How Important Is IT?

Throughout the 20^th century, particularly after 1970, the technological revolution in semiconductors, digital communications, and consumer electronics resulted in dramatic changes to our everyday lives. We work, travel, consume, and communicate differently thanks to technological innovation. As a consequence, the role of information and communications technologies (ICTs) in the global economy is often the center of popular and academic attention—to the point that the word “technology” in common parlance has come to mean “information technology.” But just how influential have ICTs become in affecting subsequent inventions compared to other technologies? This article examines the special role of ICTs in influencing technological development over the 20^th century. The analysis is intended to highlight the role of ICTs in societal accumulation of technology, not only in terms of production and consumption.

Back to Top

ICTs can be utilized as components and enablers in many different production activities. For this reason, they are thus seen as drivers of economic growth. At the national-economy level, ICTs have a strong influence on economic growth,^8,14 and although there was some early debate as to whether ICT improves productivity,^a it is now accepted among economists that at the firm level, ICTs significantly contribute to industrial productivity.^4,7 Academic economists generally agree that such productivity benefits first appear in the ICT-producing industries and gradually spread to ICT-using industrial sectors.¹⁹

Economic history suggests that innovation over long economic and technological cycles is brought about by general-purpose technologies,⁶ or those that can be adopted in diverse economic sectors and lead to efficiency and improvements. For instance, electrical engines can power cars, factory machines, or home appliances. Electricity is thus a general-purpose technology that has been adopted by the car manufacturing industry, along with other machine industries, leading to improvements and efficiencies to the car as a product. However, the economywide contribution of such general-purpose technologies on subsequent invention has not been thoroughly examined thus far, with studies focusing instead on specific technologies (such as broadband).^2,3 There are also several studies that link adoption of ICTs with subsequent invention activities,^1,10 while others have considered the processes of “co-invention” associated with information technologies.⁵ Unlike these earlier studies, in this article we empirically assess whether ICTs have had a greater influence on cumulative technological change compared to other technological fields of the 20^th century.

We undertake our investigation using patent data, in particular the citations between individual patents. Patent data, with its global reach and wide technological spectrum, is an ideal testbed for describing how (protected) knowledge from ICT sectors has influenced other non-ICT sectors. Patent citations are references to “prior art” that the focal invention builds on or relates to and thus limit the originality of the focal invention.¹¹ They are widely used as a proxy for knowledge spillovers indicating the transfer of innovative applications and ideas within and across fields of technology.¹⁸ Consequently, citation patterns have been used extensively to inform industry and technology policy making; see, for instance, Dechezlepêtre et al.⁹ and Jaffe and Trajtenberg.¹² Citations to a patent have also been used as a measure of its importance in industry comparisons, intellectual property management, and valuation of firms.¹¹

We empirically compare knowledge spillover from ICT and non-ICT inventions at the patent-application level. Utilizing the PatStat database, which consists of the entirety of patent records for more than 160 patent offices over the past 100 years, we use two methods. We first analyze the number of prior art (forward) citations per patent, controlling for the type of the underlying invention (ICT or non-ICT) to assess the differences between the two groups. Second, we compute the importance of each patent based on the Google PageRank algorithm and substitute the number of prior-art citations with the PageRank metric; based on this model we rerun our analysis to assess the difference between ICT and non-ICT patents. For both approaches we apply a wide range of patent-specific, fixed, and time-varying controls.

We confirm that ICT patents are more influential than other types of patents, observing a significant difference in the citations of ICT and non-ICT technologies. ICT patents receive up to 0.406 more citations and a considerably higher PageRank than non-ICT patents. These findings quantify the influence of ICT inventions on other technological inventions. Moreover, the PageRank method provides a quality-adjusted indicator that helps measure the true influence of inventions. We suggest the exceptional influence of ICTs is due to their openness and flexibility enabling complementary invention and the fundamental roles of information and communication in the very process of invention.

