Abstract:
In this study an historical statistic is defined as a number purporting to measure some aspect of reality which it is impracticable to remeasure, and the evaluation of an historical statistic is defined as determining its conceptual nature (i.e. true value defined) and its veracity (i.e. how closely it matches its true value). The difficulties associated with determining the conceptual nature of an historical statistic are discussed, and four approaches to determining veracity – the causal, consistency checking, contemporary comments, and revisions in statistics approaches – are outlined and scrutinized. A classification of non-sampling errors in statistics into coverage, response, and processing errors is presented, and the relationship between the collection and processing methods by which statistics are generated and the occurrence of errors of each of these types are discussed. Finally, the statistics generated by three important long-term New Zealand statistical collections: the annual sheep returns, 1879-1940, the annual agricultural statistics, 1868/69-1939/40, and the official trade statistics, 1853-1940, are evaluated as case studies.