, y = from.y, shape = as.factor(cluster)), size = 8)+
  geom_point(aes(x = to.x, y = to.y), size = 8)+
  geom_text(data = output$edgelist_frame, aes(x = from.x, y = from.y, label = from),
            size = 3, color = "white")+
  geom_text(data = output$edgelist_frame, aes(x = to.x, y = to.y, label = to),
            size = 3, color = "white")+
  ggtitle("ResIN example  plot")+
  theme_dark()+
  theme(axis.text.x = element_blank(), axis.title.x = element_blank(),
        axis.text.y = element_blank(), axis.title.y = element_blank(),
        axis.ticks = element_blank(), panel.grid.major = element_blank(),
        panel.grid.minor = element_blank(), legend.position = "none",
        legend.text = element_blank(), plot.title = element_text(hjust = 0.5))

ResIN_plot