Identification of local interaction models with imperfect location data

This paper considers the problem of estimating a local interaction model defined at the level of individual agents, in the absence of perfect information about agent locations in the relevant socio-economic space. We consider two types of data limitations: one in which individual locations are measured with error, the other in which location information is correct but only up to some spatial region (e.g., census tracts or zip codes). Our results indicate that the model parameters are not identified almost surely in the former situation, whereas identification is preserved in the latter. We also suggest a potential estimation strategy that can yield useful information about the likely range of parameters even when point identification fails.