It is increasingly common for firms to break up their production process across different regions and countries. While the existing literature credits improvements in communication technology with enabling fragmentation, there is practically no empirical evidence on the determinants of firms' sourcing strategies. In this dissertation, I document the roles of communication technology, distance to suppliers, and labor cost differences in a firm's decisions about i) whether to fragment production across distinct geographic locations; ii) whether to offshore fragmented production; and iii) how much to offshore. To do so, I construct an original data set about U.S. plants' and firms' fragmentation choices. The data yield a rich set of new facts about firms' sourcing decisions which I incorporate into a theoretical model. I then use the new data to test the model's predicted equilibrium relationships and to assess the relative importance of communication technology, distance, and labor costs in firms' sourcing strategies.

I construct the fragmentation dataset using new information about U.S. manufacturing plants' decision to contract for manufacturing services from domestic or foreign suppliers in 2007. Chapter two describes these data and presents ten new stylized facts about plants' and firms' sourcing decisions. Most notably, the data show that fragmentation: (i) is 13 times more prevalent from domestic than foreign suppliers; (ii) is done mostly by larger and more productive plants; and (iii) varies substantially within industries. The new facts are also consistent with the premise that firms fragment to access cheaper labor but that they pay a fixed cost to do so.

In the third chapter, I incorporate these new facts into a model of heterogeneous firms that decide where to locate the various stages of their production process. Firms fragment production to access cheaper labor, but breaking up production is costly. Firms incur a fixed cost to establish a supply network and additional per-task costs to coordinate production and transport inputs. The fixed costs deliver standard productivity sorting predictions, while the marginal costs add a new dimension of heterogeneity in firms' organization of production. In particular, firms with access to better communication technology, or in locations closer to their potential suppliers, will find fragmentation relatively more profitable. The model also shows that firms in high wage locations have more to gain from fragmentation, while firms in low wage states must offshore to access cheaper wages.

The fourth chapter provides an empirical assessment of the model's predicted equilibrium relationships. I estimate the relative importance of labor cost savings, technology, and distance to suppliers in a plant's decision to fragment production. The estimates indicate that plant use of electronic networks (as a proxy for communication technology) is associated with an 18 percentage point increase in the probability of fragmentation, and a ten point increase in the probability of locating fragmented production offshore. While wage differences and distance to suppliers also have statistically significant relationships with plants' sourcing strategies, communication technology accounts for five times more of the explained variation than wages and distance combined. In contrast, for the decision about how much to offshore, wage differences are relatively more important than distance, and technology explains almost none of the observed variation. Because plant technology may be endogenous, I estimate the differential impact of plants' use of electronic networks on fragmentation in industries whose production process can be codified electronically more easily. As expected, plant use of electronic networks has a bigger impact on fragmentation in industries that are better able to specify production processes electronically. However, plants that use networks in industries with high electronic codifiability are less likely to locate their fragmented production offshore. Estimates from firm-country level import data suggest that successful electronic communication depends upon suitable technology in the sourcing location. The results support the premise that technology facilitates production fragmentation, but uncover substantial heterogeneity in technology's effectiveness across firms, industries, and sourcing locations.