Analysis of cluster randomised stepped wedge trials with repeated cross-sectional samples