A filter model to describe bias in official statistics on alcohol-related injuries

The paper describes a filter model that can be used to explore the possible biases that occur in reporting of alcohol-related injuries. These biases occur because of loss of data at various stages of injury reporting systems. The filter model consists of four levels or incidences, with a filter between each level. Level 1 is the total incidence of alcohol-related injuries. Level 2 is the incidence of reported alcohol-related injuries. Level 3 is the incidence of reported alcohol-related injuries where the contribution of alcohol has been correctly identified. Level 4 consists of official statistics on alcohol-related injuries. Between each of these levels is a filtering mechanism that allows some but not all data to pass through to the next level. The paper describes the nature of data at each level and the mechanisms operating at each filter to result in progressive loss of data. Estimates are provided regarding the extent of loss of data at each filter. Suggestions are provided for improving the quality of official statistics on alcohol-related injuries.