Entrepreneurship, the initial labor force, and the location of new firms

The definition of ‘spinoffs’ used here follows the one adopted by Garvin (1983) and Klepper (2009)—i.e. independent startups founded by one or more individuals straight after leaving employment in an incumbent (parent company) in the same industry. These startups have no equity relationship with incumbents. This event is akin to the “spinout” concept described by Agarwal et al. (2004).

Throughout the remainder of the text we will refer to the location of the incumbent firm—i.e., where its founder previously worked—as the home region of the founder and also the new firm; we will refer to the industry of the incumbent firm—i.e., the industry in which the founder previously worked—as the home industry of the founder and the new firm.

We assume that any region where the firm might locate has enough suitable workers matching its needs, so agglomeration costs or scarcity of labor do not play a determinant role in the selection or exclusion of a region.

While we acknowledge that this assumption is somewhat unrealistic, we submit that the everyday contact with a co-worker should provide significantly greater knowledge about performance than second-hand information acquired in industry and professional networks.

Tracing the history of founders and employees required us to search earlier years. The industry codes in 1995 and 1996 are the same but this was not true for some industry codes before 1995, which required the use of an algorithm (based on how the majority of firms changed industry codes from 1994 to 1995) to match industry codes. To minimize errors while preserving observations we included only entrants from 1996 onwards.

We included entrants in 1999 even though we had no information on their third-year hires because the number of hires in the third year is small relative to the prior two years and we wanted to keep the sample of entrants as large as possible.

The total number of new entrants in our sample period was 88,981, so our criteria resulted in retaining about 11% of the firms. We compared our sample with the full sample. The distribution of entrants by location and industry was not significantly different for the two samples. The main difference was that average number of initial members of the firm, including owner-employees, was greater (by one) than the full sample, reflecting the requirement that firms in our sample had at least one employee in their first year.

This is less than the sample of all hires (for instance, in the first year, firms hired 32,968 workers whereas our sample in the first year is composed of 27,282 workers). Some workers had to be removed from the sample because the firm failed to input the workers’ correct social security number.

Currently, Portugal (excluding islands) is subdivided into 278 counties. Three counties were introduced in 1998 and correspond to subdivisions of previously existing counties. Throughout the analysis we merged the new counties with the older ones and so considered only the 275 counties that existed as of 1996.

We traced the work history of every founder from 1987 (the first year of the dataset) onwards to determine tenure in the home region.

In Portugal, the time to commute increases discontinuously with the number of county borders one has to cross. This is true both for commuters using public transportation and for commuters that use their own vehicle. Public transportation systems are usually county-bounded and integration mechanisms are time consuming. On the other hand, time consuming traffic lines are usual in the most densely populated metropolitan areas. For instance, a 15 km travel from a county nearby Lisboa to the county of Lisboa may take 15 min on a Sunday morning and one hour or more in rush hour, while inside Lisboa public transportation systems such as the subway ensure those discrepancies do not exist. This explains the preference of firms to stay so close to their home county.

We estimated additional models using a variable Dist, which is the number of kilometers between the center of county j and the center of the firm’s home county. These models are available upon request as an appendix to the paper.

Specifically, the estimated coefficients for variables _Homex4I, _Homex3I, and _Homex2I are significantly different.

Consistent with other studies, in our sample firm longevity was positively and significantly related to the initial number of employees of the firm. However, when Firmtenure, Highlevel, and Multiplefounders were also allowed to affect firm longevity, the effect of the initial number of employees on the firm hazard was no longer significant, while the effects of Firmtenure, Highlevel, and Multiplefounders were. Accordingly, we did not include the initial number of employees of the firm in our analyses, although its inclusion has little effect on our estimates.

The patterns are similar if we maintain the industry distinctions in Model 2, but we report the estimates for the simpler model to facilitate comparisons with the patterns for hires in years 2 and 3 reported below.

We also included a variable equal to the number of employees of the parent firm in the year prior to the new firm’s entry to test how the size of the pool of old colleagues influenced the hiring of old colleagues. Curiously, the coefficient estimate of this variable was negative and significant. We added a quadratic term for the number of employees of the parent firm, which had a positive and significant coefficient estimate, indicating that possibly the hiring of old colleagues increased with the pool to choose from. One interpretation of these findings is that small firms facilitate the kind of interactions among employees that motivate the hiring of old colleagues. The inclusion of the number of employees of the parent firm did not alter the patterns regarding the other coefficient estimates.

It should be noted that H4IR is significantly different from H3IR and H2IR.

In terms of significant differences, _H4IR, _H3IR, _H2IR are significantly different than _HOIR.

Standard errors are in parentheses.

Given that the percentages of hires that were old colleagues and workers from the home industry and region are similar for firms that entered in their home county but not their home industry, and home industry but not home county (in contrast to what we predicted), it might have been expected that _HR and _H4I would both be negative and significant, and similar in magnitude. We probed this by allowing C1, C2, C3, WHIR1, WHIR2, and WHIR3 in Model 3 of the worker hazard to have different effects for the four groups of firms (corresponding to whether they entered their home county and/or their home industry). These estimates revealed that, except for the old colleagues hired in year 1, the hazards of the old colleagues and workers from the home industry and region were not nearly as low for the firms that entered their home industry and not home county compared with the other three groups of firms. This result suggests that firms that do not enter in their home region, whether they entered their home industry or not, do not benefit by hiring old colleagues and workers from the home industry and region as much as firms that enter in their home region.

When just C1, C2, C3 are added, _H4IR, _HR and _H4I equal -0.044 (0.029), -0.022 (0.029), 0.023 (0.024), and when WHIR1, WHIR2, and WHIR3 are added as well then _H4IR, _HR and _H4I equal -0.023 (0.029), -0.018 (0.029), 0.029 (0.039). This suggests that hiring of old colleagues is the most important factor differentiating the worker hazards of firms, but hiring workers from the home industry and region is also important. As mentioned in the previous footnote, these estimates are available from the authors upon request.