We know that "cities are the engines of growth", that entrepreneurial activity is key, that human capital is crucial, that there are resulting spillover opportunities -- and that spatial concentration is the result. Statistical tests are difficult because the web of influences is complex. But this has not stopped researchers and a large amount of published research has recently become available.
Much of this is conveniently surveyed in Entrepreneurship, Geography, and American Economic Growth by Zoltan Acs and Catherine Armington.
The authors also tap into an amazing data set on 14 million firms across the 394 U.S. Labor Market Areas (Longitudinal Establishment and Enterprise Microdata). Their data include new firm formation which is their proxy for entrpreneurship. Adding the regional dimension is very helpful and provides opportunities for testing across a variety of conditions.
What is missing (not the authors' fault) is how this plays out at the spatial micro-level. Why have land markets? So that locational choices that exploit spillover opportunities can be realized. This is why cities are the engines of growth.
We are not yet there but this volume gets us a little closer.