Routinizability. We have also looked at how the di erent tasks are organized across hierarchical lev- els within groups. To that extent, we have exploited a routinizability index sourced by a questionnaire developed for the Princeton Data Improvement Initiative (PDII henceforth) and discussed by Xxxxxxx and Xxxxxxx (2013). In the PDII survey, respondents are asked to rank their tasks on a scale from 1 to 4, where 1 represents a high share of workday involved in short and repetitive tasks, while 4 represents a small share of workday involved in short, repetitive tasks. Therefore, a low number on this scale is associated with a high degree of routinizability.16 We then compute the mean answer by the 3-digit NAICS 2002 sector in which respondents are employed, and use it as a measure of routinizability of the industry in which the subsidiary is active. Out of a sample of 3, 483, 354 observations with information on NAICS 2002 sector, we manage to match the routinizability index for 3, 478, 192 subsidiaries. In Table 5 we provide initial evidence on routinizability. Speci cally, we regress the 16 Speci cally, we refer to question Q25b of the PDII survey. Σ routinizability associated to each subsidiary on the position of the same subsidiary on the hierarchy, controlling for both parent xed e ect and a xed e ect for the country in which the subsidiary is located. The detailed speci cation is as follows: Routinizabilityspic = βll + FEp + FEc + εspic l∈L where (s, p, i, c) denote respectively subsidiary, parent company, industry and country in which the subsidiary is located, l indicates the hierarchical level at which the subsidiary is placed, L indicates the maximum hierarchical level of the BG, FEp is the parent xed e ect and FEc is the xed e ect for the country in which the subsidiary is located. Please bear in mind that we unify all levels higher than 7 under the same category, 8, so that L ≥ 8.17 To ease the interpretation of this stylized fact, we plot the regression xxx - cients and their 95% con dence intervals in Figure 6. This preliminary evidence suggests that, within groups (as we employ parent FE), the complexity of tasks performed by sub- sidiaries positioned in each layer decreases as layers are added to the rst one, since a small value of the routinizability index is associated to a high degree of routinizability of tasks. Again, this is in line with insights from the knowledge-based theory of xxxxxxxxxxx, hinting at the idea that lower positions in the hierarchy ...
