Imputation of missing information in worldwide patent data

How can we effectively impute missing information in the Worldwide Patent Statistical Database (PATSTAT)?

Accurate and comprehensive patent data are crucial for assessing technological activities globally. Many researchers rely on incomplete data, which can lead to erroneous analyses and policy decisions. By developing a robust imputation method for missing data in PATSTAT, we can significantly enhance the quality of insights drawn from this essential resource.

We developed a SQL algorithm to impute missing country codes and technology classifications in PATSTAT. We used comprehensive institutional knowledge of the international patent system and created two datasets—one for missing country codes of inventors and applicants, and another for missing technology classifications. Our method leverages information from both first and subsequent filings within patent families, improving data completeness for analysis.

Our imputation method substantially increased the completeness of the patent data, with significant performance improvements. For example, imputation raised the share of inventor country information from 1.1% to 49.8% for France in 1990. Moreover, IPC code imputation achieved 100% coverage across patent offices. The algorithm is versatile and can be adapted for other missing information in patent data.

• We generated two datasets: one containing information on missing country codes for inventors and applicants, and the other addressing missing technology classifications. Each dataset corresponds to a comprehensive output table derived from queries on the PATSTAT database.
• Across the two datasets, the total number of first filings included varies, encompassing numerous examples from different patent offices, specifically selected for analysis purposes.

• We assume that subsequent filings in a patent family can provide valid information to fill gaps in the first filing.
• We assume the patent system operates consistently across jurisdictions, enabling reliable data retrieval based on patent family relationships.
• We assume the imputation process can rely on sufficient institutional knowledge regarding patent classifications and applicable filing rules.