INTERNATIONAL KNOWLEDGE FLOWS: CONCLUSIONS 3

Patent citations offer a rich repository of information about the locus of technological activity, and the relationships among activities in different places. Systematic use of these data requires, however, careful attention to the need to control for time and technology field effects that otherwise have an impact on simple comparisons across countries or other units of observation. Fortunately, the patent data are sufficiently numerous that detailed controls can, in fact, be implemented. Though any model obviously imposes structure on the data, one can allow for complex patterns of interactions among effects. Indeed, readers of this paper have no doubt already thought of additional interactions that we could have estimated with our data but did not. We hope that, as these data become more widely available, other researchers will pursue questions that we have not considered read.

APPENDIX A

DERIVATION OF EXPECTED CITATION FREQUENCY FOR A “CELL”

Let s index the patent classes represented by patents with the t, I, g attributes, and S index the patent classes represented in the set of patents with the T,L attributes. Let Ntg& represent the number of cited patents in a given class s, NTLS the number of citing patents in a given class S, and C„gsXLS be the total number of citations from class S in year T and country L to class s in year t, country £and field g. Starting from Eq. 1, the expected value of the citation count for a given “cell” is:
w6507-19