Soil moisture data were analysed with mixed effects models, using location, watering treatment and time as fixed effects, and plot pair (control and watered) as the random effect. The dependent variable was the mean value of the five soil moisture measurements per plot recorded each week. The data were logtransformed
to meet model assumptions. Temporal dependence of errors was modelled with a first order autoregressive correlation structure. The best model was selected using the AIC score, by stepwise removal of terms from the saturated model. In addition, edge watered and control plots and interior watered plots were compared with the interior control plots using major axis regression, to examine how soil moisture varied among
habitats.