This paper describes an empirical study of effort estimation in agile software development. Estimated effort and actual effort of a 46-iteration project are collected and analyzed. The results show that estimation in agile development is more accurate than that in traditional development even though agile developers still underestimate the effort. However, estimation accuracy is not improved over time as expected by agile communities.
Cao, Lan, "Estimating Agile Software Project Effort: An Empirical Study" (2008). AMCIS 2008 Proceedings. 401.