Which estimation method to choose in network psychometrics? Deriving guidelines for applied researchers

Area of Science:

• Psychological research
• Network analysis
• Statistical modeling

Background:

• Gaussian graphical models (GGMs) are increasingly used in psychology.
• Existing estimation methods vary in performance and suitability for psychological data challenges.
• No consensus exists on the best GGM estimation procedure for specific research settings.

Purpose of the Study:

• To compare the performance of various GGM estimation algorithms for Gaussian and skewed ordered categorical data.
• To provide concrete guidelines for applied researchers on selecting appropriate estimation methods.
• To address challenges in psychological research, such as weak edge detection and limited sample sizes.

Main Methods:

• A large-scale simulation study involving 564,000 datasets.
• Evaluation of 60 different performance metrics across diverse settings.
• Comparison of algorithms suitable for both continuous and ordered categorical data.

Main Results:

• Performance of GGM estimation algorithms differs significantly across settings.
• A trade-off exists between statistical discovery (sensitivity, correlation) and caution (specificity, precision).
• Optimal algorithm choice depends on the specific research question and data properties.

Conclusions:

• Selecting the best GGM estimation method requires careful consideration of research goals and data characteristics.
• Achieving perfect replicability, balancing discovery and caution, is challenging.
• Guidelines are provided to aid researchers in choosing appropriate GGM estimation techniques for psychological network analysis.