The IAB innovation database

In light of the weaknesses of patents as innovation indicators (Griliches, 1990), the use of literature-based innovation indicators has been postulated for quite some time (see Kleinknecht and Reijnen, 1993; Link, 1995). Nevertheless, only few are available to date.

Examples include the Science Policy Research Unit (SPRU) innovation database containing major innovations in the UK since 1945 (cited in Ratanawaraha and Polenske, 2007), the US Small Business Administration (SBA) survey for 1982 (Edwards and Gordon, 1984), or the database by Coombs, Narandren and Richards (1995) on British innovations. All of those, however, are either restricted to specific countries or periods. Recently, Metz (2003) has made a new database available that is a lot more promising than the previously mentioned alternatives. It contains innovation events throughout human history.⁷⁶ Each observation identifies an event that was mentioned in at least one of more than 1,300 references, primarily from the field of history of technology. Close to 15,000 such instances have entered the list (13,764 for the period 1500-1990). Assuming that citation of an event in the literature qualifies for being viewed as a significant contribution to technological advancement, the database may be argued to mirror the entire history of human technological progress. Details available for each event include - among other things - the name of the person or institutions who was accorded responsibility for the occurrence of the event, his/her place of birth or the location of the respective institution (place and country), year and location of the event (place and country), as well as the type and a description of the event. Table 4.1 provides an excerpt from the database. It lists all events reported for the year 1769. The most well-known are probably the grant of the patent for the steam engine and the construction of Cugnot’s steam car.

76 Originally, the database was assembled by the Institut für Arbeitsmarkt- und Berufsforschung (IAB) in Nürnberg. It is now stored at Zentralarchiv für empirische Sozialforschung, Cologne.

An important strength of the database is the broad notion of an innovation event. Note that usually an innovation is understood as a new product or specific method that is ready for market. Here, however, single incidences have been chosen to document the innovation process. In addition to actual inventions or discoveries, a whole range of other types of events have been considered as crucial occurrences in the innovation process, such as the first idea, an early draft, a publication or a successful trial run. That is, there may be two, three, or even more entries related to a specific innovation. The steam car, for instance, is represented in the database by three entries: the construction and the presentation of the car, as well as the first ride with it. In this case, the events occurred in the same year, but in other cases they are dispersed over many years. I maintain that the likelihood for manifold occurrences being documented with respect to just one innovation is higher the more attention an innovation received by society. This, in turn, is probably closely related to its significance in terms of economic growth. Hence, the database provides a promising alternative to traditional innovation measures, such as patents, which must be weighted in terms of their importance for economic growth.⁷⁷ The respective procedures, however, may entail arbitrariness.⁷⁸ Table 4.1 reveals that institutional innovations are also considered as a part of the list, if they are connected with technological development. Among others, those include laws (e.g. regulating the protection and organization of labor and intellectual property rights) and foundations (e.g.

academic societies, universities). Robustness checks of the empirical results should therefore test subsamples, which exclude those types of innovation events.

77 Actually, in some cases the exact same incidence is listed twice. Table 4.1 encloses two of those cases: the patent grant for the steam engine, and the installation of the first arrester in Germany. Probably being a coding error, it seems plausible to assume that such a mistake is more likely to occur, the more often an event is cited in the literature, hence the more important it is. So, the overall innovation frequencies series still reflect the technological advancement generated in every period. This view is supported by Metz and Watteler (2002).

78 In fact, some studies have used patent citations to identify important patents (see Hall, Jaffe, and Trajtenberg, 2000).

Table 4.1 - Documented innovation events in the year 1769

Notes: The original data is documented in German language. Hence, for the purpose of presenting an example, the information in this table have been translated into English. The translation, however, does not reflect the information from the original database with 100% accuracy. In some cases the terms may deviate from what would be the correct technical label, in other cases information was omitted from the description to keep it concise.

a The exact term could not be translated.

b Information completed as described by appendix B.1.

No. Person

event Type of event Description of event

5428 Arkwright, R. England^b Patent Spinning machine

5426 Beccaria Italy^b Succesful Trial Electrical charging of glass and other materials 12039 Beckmann, J. Hoya Germany Göttingen Germany^b Publication First agricultural schoolbook

5427 Bergman, T.O. Sweden Sweden^b Hypotheses Crystals result from chemical reaction in ocean water 5907 Born, I.E. von Karlsburg Prag Czech R. ^b Foundation .

5905 Reimarus, J.A.H. Hamburg Germany Installation First lightning arrester in Germany 14131 Reimarus, J.A.H. Hamburg Germany Installation First lightning arrester in Germany 5431 Scheele, K.W. Stralsund Sweden Sweden^b Discovery Tartaric acid

12276 Sivrac, Graf de Paris France Invention Celerifere/Velocifere: precursor to the bicycle.

4063 Smeaton, J. England^b Construction Early drilling machine

4208 Vevers, J. England Construction Mechanical carriage with treadles

4062 Watt, J. Scotland^b Patent Improved steam engine

12722 Watt, J. Scotland^b Delivery First steam engines delivered

14460 Watt, J. Scotland^b Patent Grant Steam engine

12831 Watt; Boulton, M. England^b Patent Self-regulating steam engine, first steam engine plant in Birmingham.

5909 Wise, S. England^b Patent Mechanical production device^a

3599 . England Law Law to prevent demoliton of machines and factory buildings

5430 . USA Foundation American Philosophical Society of Philadelphia

5906 . England Law Law to prevent any action against the introduction of machines

Unfortunately, the country is specified for only 34% of the cases in the original database. In order to extract series of innovation frequencies for a large set of countries, some effort had to be expended to recode this variable for as many observations as possible. Using other available details regarding the inventor or the city of an event, country information could be derived for roughly 85% of the events. Partly, internet sources were utilized to obtain this information. In fact, for most cases, it was possible to distinguish between the place of an event and the place where the respective inventor received his/her education. A detailed description of the steps taken to complete this information is provided in Appendix B.1.

