Random errors in egocentric networks.

Area of Science:

• Social network analysis
• Sociology
• Computational social science

Background:

• Egocentric network analysis traditionally focuses on systematic errors from human memory and survey design.
• Random error analysis in egocentric networks remains understudied.
• Existing research lacks a comprehensive understanding of how random errors impact network measures and statistical models.

Purpose of the Study:

• To analyze the effects of random errors on egocentric network analysis.
• To investigate the impact of using egocentric network measures as predictors in linear models.
• To quantify the consequences of false-positive and false-negative errors in egocentric network data.

Main Methods:

• Simulation analysis using a ground truth egocentric network dataset derived from Facebook friendships.
• Examination of standard network measures under varying error rates.
• Assessment of linear models incorporating egocentric network measures with simulated errors.

Main Results:

• Moderate error rates (5-20%), common in ego network data, significantly misestimate network properties.
• The presence of random errors leads to substantial misestimation of regression parameters when using egonet measures as predictors.
• Both false-positive and false-negative errors demonstrably impact egocentric network analysis and subsequent statistical modeling.

Conclusions:

• Random errors pose a significant threat to the validity of egocentric network analysis and associated statistical inferences.
• Researchers must account for potential random errors in ego network data to avoid misleading conclusions.
• Future research should focus on developing methods to mitigate or correct for random errors in egocentric network studies.