observations is huge, as in our case. In contrast to t-values, partial R2s enable us to compare
the relative importance of continuous variables with the relative importance of categorical ones,
such as industry or location.
Our econometric analysis is based on data from the Cost Structure Census of the German
Federal Statistical Office. This is a unique and representative micro-panel dataset containing
approximately 39,000 firms and covering 40 percent of all manufacturing firms in Germany
over the period from 1992 to 2005. We estimate efficiencies as firm-specific fixed effects, as
proposed by Schmidt and Sickles (1984). The major advantage of this approach, compared to
other stochastic frontier frameworks, is that it does not require any a priori assumption regard-
ing the distribution of efficiency across firms. Such distributional assumptions are often quite
restrictive and sometimes unsupported by the data.
The analysis yields some important results. (1) Industry affiliation is the most important
factor for explaining efficiency at the firm level, contributing almost half of the model’s ex-
planatory power for the level, and even more so for the development, of efficiency. (2) Firm
size and headquarter location contribute approximately 20 and 15 percent, respectively. (3)
Other factors such as R&D, organization of production, and relative size (production share in
domestic industry) have only negligible explanatory power, which is surprising given that these
factors have been emphasized as important in previous studies (e.g., Ornaghi, 2006). This pa-
per has mainly an explorative character; fundamental explanations of the influence mechanisms
behind the various factors lies beyond its scope. Nevertheless, we provide novel insights into
the importance of certain factors for explaining productive efficiency and its development.
The paper is structured as follows. Section 2 discusses hypotheses regarding the determi-
nants of efficiency, which are tested in the empirical analysis. Section 3 describes the method-
ology for assessing productive efficiency, gives specifics on the data used to estimate the pro-
duction function and efficiency scores, and discusses the obtained results. Section 4 reports the
analysis of the determinants of productive efficiency, sets out our reasons for using the partial
R2 concept, and describes the variables of the Cost Structure Census dataset used in the second
step of the analysis. Section 5 deals with the analysis of the dynamics of efficiency at the firm
level during the period 1992-2005. Section 6 provides a summary of empirical findings and
concluding remarks.
2 Productive efficiency of manufacturing firms
The classical microeconomic textbook treats all manufacturing firms as homogeneous produc-
ing units and, therefore, assume that all firms operate at the same level of efficiency. However,
empirical studies frequently show that in the real world some firms are more efficient than others
(e.g., Caves, 1989). Productive efficiency characterizes the firm’s ability to derive the maximum