Upon this effort, two types of innovation frequency series were derived from the data for each country that was mentioned at least once as the origin of an innovation event. The first is based on the place/s of an event, and the second on the place/s of birth (and allegedly education) of the person/s owning responsibility for the events. The latter would be the natural basis of an analysis, which is mainly concerned with the effect of human capital.

Provided a stronger interest in the effect of national institutions on innovation success, for instance, one might choose to base the analysis on the former series. For events associated with more than one country, be it because it occurred simultaneously - but independently - in different states, or because persons from different states collaborated on a project, each involved country was assigned the respective share of the event. That is, for joint events of two countries, each was assigned one half, for joint events of three countries one third, and so forth.⁷⁹

79 This approach discriminates against countries with many shared events in comparison to those with few shared events. Alternatively, a shared observation could be assigned fully to each of the involved countries.

Which approach is chosen should not affect the results of an empirical analysis as long as it takes account of unobserved individual effects.

Innovation and Growth on a Macro Level, 1500-1990 103

0 50 100 150 200 250

events/decade

1500 1600 1700 1800 1900 2000

UK US FR

Figure 4.1. Innovation frequencies by countries, 1400-1990⁸⁰

Source: Metz (2003).

Figure 4.1 provides an overview of innovation frequencies by countries between 1400 and 1990. The frequency for a country j is given by the number of documented events per decade, I&_j. Only the lines for the UK, the US, and France are highlighted, all other lines are displayed in light grey color. The main message from this chart is that technological progress has been generated by very few countries. Depending on the period, technological leadership changed (e.g. Italy around 1500, UK around 1800, and the US after 1850). But when picking

80 For the purpose of a neater presentation, not all the countries have been included in the chart. The light grey lines represent Australia, Austria, Belgium, Czech Republic, Switzerland, Denmark, Estonia, Spain, Ireland, Italy, Greece, Hungary, Netherlands, New Zealand, Poland, Portugal, Russia, and Sweden. Germany is excluded from the chart, because it is suspected that the database entails a bias in favor of German innovations given that only German literature was used by the assemblers of the database.

a specific period, essentially a few nations created most of the innovations and hence drove technological advancement while the rest of the world obviously adopted it.⁸¹ Further, the sudden take-off of innovative activity after 1750, as well as the quick reaction of France to the Industrial Revolution in Great Britain, seem hard to explain by the smooth and continuous development of population and human capital. Rather, shocks to the institutional environment, or the occurrence of some crucial technological advances are imposed as plausible explanations.

0 200 400 600 800

events/decade

1500 1600 1700 1800 1900 2000

Figure 4.2. Worldwide innovation frequency, 1400-1990

Source: Metz (2003).

81 It is suspected that the database entails a bias in favor of German innovations, because only German literature was used by the assemblers of the database. That is why Germany is excluded from this chart.

Innovation and Growth on a Macro Level, 1500-1990 105

Figure 4.2 depicts the development of aggregate worldwide innovation frequency, I&^*. The pattern implies a declining growth rate I&^* I^* . However, it does not necessarily imply a decline of actual knowledge growth. The observed pattern may be explainable by a declining probability of an event of being documented in written form and be remembered as an important technological contribution. Consider the following arguments: Two of the most frequently cited references date from the early 20^th century (see Metz, 1999). Obviously, those do not include events from the subsequent years, which could create a bias in favor of events that occurred prior to 1900. At the same time, however, possibilities of documentation certainly improved over time along with technological development, such that later publications can contain a larger share of the significant contributions to technological progress. Both effects might compensate each other. Anyhow, subsample estimates for years prior to 1900 are an easy way to control for a potential distortion. Further, one might argue that literature is less likely to reach a consensus about the significance of very recent events.

Hence, the documented number of incidences associated with an innovation (e.g. idea, first draft, trial runs etc.) may be smaller for more recent cases. On the other hand, recent documentation may be broader in the sense that a wider range of events appears in the literature, part of which vanishes after a while. Both these effects are not unlikely to cancel each other out, which would make them unproblematic. Finally, assume that, in reality, the stock of knowledge grows boundlessly at an exponential rate. In that case, the number of ideas generated in each period would soon exceed the documentation capacities, requiring that only the most significant events, i.e. the ones that receive most attention, be put on record.

Hence, the fraction of events that can actually be documented would have to decline steadily over time. Consequently, the documented events would gain in relative significance. This could explain the revealed pattern even in the presence of constant knowledge growth rates.

Because attributed significance or attention is likely to be correlated with the strength of the

productivity improvement effect, a database event may reflect greater shifts of the technological frontier in later years than in earlier years. On the other hand, one might argue that recording is not reduced over time at all, and innovations do in fact occur more rarely.

This view receives support from the observation that patent statistics, too, reveal a similar stagnating pattern, at least during the last decades (see Kortum, 1997). Also, other - much less comprehensive - long-run innovation counts (e.g. Baker, 1976) are well in line with the IAB data. Metz and Watteler (1999) provide graphical comparisons; nothing in their charts suggests that recent innovations could be unrepresented by the IAB data.

A possible explanation is that increasing complexity might make innovations more and more difficult, causing diminishing returns of existing knowledge in the generation of new ideas. Whichever interpretation is correct, the issue of potential changes in documentation behavior should be kept in mind, because it makes interpretation of the empirical results regarding the returns of existing knowledge less straightforward.