| dc.description.abstract | Background - Missing data is a problem that is common. It affects the accuracy and introduces
biases in social network analysis, which can have a significant effect on the interpretation of
findings. Relational event history (REH) data, a type of social network data, is becoming
increasingly available due to new technological developments and can enhance the
understanding of dynamic social networks. However, research on handling missing values in
social network data is limited and statistical tools for incomplete REH data are underdeveloped.
This paper focuses on using multiple imputation to handle missing values within REH data.
Methods – Relational event history model analysis is first performed on the fully observed
dataset to produce true estimates. Next, a simulation study is conducted to introduce
missingness to this fully observed dataset, assuming missing completely at random (MCAR) and
right-tailed missing at random (MAR). After multiple imputation, the relational event model is
applied on the simulations and the results are compared to the analysis of the fully observed
dataset.
Results – The results of the relational event model of the simulations and the true estimates
show inconsistency in the significance of the results. The simulations generally have a low bias,
good coverage rate an low average width. A higher proportion of missingness resulted in a
decrease in the performance. Multiple imputation thus produces unbiased inferences under the
MCAR and MAR mechanism, however unexpected significant results are found.
Conclusion – This study provides insights into the use of multiple imputation for producing valid
inferences when applied on REH data. It shows that under the assumption of MCAR and MAR,
multiple imputation can be a valid method for missing data in REH data when the percentage of
missingness is not too high. Further research is needed confirm an expand upon the results
obtained in this study. | |