Back to Top

Method

Our data source is PatStat, a comprehensive dataset available from the European Patent Office, with data from more than 160 publication authorities, 90 million awarded patents, and 160 million citations for the period 1900 to 2014. This information is also linked to detailed data about patent-level publication claims, patent families, technology fields, and classification data. Table 1 details the PatStat organization of technology sectors and fields. We use the patent universe (all listed patents from the PatStat dataset) to identify the complete network of prior-art citations. Table 1 also details our identification of ICT patents.^b As ICT is not solely the domain of the electrical-engineering sector, we code each technology class as ICT or non-ICT, following Kim and Hwang.^{13, c}

We estimate a simple model at the patent level where the total number of forward citations it has received (a count) is regressed against the type of patent (ICT or non-ICT) and other controls X_it. These controls include patent office, year of grant, patent family, extended patent family,^d stock of published patents by sector and year, count of citations by year and sector, and number of citations added by examiners. The controls permit us to alleviate sector-specific concerns (such as the potential ease in categorizing and finding during patent search), which lead to more complete referencing by patent officers, as well as rapid growth of the ICT sector itself, which leads to yet more patents and hence citations. Moreover, they enable us to indirectly control for regional and temporal effects regarding the “ease” of publishing a patent.^e Self-citations by the same assignee(s) are excluded from the counts. The model is

where C_i is the count of all citations received by patent i, ICT_i is a dummy equal to 1 for ICT patents and 0 otherwise, and X_it is a vector of patent characteristics.

To reduce the potential bias of other potentially confounding effects in patent citations we also look at the data as a directed graph that evolves over time. This approach resembles the PageRank algorithm initially proposed by Google as a measure of ranking webpage influence.¹⁶ Each patent represents a node in the graph, and each citation represents an edge that originates from the citing patent and ends at the cited patent. Ranking patents in terms of their influence instead of the count of incoming citations (in-degree) requires an iterative process that goes through each node of the graph and determines its rank as the sum of the ranks of other patents citing the focal patent. We compute the PageRank for each patent, then use the same model from the first regression, replacing the dependent variable with PageRank. As the PageRank method requires, we apply a damping factor to reduce the importance of nodes that are farther away.^f As a further robustness test we compute the PageRank for the entire network by decade and observe the changing influence of ICT patents.

As the size of the dataset exceeds common computing capacities, the bulk of this analysis has taken place using c4.8xlarge compute-optimized instances and r3.8xlarge memory-optimized instances on the Amazon Cloud.

Back to Top

Results

We first plot the mean citations for ICT and non-ICT sectors over the period of study in Figure 1. ICT patents clearly display higher citation counts for most of the 20^th century. We observe a peak in the 1980s and 1990s when the sector difference increased dramatically. Citation counts drop dramatically after 2005; we attribute it to both a smaller window of observations reducing the citation records and right censorship.^g We thus consider results later than 2005 to be unreliable.

We also conducted a simple t-test on the citation counts to see whether a difference exists between ICT and non-ICT patent citations (see Table 2). The t-test clearly demonstrated the substantial differences between the technology sectors; ICT patents receive 0.842 more citations than other patents. This result can be attributed to a number of factors: ICT fields may produce more patents than others, increasing their citation counts; ICT patents might cite themselves or other ICT patents more frequently, further increasing the counts; and ICT patents might be associated with more developed international presence through larger patent families that might boost citation counts as well.

To account for these confounding factors, we include a number of control variables in our models in Table 3. In model 1, considering the total number of citations received by each patent, we observe that ICT patents now receive only 0.406 more citations than non-ICT; that is, ICT patents receive 27.6% more citations than other technologies.^h The controls help explain almost half the effect found earlier in the simple t-statistic, which is both expected and reassuring about our method. In model 2 we limit the period of analysis to five years following publication, measuring the immediate impact of each invention rather than the longer-term effects. ICT patents attract 0.31 more citations than other technologies in their first five years, a finding that suggests that approximately three quarters of citation effects in ICT patents have already materialized in just five years following publication. As a further check, we find that the controls have the expected signs and significance, with the stock of patents positively affecting total counts and more popular sectors experiencing higher citation counts.

We now perform the computationally demanding iterative process of measuring the PageRank of each patent in our 90-million-patent dataset. This involves loading the edge list (approximately 160 million) of all citations and looping over citation networks until the algorithm converges under the specified accuracy thresholds. The PageRank offers a more informative metric compared to simple citation counts. Using the PageRank approach, two patents with identical citation counts are not necessarily equal; the importance of the citing patents instead reflects the importance of the focal invention.

Figure 2 shows the average PageRank by decade for selected technology sectors.ⁱ The effect we see with ICT patents is clearly reflected in the electrical-engineering sector where there was a distinct increase in importance from the 1970s onward. Moreover, these results seem to suggest the effect is ongoing. Emphasizing the continued effect of electrical engineering, mechanical engineering appears to be decreasing in importance, while chemistry and instruments appear to have remained at a constant (or slightly decreasing) level of importance during the period of study. The other-sectors category, including consumer goods and furniture and games, seem to be gradually increasing in importance, albeit from a more modest starting point.

