In lieu of an abstract, here is a brief excerpt of the content:

a p p e n d i x Selected Primary Sources and Data Sets The contentions of this book rest on evidence from dozens of data sets and firm records , many used in several chapters. The specific kinds of evidence were discussed when first used, but an overview of the evidence and key assumptions might prove a useful reference. The appendix examines principal primary data sets on patents, inventions, patent assignments, output, and other activities as well as listing firm records consulted. For each it identifies the chapters in which the sources were used. Patents The book uses over 60 distinct sets of patent and invention data. Table A.1 lists these sets, their selection method, the chapters in which they were used, and the number of relevant patents, patentees, or individuals surveyed. One group, called the technology samples, is drawn from a survey of patents. A few of them are based upon U.S. Patent Office official classifications, including sewing machines, shoe machines, and telegraphs . The official classifications begin in 1836, when patents were first numbered, and in these three cases, all relevant patents were issued from 1836 forward. Most technology data sets selected patents based on keywords (such as spinning, weaving, looms, and carding for textiles, or locomotives, railroads, railroad cars, switches, and brakes for railroads), using an 1873 listing of all U.S. patent grants. This method omits many patents that are picked up in the classifications and so cannot claim to be complete . But it does allow the formation of a consistent set of patents from 1790 to 1865, not simply from 1836 on, and so can examine the growth of patents over time. It also catches many relevant patents not listed in the narrow classifications. A comparison of the keyword selection procedure with the use of patent classifications suggests that the procedure introduces no biases. (The sources for all data sets were listed the first time the sets were used in each chapter.) Other data sets are drawn from a survey of individuals, not patents. These sets determine all patents,usually from 1790 through 1865,for each inventor.The individuals are selected from (a) the technology samples, (b) a random sample of all inventors from 1790 through 1865, selected from lists of patentees (called the all-inventor sample ), (c) a study of major innovators listed in biographical dictionaries, (d) a study 330 Appendix of selected groups including principals of machinery firms, engineers, patents agents, patent examiners, and Franklin Institute leaders. The sources used posed particular problems for achieving a random sample of inventors in the all-inventor sample. The sources were alphabetical lists of patentees in each year from 1847 through 1865 and in a compilation of all patents from 1790 through 1846. A random sample of patents could not be used because it would bias the selection toward those with more than one patent in any year. My procedure was to pick randomly distributed locations in the reports (such as page 79, 8 centimeters from the top) and to pick the inventor after the one in that position. To have picked the one in that position would have biased the selection toward patentees with more patents (and hence taking up more space) in any report. The number of patentees chosen in each year was proportional to the number of patents, except that I oversampled in the years 1790–1846 to get enough patentees for meaningful comparisons over time. Although this procedure randomly selected inventors within any year, it did not overcome the problem that those patenting in more than one year had more chances of being selected. In a random sample each patentee from 1790 through 1865 would have an equal chance of being selected. My sampling procedure does so within each year, but because inventors could have been selected from 20 different sources—the cumulative index of patents through 1846 and annual reports from 1847 through 1865—patentees listed in more than one of these sources had more chances to be selected. I compensate by weighing the patents of inventors by the reciprocal of the number of years in which they invented. An inventor who patented in 10 different years (including the 1790–1846 index as one “year”), for example, had 10 times the likelihood of being selected as an inventor in only one year.By the procedure adopted, this inventor would be weighed at 0.1 if sampled in one year, 0.2 if sampled in two years, and so forth. An...

Share