Expanding Network Analysis Tools in Psychological Networks: Minimal Spanning Trees, Participation Coefficients, and Motif Analysis Applied to a Network of 26 Psychological Attributes
Table 4
An overview of the three methods.
MINIMUM SPANNING TREE
Recommended for network
Dense networks and/or networks with many small edges
Important procedure steps
(1) Selecting a distance measure: (i) Gower’s distance (ii) Distance inversely proportional to the shared variance (2) Centrality analysis (by inspection or/and with standard centrality measures computed)
Output
MST – the filtered network
Methodological considerations
Distance measures (i) and (ii) will produce different MSTs if a network has negative
Other analytical possibilities
Looking at MST branches as communities Using MST to test the robustness of the network estimation of the most essential edges MST can include distance metric as weight for further analysis
Effect of reverse-coding variables
(i) affecte (ii) not affected
Hypothesis/Research question
RQ: Which node is the most central? Which (overlapping) communities exist in the network?
PARTICIPATION COEFFICIENT as a corrective
Recommended for network
(1) Pre-existing differences in the kinds of nodes (2) Networks with communities
Important procedure steps
If (1) is true: (a) Defining the node groups (b) Calculating PC (c) Choosing the centrality measure to be corrected with PC (optional) If (2) is true: (a) Data-driven detection of communities (b) Calculating PC (c) Choosing the centrality measure to be corrected with PC (d) Comparing the rank order of chosen centrality measure before and after the correction
Output
PC values for each node (in (1)(b) and (2)(b); The corrected centrality measure (for (1)(c) and (2)(c))
Methodological considerations
Communities should not overlap ((1) and (2)) (1) a Group affiliations may be ambiguous (2) a Decision about appropriate community detection algorithm
Other analytical possibilities
PC version that treats positive and negative edges separately
Effect of reverse-coding variables
Not affected if signs are not taken in the account when calculating PC
Hypothesis/Research question
centrality measure is not affected by (pre-existing or data-driven) communities
MOTIF ANALYSIS
Recommended for network
(1) Not for networks with small number of nodes and/or very low density (2) Has negative and positive ties (3) If weighted, additional steps in the procedure
Important procedure steps
(a) Defining motifs of interest and the null model (b) Motif identification and frequency (c) Significance testing of motif frequency If (3) is true: (d) Motif intensity (and coherence) (e) Significance testing of motif intensity (and coherence)
Output
Identified motifs; Motif Frequency; P-values for frequencies; Motif intensity (and coherence); P-values for the intensities (and coherence)
Methodological considerations
The definition of the null (reference) model
Other analytical possibilities
Other motif structures, e.g. that include more nodes
Effect of reverse-coding variables
Identified motifs and motif frequency will be differen, but onclusions about significance (of frequency, intensity, and coherence) will tend to converge
Hypothesis/Research question
Many research questions and hypotheses possible. In this study: Signed edges will tend to cluster in the line with what is observed in social networks and correlational networks (balance theory, forbidden triads, imbalanced triplets):= P0P, NNN, and PPN will be less frequent than expected by chance. RQ1 = Do same pattern of results holds true when only relatively stronger motifs are considered? RQ2= Are Intensity and Coherence measure following the same pattern of results as the frequency of motifs?
his should be the case, but there is a possibility that distances related to negative edges are present in a network in such a way that will not affect the MST construction, e.g. a weak negative tie that exists among two peripheral nodes that have ties to other more central node.