Table 3 (model 3) lists the regression results of the PageRank analysis against all controls. We find ICT patents received 10% higher PageRanks than other technologies.^j Although this appears to be a smaller difference than that of simple citations (27.6%), we highlight that the distribution of citation counts and PageRanks differ significantly (see Table 4).

In particular, the distribution of PageRanks is less dispersed than the simple counts, and its coefficient of variation (the standard deviation divided by the mean) highlights this phenomenon, as in Table 4. Simple counts have a coefficient of variation equal to 4.66, while PageRanks had a coefficient of variation of 2.12, or less than half the dispersion of the citation distribution; that is, a 1% increase in PageRank is equivalent to a 2.2% increase in citation count.^k The 10% increase in PageRank ICT patents receive was thus equivalent to a 22% increase in citation counts, only slightly less than the 27.6% we found, confirming the effect.

Figure 3 and Figure 4 plot the kernel densities for the logged distributions of citations and PageRanks, with the y-axis indicating the probability of each value in the x-axis. For easier presentation and comparison, and, as these scores are positive and nonzero, we use the logarithmic form. Probability distributions help us further understand the difference between the ICT and non-ICT results by presenting the probability of a particular citation count or PageRank value. Figure 3 shows the ICT effects remain strong for both citations and PageRanks, with the gap between ICT and non-ICT widening as we move toward the most influential patents, or those with greater citation counts or PageRanks. Although the difference between PageRank and log(citations) as a measure of influence appears to be relatively minor at the very high end of the distribution, Figure 4, presenting the kernel density of the most influential 1% of patents, shows the difference between ICT & non-ICT is substantial. Figure 4 also shows there is greater probability of ICT patents with high citation counts and PageRank values than non-ICT patents.

Illustrating this effect, Table 5a includes the summary statistics comparison of the PageRank of the top 1%, 0.1%, and 0.01% for ICT and non-ICT patents, and Table 5b shows the PageRank comparison of the top 20 patents in our data. These examples help illustrate the greater PageRanks for ICT patents, as well as the types of patents that drive this effect.

We conclude that our analysis points to an outsized influence of ICT patents on subsequent patents from other technology sectors.

Back to Top

Conclusion

Using complete patent data available for more than 100 years from 160 countries comprising 90 million patents, we have shown that ICT patents are consistently more influential than patents from other industrial sectors. We used simple, iterative metrics to support our findings using the full lifetime of patents or shorter time spans. Whereas other studies have highlighted the importance of ICT for productivity and economic development, we have quantified the direct influence of ICT inventions on other technologies.

We emphasize the results obtained through the PageRank approach because it most accurately measures the influence of inventions. PageRank highlights the role of specific ICT inventions in enabling the invention of other influential technologies. With it, ICT patents do not appear quite as influential as with the coarser methods of descriptive statistics, but we still find a statistically and economically significant difference between ICT patents and the patents from other sectors. While the propensity for a greater number of patents and patent citation within the ICT sector may exaggerate the influence of ICT, we use PageRank to capture “cumulative” influence and still find ICT patents have an outstanding effect on subsequent technological development. Our findings thus complement and extend earlier studies of the societal effect of ICTs.

Our analysis is not directly able to explain conclusively why ICT inventions are substantially more influential than others but points to three possible factors we hope open avenues for further research: ICTs may be more cumulative than other technology fields, whereby subsequent inventions build more closely on previous inventions; as a field with a strong scientific knowledgebase, ICT inventions may depend on scientific breakthroughs that enable cumulative invention of industrial applications and lead to highly cumulative patterns of patent citations. ICTs may be more “generative” than other technologies; such generativity stems from the openness of ICT systems, by design, to enable complementary applications, and from the inherent flexibility of ICTs that creates technological opportunity for invention. And finally, ICTs may have exceptional influence on invention in a range of technological fields because they enable the capture, manipulation, and communication of information itself, and information is the fundamental ingredient of invention. In a related study, we call ICTs “invention machines” and further explore the nature of ICTs as general-purpose invention technologies.¹⁵

Back to Top

Back to Top

Back to Top

Back to Top

Back to Top

Back to top