Exogenous, unobserved factors often confound the effects of alliance networks. More capable farmers might be less likely to exit and more likely to have a large number of alliances. In this case the negative correlation between alliance network size and exit likelihood is due to the unobserved confounder--farmer ability--not the effect of network size on exit likelihood. Recognizing the endogeneity of alliance network size when determining a farm’s survival likelihood, we employ an empirical model that accounts for the bias caused by unobserved effects. We account for time-invariant unobserved effects with individual fixed effects. We control for county-level confounding factors with a time-varying county effect. Finally, we address unobserved, time-varying confounding factors by including alliance network size as a control variable. As a control variable, alliance network size captures all remaining confounding factors. We estimate the alliance network effect by interacting the alliance network size with a weather index. In this way we isolate the “insurance effect” of having a large alliance network. We answer the question, “given a farmer’s alliance network size, what is the likelihood of exiting when experiencing an extreme weather event.” We employ USDA-Farm Services Agency administrative data to measure farmer alliances and farm entry and exit. The data provide detailed information on every farm-payment recipient since 1990, including location and organizational structure. We use the organizational-structure information to map farmer alliances and characterize alliance networks according to their size (number of nodes and number of members in each node), contractual choice (i.e. joint ventures), geographic scope, and commodity scope. Our preliminary results confirm the positive correlation between network size and farm survival.