Findings, implications and lessons learned

• The findings from the statistical analysis comparing the shapes of the lactation curves between the three periods essentially supported the conclusions reported by Romney et al., but our methods were somewhat sounder and provided more reasonable representations of the data.

• The statistical analysis of the differences in patterns of associations between milk yield and concentrate provided more information on the ways that patterns varied with stage of lactation, and gave some insight into the degrees to which farmers were adhering to the recommended feeding strategy.

• As a reader of a scientific publication it is imperative that one is able to review critically the contents of the paper and not accept at face value the authors' conclusions. It is particularly important that a reader is able to question the validity of any statistical methods that the authors may have applied.

• A strategy for tackling a research project is not cast in stone. The researcher needs to be ready for unexpected findings, unexpected delays, disappointments when studies do not work out as planned, and so on. When farmers decided not to follow precisely the recommendations given in this project, the researchers had to revisit their strategy for analysing the data.

• Furthermore, the realisation that additional data were required that had not been planned meant that additional information based on farmer recall (less reliable than from current observations) had to be collected by formal questionnaire at the time of the final cross-sectional survey.