We address this problem by measuring *exposure* to the program through the network of commuting flows between all neighbourhoods before and after the program was rolled out. For this, we collected detailed commuting data for everyone in our sample, which is rare in this context.

Although the resulting approach looks similar to a shift-share design used in trade and migration economics, it’s rarely used in RCTs of social programs. Our approach yields significantly larger estimates than commonly used approaches. Wages increased by nearly 20% citywide.

A core insight of the paper is our approach to estimating those wage effects that addresses spatial spillovers -- potentially useful for researchers studying other social protection and labour policies.

We use data from a large survey of households, a randomized experiment at scale, and a spatial equilibrium model. We quantify i) the direct effects of program earnings on beneficiary households, ii) equilibrium wage effects, and iii) improvements in urban amenities from the work.