Localized Spillovers and Knowledge Flows: How Does Proximity Influence the Performance of Plants?

abstract

By means of a unique longitudinal database with information on all industrial plants and employees in the Swedish economy, this article analyzes how geographic proximity influences the impact of spillovers and knowledge flows on the growth in productivity of plants. Concerning the effects of spillovers, it shows that the density of economic activities contributes mainly to the performance of plants within a short distance and that the composition of economic activities is more influential farther away. Regarding the influence of the local industrial setup, proximity increases the need to be located near different, but related, industries, whereas increased distance implies a greater effect of intraindustry spillovers. The analyses also demonstrate that knowledge flows via the mobility of skilled labor are primarily a subregional phenomenon. Only inflows of skills that are related to the existing knowledge base of plants and come from fewer than 50 kilometers away have a positive effect on the performance of plants. Concerning outflows of skills, the results indicate that it is less harmful for a dispatching plant if a former employee remains within the local economy rather than leaves it for a job in another part of the national economy.

Key Words:

• agglomeration economies
• knowledge spillovers
• labor mobility
• plant performance
• geographic proximity
• related variety

Acknowledgments

Financial support was granted from the Swedish Research Council (no. 421-2010-1597). The author is grateful for the assistance of Erik Bäckström on database issues and for the valuable comments of Guido Buenstorf and Bram Timmermans on an earlier draft of the article that was presented at the 2009 DRUID-DIME Academy Winter PhD Conference in Aalborg. The author also thanks the three anonymous reviewers whose comments substantially improved the article. The usual disclaimers apply.

Notes

1 For recent studies on knowledge spillovers overcoming the issue of spatial biases by using spatial econometrics, see, for example, RodrÍguez-Pose and Crescenzi (2008) and Sonn and Storper (2008).

2 For a more detailed discussion of the difference between routines and institutions, see MacKinnon et al. (2009) and Boschma and Frenken (2009).

3 Since the sampling procedure may influence the parameter estimates, the empirical models were also run on all plants in the original sample and on manufacturing plants only. Although the estimates on the entire population implied that the relative effect of inflows increased, it also resulted in a relative lower effect of the different types of inflows. When only manufacturing units were modeled, the outcome of the key variables was not affected. Taken together, these findings imply that the models show signs of being relatively robust.

4 To control for part-time work and increased efficiency, which would have been possible with information on hours of work, a proxy for part-time was created. It held information on the per capita social benefits received for all employees at each workplace (including parental leave, unemployment insurances, and sick leave), which implicitly account for the relative share of absence from work during 2001 (Eriksson and Lindgren 2009). This variable did not affect the estimates and was omitted from the final model.

5 The two groups with either observed or estimated productivity were estimated separately to check for the robustness of this indicator. The outcomes of the key variables did not differ substantially, which means that they can be interpreted with confidence.

6 It should be noted that the entropy within each two-digit category was also calculated, but this calculation did not change the effect of related variety in any of the models.

7 Because of the decomposable nature of the entropy measure, differentiating variety at various digit levels, this variable should not be interpreted as the inverse of the similarity variable (see Frenken 2007 for more details).

8 For example, except for the correlation between PopDensKm² and PopDensKm²∧2, the only correlation higher than 60 percent and significant at the 5 percent level was between the similar and unrelated variety variables calculated for the municipalities (correlation = 0.74). This finding was confirmed by testing the variance inflation factor in all the models, where only the two variables on population density show a tolerance below 0.2.

9 These effects were separated for two main reasons: first, there is collinearity between the measurements of similarity and unrelatedness, and, second, it is then possible to analyze explicitly how geographic proximity influences both intra- and interindustry spillovers. It should be noted that only models that estimated related or unrelated variety also were calculated because these variables could affect each other. Since neither the sign nor the levels of significance of the covariates were affected by this procedure, the results indicate robustness and can be interpreted with some confidence.