| This study is chiefly concerned with the impact of commonly employed variations in the specification of econometric models on the outcome of those models. Three previously published simple time series models of the determinants of coal mine fatality and injury rates are replicated. By varying such factors as the definition of the dependent variables, the manner of operationalization of the concepts which serve as the independent variables, the time period modeled, and the presence of trend terms, nearly 3,000 unique model specifications are developed and comparatively evaluated. The results suggest the following: (1) scholars have traditionally paid too little attention to the impact of alternate measures of concepts on model outcome; (2) the published literature fails to deal adequately with the deleterious effects of collinearity on model outcomes; (3) studies purporting to test micro-level theories often commit ecological fallacies by analyzing macro-level data; (4) the simple time series econometric technique may have much more limited utility than has generally been assumed due to its tendency to be massively unstable contingent on even minor changes in model specification; (5) sensitivity analysis is an important and under-utilized tool which has great potential to improve the quality of social science research.;Included are reviews of the background and context of coal mine safety regulation, a review of the literature from related fields of study, and an analysis of the theoretical issues of interest. Particular emphasis is put upon contrasting the predictions of economic theory and a behavioral approach, although no clear test of the two approaches is included. There is also an extensive review of the methodological issues which are relevant to least squares regression analysis. |
