#### pollinator behaviour

The hawkmoth *B. schenki* visited *S. longicauda* inflorescences only during the first 30 min after dusk. Hawkmoths typically started foraging from the lowermost flowers on inflorescences and always moved upwards on inflorescences, but on several occasions feeding commenced on the middle or upper flowers.

Overall, hawkmoths behaved similarly on natural and experimental inflorescences. Moths probed equivalent numbers of flowers on protandrous (mean ± SE = 4·1 ± 0·8), non-protandrous (5·7 ± 0·9) and natural inflorescences (5·9 ± 0·8; single-factor anova, *F*_{2,56} = 1·66, *P* > 0·05). In addition, their probing time (*s*) per flower did not differ significantly between protandrous (2·2 ± 0·2) and non-protandrous inflorescences (2·5 ± 0·2; *t*-test, *t* = −0·98, df = 25, *P* > 0·05).

#### pollen fates

Moths removed more pollinaria from protandrous inflorescences than from non-protandrous ones, but this difference was not significant (Fig. 3a). Flower position within the inflorescence affected pollinarium removal similarly for both protandrous and non-protandrous inflorescences; the proportion of flowers with at least one pollinium removed did not vary significantly with flower position (interaction between inflorescence type and position: quasi-*F*_{1,67} = 0·003, *P* > 0·05; Fig. 4a).

We found strongly contrasting patterns in the incidence of receipt of self- and cross-pollen in the two inflorescence types. Non-protandrous inflorescences received twice as many self-massulae as protandrous inflorescences (Fig. 3b). The median proportion of removed pollen deposited on self-stigmas in the case of non-protandrous inflorescences was 2·8 times greater than for protandrous inflorescences (Fig. 3c). However, because non-protandrous inflorescences also received more cross-pollen than protandrous inflorescences, the proportion of self-pollen in the total pollen load deposited on stigmas did not differ significantly between inflorescence types (Fig. 3e). Protandry decreased the incidence of geitonogamy. The minimum estimate of geitonogamy, based on the number of flowers that received self-pollen, but from which no pollinarium was removed, was significantly lower in protandrous inflorescences (median = 1, lower quartile = 1, upper quartile = 2, *N* = 29) than in non-protandrous ones (median = 2, lower quartile = 1, upper quartile = 3, *N* = 34; Mann–Whitney *U*-test: *Z* = 2·24, *P* < 0·05).

Self-pollination varied positively with pollen removal for both inflorescence types (protandrous: *R*^{2} = 0·22, *P* < 0·05; non-protandrous: *R*^{2} = 0·34, *P* < 0·005; Fig. 5a). The slopes of these relationships did not differ significantly (*F*_{1,46} = 0·34, *P* > 0·05). The proportion of flowers that received at least one self-massula differed significantly according to position in the two inflorescence types (interaction between inflorescence type and position: quasi *F*_{1,67} = 33·1, *P* < 0·0001; Fig. 4b), with self-pollination decreasing towards the top in protandrous inflorescences, but increasing towards the top in non-protandrous inflorescences.

The proportion of removed pollen that was exported to stigmas of other plants did not differ significantly between inflorescence types (Fig. 3d). However, the proportion of pollen exported to stigmas of other plants as a fraction of total pollen export to all stigmas was significantly higher for protandrous plants (Fig. 3f). Pollen export to other plants increased with pollen removal for both inflorescence types (protandrous: *R*^{2} = 0·49, *P* < 0·001; non-protandrous: *R*^{2} = 0·28, *P* < 0·01; Fig. 5b), but pollen export increased more strongly with pollen removal in protandrous plants (regression slopes: 2·04 *vs* 0·87; *F*_{1,46} = 5·23, *P* < 0·05).

In general, pollen export increased with self-pollination for both inflorescence types (protandrous: *R*^{2} = 0·25, *P* < 0·01; non-protandrous: *R*^{2} = 0·33, *P* < 0·005; Fig. 5c). The slopes of these relationships did not differ significantly (*F*_{1,46} = 0·02, *P* > 0·05) and were <1, indicating that pollen export declined proportionally with increasing self-pollination. The relationship between pollen export and self-pollination remained significant when pollen removal and treatment were included as predictor variables using multiple regression. The partial regression coefficient for pollen removal (*b* ± SE = 1·01 ± 0·29) did not differ significantly from 1 (*t*_{46} = 0·03, *P* > 0·05), indicating that pollen export used the same proportion of removed massulae, regardless of the number of massulae removed. The partial regression coefficient associated with self-pollination (*b* ± SE = 0·28 ± 0·12) was significantly <1 (*t*_{46} = −5·9, *P* < 0·0001), so the ratio of exported massulae to self-deposited massulae declined with increasing self-pollination. This indicates that self-pollination significantly reduced a plant's opportunities to export pollen.

Plants from which hawkmoths removed pollinaria exported pollen to as many as eight recipient plants, with a median of three recipients. The numbers of recipient plants did not differ significantly between the two inflorescence types (Mann–Whitney *U*-test: *Z* = 0·13, *P* > 0·05). Protandrous inflorescences dispersed pollen to more flowers (median = 10, lower quartile = 9, upper quartile = 15) than did non-protandrous inflorescences (median = 8, lower quartile = 5, upper quartile = 17), but this difference was not significant (Mann–Whitney *U*-test: *Z* = 2·20, *P* > 0·05). Most pollen was exported to nearby plants, with 61 and 64% of all dispersals occurring within a 30-cm radius around protandrous and non-protandrous inflorescences, respectively (Fig. 6). This localized pollen dispersal probably reflects the combined effects of hawkmoths moving short distances between plants and limited carry-over of pollen between successively visited flowers.