Measuring Research Intensity from Anonymized Data: Does Multiplicative Noise with Factor Structure Save Results Regarding Quotients?

Summary Economic researchers often consider quotients like R&D investment divided by sales which could be used to measure "research intensity" of firms if available. However, data on research in particular are highly confidential and would not be released in original form. Therefore scientific use files have to be generated from anonymized micro data. The paper considers joint anonymization of all variables by multiplicative noise which stems from a bimodal mixture distribution and can be regarded as an error model with factor structure. It is shown that quotients such as research intensity are not modified considerably by this procedure. However, already quotients from original data can give quite misleading results which is illustrated by simulation results and an empirical example using the German Cost Structure Survey.