example, with whom DuPont scientists are publishing and upon whose research DuPont scientists are capitalizing. A similar analysis can be done of patents. The result is a visual map of the breadth and depth of the intellectual network in which DuPont operates—the institutions with which they collaborate in research and patenting and how often and the institutions whose work they have used or absorbed and vice versa, again with what frequency. The analysis shows how interlinked the structure of scientific research is.
Hicks cautioned, however, that the effort to pull the data together was very intensive and expensive, and a considerable investment would be needed to extend the analysis to more companies and include human resources data. Hicks estimated that it might take $3 million to $4 million to clean and integrate up to 10 years of data from the various citation and human resources databases and another $25,000 a year to maintain it, excluding the cost of getting data on sources of research funding. Others said that estimate might be low, perhaps very low, especially for annual operating costs. On the other hand, Stephan observed that there is a substantial unmeasured cost in failing to improve the data already collected and their linkages, because they are revealing less and less about trends in industrial innovation.