The economics of entrepreneurship is notable for its careful use of econometric methods, and for eschewing the practice of asking entrepreneurs or other agents what they think they will do in various situations. Responses to these kinds of question are known to be prone to self-serving bias, or ‘cheap talk’, and respondents are liable to give answers which they think interviewers regard as ‘appropriate’. Instead, the ‘revealed preference’ principle trains economists to distrust individuals' declared intentions, forcing them to undertake the harder but more objective task of inferring preferences and constraints from their actual behaviour. This often necessitates the use of advanced statistical techniques to overcome econometric problems that might otherwise vitiate empirical estimates. Examples of such problems include:

• Sample selection bias, whereby sample membership is not random but instead is generated by some (at least partially) observable systematic process;

• Unobserved heterogeneity, whereby important unmeasured idiosyncratic variables are omitted from a regression model;

• Endogeneity, whereby an ‘independent’ variable is itself codetermined within the structural model of interest; and

• Non-stationarity, whereby time-series variables follow unit root processes that violate a key assumption of the classical linear regression model and lead to invalid statistical inference.

The present chapter outlines several ‘canonical’ empirical models in the economics of entrepreneurship which address all of the problems on this list. Throughout, knowledge of regression analysis is taken as given; readers not versed in it are referred to Greene (2003) or one of innumerable alternative standard textbooks.