The literature has shown that network architecture depends crucially on whether links are formed unilaterally or bilaterally, that is, on whether the consent of both nodes is required for a link to be formed. They propose a test of whether network data is best seen as an actual link or willingness to link and, in the latter case, whether this link is generated by an unilateral o rbilateral link formation process. They illustrate this test using survey answers to a risk-sharing question inTanzania. They find that the bilateral link formation model fits the data better than the unilateral model, but the data are best interpreted as willingness to link rather than an actual link. They then expand the model to include self-censoring and find that models with self-censoring fit the data best. [Working Paper No. 236]
