Crawling subsampling for multivariate spatial autoregression model in large-scale networks (Q2233551)

Usually, multivariate spatial autoregression models are used to analyze the autocorrelation among multiple responses in network data analysis. In this paper, the authors use crawling subsampling for a multivariate spatial autoregression model in large-scale networks. They construct a least-squares objective function on the sampled data and investigate the sample size condition and asymptotic properties of the least-squares estimators. To validate their results, the authors compare their approach with three sampling methods including simple random sampling (SRS), snowball sampling (SN), and Metropolis-Hastings random walk (MHRW) on three different networks: dyad independence network, stochastic block network and power-law distribution network.