The old age population in developed countries has been increasing remarkably, yet internationally comparable high quality data on oldest-old mortality remain relatively scarce. The Kannisto-Thatcher Old Age Mortality Database (KTD) is a unique source providing uniformly recalculated old-age mortality data for 35 countries. Our study addresses a number of data quality issues relevant to population and death statistics at the most advanced ages. Following previous studies by Väinö Kannisto, we apply the same set of measures. This allows us to identify dubious or irregular mortality patterns. Deviations such as this often suggest that the data quality has serious problems. We update previously published findings by extending the analyses made so far to thirty five countries and by adding data on longer historical periods. In addition, we propose a systematic classification of country- and period-specific data, thus simultaneously accounting for each indicator of data quality. We apply conventional procedures of hierarchical cluster analysis to distinguish four data quality clusters (best data quality, acceptable data quality, conditionally acceptable quality, and weak quality). We show that the reliability of old-age mortality estimates has been improving in time. However, the mortality indicators for the most advanced ages of a number of countries, such as Chile, Canada, and the USA should be treated with caution even for the most recent decade. Canada, Ireland, Finland, Lithuania, New Zealand (Non-Maori), Norway, Portugal, Spain, and the USA have particular problems in their historical data series. After having compared the KTD with official data, we conclude that the methods used for extinct and almost extinct generations produce more accurate population estimates than those published by national statistical offices. The most reliable official data come from the countries with fully functioning population registers.
