In this paper it is tested which of the various alternative approaches for constructing knowledge spillover pools suggested in existing literature measures the extent to which a rm can costlessly receive external knowledge best. Since knowledge spillovers are unmeasurable, a 'goodness of fit' measure is constructed using innovation survey data. It turns out that measures of the uncentered correlation of firm characteristics seem to fit actual knowledge spillovers best. Direct measures constructed from innovation survey data appear to work reasonably well while measures of the Euclidean technological distance and of the geographical distance lead to counterintuitive results. Empirical evidence is provided for both the German service sector and the manufacturing sector.


Kaiser, Ulrich


knowledge spillovers, technological distance, geographical distance, CIS-II data, Kernel density estimation, ordered probit estimation