Biases in approximating log production

Most empirical work in economic growth assumes either a Cobb-Douglas production function expressed in logs or a log-approximated constant elasticity of substitution specification. Estimates from each are likely biased due to logging the model and the latter can also suffer from approximation bias. We illustrate this with a successful replication of Masanjala and Papagerogiou (2004) and then estimate both models in levels to avoid these biases. Our estimation in levels gives results in line with conventional wisdom.