In economics, the number of observations available for empirical work is often predetermined. Researchers assume some large sample distribution and carry through with measurement and testing applied to data sets of varying sizes. The consequences of sampling variability are generally ignored. It is shown in a re-sampling experiment, using data sets of different sizes and estimating log-linear male labour supply equations, that a wide range of what appears to be statistically supported estimates of the wage elasticity of labour supply are generated. Testing based on bootstrapped estimates shows that 4000 observations are required to reduce sampling variability to statistically acceptable levels.