What are knowledge spillovers and how can you measure them?
knowledge may spill-over through non-market mediated actions (no price involved in the transaction) to benefit others – individuals and firms.
They produce:
- Increasing returns and growth (less costly knowledge)
- Free riding, wait for others to produce basic science instead of paying for the development.
- Inequalities among regions if spillovers have local nature.
It is difficult to measure.
What are the mechanisms through which knowledge spills-over?
Mechanisms from literature:
- Intra-firm/organization knowledge exchange. General consensus
- Geographical proximity. No general consensus
Alfred Marshall (1890) notes that firms that use similar knowledge tend to co-locate. Debate issue because it’s hard to empirically prove the existence and geographical extent of knowledge spillovers.
Marshall (1920): advantages of co-location activities, agglomeration economies given by:
- Pool labour market for specialized skills. (Demand and Supply)
- Pecuniary externalities, concerning the provision of specialised intermediate inputs and lower costs.
- -> Companies will form locally that provide specialized inputs to the already present larger firms. Benefits both specialized suppliers and the original firms from economies of scale. - Knowledge spillovers, as a non-market mediated (or involuntary) knowledge flows, particularly important for innovation and technological activities.
Talk about JTH 1993 study
JTH’s idea is to compare geographical location of inventors of the citing and cited patents.
(1) citations–> interpreted as a “knowledge spillover”
(2) Geographical address of inventors –> tracing the geography of the spillover
Research Hypothesis: if knowledge spillovers benefit from physical proximity, then citations must come disproportionally from nearby inventors (same country, region or state).
Challenge: if most firms active in a technology are located in one region only, then patents would be from that region anyway, how do you distinguish from citation that come from knowledge spillovers and simply co-located firms?
Specifically, JTH constructed three samples of patents:
- Sample of originating patents: patents belonging to Universities, top 10 corporate applicants and other companies. (in 1975 and 1980)
- Sample citing patents: patents citing the “originating patents”.
- Sample of control patents (key!): for each citing patent JTH find a “twin” patent that is as similar as possible to the citing patent but it DOES NOT cite the originating patent.
Method:
- Originating + Citing patents: Compute the geographic matching frequency (share of citation that come from same geographical area)
- Originating + Control patents: Compute the geographic matching frequency
Results:
1. Descriptive statistics
Newer data on citations is lower, makes sense = takes time to produce things that cite others
2.T est for localisation - Country level:
Geographic matching frequency is higher for the citing patents VS control patents. Confirms hypothesis
- Test for localisation - State Level
Confirm hypothesis
matching is twice as likely than control patents! - Test for localisation - City Level
Confirms hypothesis
matching is three-to four times more likely than controls!
Results/Evidence: geography matters for knowledge spillovers. Importance of geography becomes higher as the unit of analysis gets smaller.
Critiques:
a. Indicators: How good are patent citations to measure knowledge flows?… Citations may not be valid since:
Strategic intent in adding citations. Patent examiners add the citations.
b. Measurement: is the empirical evidence robust to the construction of different control samples? Paper: If you change the level of aggregation of technological classes, the results change.
c. Mechanisms: spillovers are not merely “in the air” and knowledge flows take place through market meditated mechanisms. Mobility of individuals, personal characteristics, social/personal networks
Talk about A&F 96 study
If they exist, they should be more relevant in R&D intensive industries. If this is the case, high-R&D industries should cluster even more than low-R&D industries.
Compare concentration indexes across sectors.
Challenge: similar as before, clustering could be due to production activities simply being more concentrated for other reasons (and R&D activities following production activities), other than knowledge spillovers.
Hypotheses: if R&D activities benefit from localized knowledge spillovers, innovation activities in R&D-intensive industries are more concentrated geographically compared to less-R&D intensive, after controlling for the spatial distribution of production activities.
They use the Gini of innovation = indicator of the degree of concentration of innovative activities at state level. They regress the gini on sets of controls.
Results:
Innovation concentrations higher in places where there are more sources of knowledge (skilled labour, university research). Even more so in R&D intensive sectors. Even after controlling for the geographical distribution of production activities.
Suggesting that production activities is not the force that is driving this correlations.
There might be other factors may affect the concentration of innovation activities (other than knowledge spillovers):
- transportation cost
- composition of the industry workforce
- proximity to natural resources
- proximity to the science base
- Firm characteristics, and in particular capital intensity (MES)
What are the final conclusions on the results of the two final milestone studies?
Evidence that geographical proximity facilitates the transmission of informal and unintentional knowledge spillovers.
- JTH (1993) the geographic matching frequency is higher for patent citation than for control patents, especially at the regional level.
- A&F (1996) the propensity for innovative activities to cluster is higher in R&D intensive industries, even after controlling for the geographical distribution of production activities.
