This function compares correlations in a psychometric network in order to identify nodes which most likely measure the same underlying construct (i.e., are colinear)
goldbricker(data, p = 0.05, method = "hittner2003", threshold = 0.25,
corMin = 0.5, progressbar = TRUE)a data frame consisting of n rows (participants) and j columns (variables)
a p-value threshold for determining if correlation pairs are "significantly different"
method for comparing correlations. See ?cocor.dep.groups.overlap for a full list
variable pairs which have less than the threshold proportion of significantly different correlations will be considered "bad pairs"
the minimum zero-order correlation between two items to be considered "bad pairs". Items that are uncorrelated are unlikely to represent the same underlying construct
logical. prints a progress bar in the console
goldbricker returns a list of class "goldbricker" which contains:
$proportion_matrix - a j x j matrix of proportions. Each proportion signifies the amount of significantly different correlations between the given node pair (j x j) $suggested_reductions - a vector of "bad pairs" (names) and their proportions (values) $p - p value from input $threshold - threshold from input
In a given psychometric network, two nodes may be redundantly measuring the same underlying construct. If this is the case, the correlations between those two variables and all other variables should be highly similar. That is, they should correlate to the same degree with other variables.
The cocor package uses a p-value threshold to determine whether a pair of correlations to a third variable are significantly different from each other. Goldbricker wraps the cocor package to compare every possible combination of correlations in a psychometric network. It calculates the proportion of correlations which are significantly different for each different pair of nodes.
Using the threshold argument, one can set the proportion of correlations which is deemed "too low". All pairs of nodes which fall below this threshold are returned as defined "bad pairs".
Pairs can then be combined using the reduce_net function
Note: to quickly change the threshold, one may simply enter an object of class "goldbricker" in the data argument, and change the threshold. The p-value cannot be modified in the same fashion, as re-computation is necessary.
# NOT RUN {
gb_depression <- goldbricker(depression, threshold=0.5)
reduced_depression <- net_reduce(data=depression, badpairs=gb_depression)
## Set a new threshold quickly
gb_depression_60 <- goldbricker(data=gb_depression, threshold=0.6)
# }
Run the code above in your browser using DataCamp Workspace