The quality of ideas: measuring innovation with multiple indicators

We model early expectations about the value and technological importance ('quality') of a patented innovation as a latent variable common to a set of four indicators: the number of patent claims, forward citations, backward citations and family size. The model is estimated for four technology areas using a sample of about 8000 U.S. patents applied for during 1960-91. We measure how much noise' each individual indicator contains and construct a more informative, composite measure of quality. The variance in quality', conditional on the four indicators, is just one-third of the unconditional variance. We show the variance reduction generated by subsets of indicators, and find forward citations to be particularly important. Our measure of quality is significantly related to subsequent decisions to renew a patent and to litigate infringements. Using patent and R&D data for 100 U.S. manufacturing firms, we find that adjusting for quality removes much of the apparent decline in research productivity (patent counts per R&D) observed at the aggregate level