Fig. 7. Revisits to previously emptied feeders in comparison to revisit frequency expected under random choice of feeders. (A) Between-trials: revisits during successive trials of the experiment. The drop within the first three trials indicates the transition from a brief win-stay to a win-shift strategy. For calculations, see text. Revisits expected under random choice (line symbol) are a function of the number of feeders emptied during a trial; this number increases (from trials 1–3 to trials 11–20) as more feeders are emptied within a trial. Error bars are 95% confidence intervals of the difference (paired t-test) between measured revisits and revisits expected assuming random choice of feeders. Differences were significant for trials 1–3 (t=5.3, d.f.=14, P<0.001) and trials 11–20 (t=–2.8, d.f.=15, P=0.014) but not for trials 4–10 (t=0.5, d.f.=15, P>0.05). Data shown are means of the individual values for N=16 individuals (one individual with less than 20 visits during each of trials 1–3 was excluded from this specific analysis). (B) Within-trials: ratio of revisits to revisits expected under random choice within a trial. Data are for successive blocks of five visits within trials (based on trials 11–20, N=16 individuals, for calculation, see text). Data shown are the means of individual values (±S.E.M.). Frequency of revisits differed significantly from random expectation up to the 20–25th visit during a trial (Wilcoxon test, Z>2.1, N=16, P<0.05).