This paper explains why the case study method need not suffer from the signal-to-noise problem which so badly afflicts the majority of econometric studies. Econometric methods suffer from this problem because they use indirect inference to estimate model parameters from data on all model variables. In a multivariate model, regression techniques can only estimate model parameters by using that part of the variance of each regressor which is independent of all other explanatory variables. When explanatory variables are closely related to each other, this partial variance is much smaller than the full variance of x, and hence we have a low signal. Moreover, it is not practical to include all possible explanatory variables in an econometric model, so the conventional noise variable can have a substantial variance. The case study does not suffer from this problem, because case studies do not just collect data on variables (as in econometrics) but also evidence on the nature and magnitude of relationships. In the case study, therefore, parameters are not estimated by indirect inference but by direct description. For that reason, the case study can use the full variance of each variable, and not just the partial variance. In a case study, moreover, it is practical to take account of a much larger number of possible explanatory variables than in an econometric model, and for that reason the noise is lower than would be the case in an econometric study. These observations do not, in themselves, negate some of the shortcomings of the case study, but the high signal-to-noise ratio means we can be confident that a comprehensive case study will identify the explanatory factors (or variables) that matter in that particular case – even if they relate only to a single case. |