The internet and other large textual databases contain billions of documents: is there useful information in the number of documents written about different topics? We propose, based on the premise that the occurrence of a phenomenon increases the likelihood that people write about it, that the relative frequency of documents discussing a phenomenon can be used to proxy for the corresponding occurrence-frequency. After establishing the conditions under which such proxying is likely to be successful, we construct proxies for a number of demographic variables in the US and for corruption across countries and US states and cities, obtaining average correlations with occurrence-frequencies of 0.47 and 0.61 respectively. We also replicate results from two separate published papers establishing the correlates of corruption at both the state and country level. Finally, we construct the first index of corruption in US cities and study its correlates.
Internet textual databases document-frequency proxy variables