- Email: [email protected]
Clarification of the chick reaction hypothesis
Anim. Behav., 1998, 55, 504–507
Clarification of the chick reaction hypothesis SONIA KLEINDORFER*, HERBERT HOI† & BIRGIT FESSL† *University of Washington, Seattle †Konrad Lorenz Institut fu¨r Vergleichende Verhaltensforschung (KLIVV) (Received 20 February 1997; initial acceptance 23 April 1997; final acceptance 19 September 1997; MS. number: -1100)
Halupka & Halupka (1998) cite two major errors in Kleindorfer et al. (1996): sampling error and the reasoning behind the chick reaction hypothesis. Sampling Error Although Halupka & Halupka (1998) use the term sampling error, it seems that they are actually addressing analysis error. By definition, sampling error refers to any bias in sampling that influences the reliability, validity or objectivity of a sample (see Lienert 1986), but there is no bias in data gathering that we could establish. We assume that Halupka & Halupka are referring to sampling error resulting from repeated trials at each nest, which could lead to habituation or increased risk taking given prior predator experiences. We based our methods on the results of a methodological review by Weatherhead (1989), who determined that the method of repeated observer approaches is appropriate for studying nest defence. Clearly, this topic deserves continued and detailed scrutiny given the widespread use of the observer approach method. Our data are insufficient to provide a rigorous analysis of this potential source of bias. Concerning Halupka & Halupka’s criticism of Fig. 1 in Kleindorfer et al. (1996), which shows the frequency of alarm calls for all nests, we agree that it is generally more prudent to control for potentially confounding factors such as nest, time of year and parental condition if the aim is to explore statistically which factors influence the frequency of alarm calls. However, we never offered any statistics with these results (the statistics were in relation to averages per nest). Our Correspondence: H. Hoi, KLIVV, Savoyenstrasse 1a, A-1160, Vienna, Austria (email: [email protected]. ac.at). S. Kleindorfer is at the University of Washington School of Medicine, Seattle, WA 98195, U.S.A. 0003–3472/98/020504+04 $25.00/0/ar970609
aim in Fig. 1 was to present the general pattern of alarm calling, which is similar for many passerines (reviewed in Montgomerie & Weatherhead 1988). The main aim of the paper was to investigate the day of onset of sequential alarm calling, calculated as a percentage for all nests, not the absolute frequency of calls. In fact, we never explored the frequency of alarm calling in the experimental data analysis. The emphasis of the paper shifted from the general pattern observed (Fig. 1) to a more detailed analysis of the time-dependent probability of both parental alarm calling and chick reactions during controlled predator placements near the nest. The time-frame for analysis also shifted: whereas Fig. 1 showed the results for 1-h behavioural observations, the remainder of the paper used a 2-min time-frame for an experimental set-up. As a general note, any discussion of intensity needs a specific reference to the timeframe used in the analysis (i.e. number of calls per time unit). In our case, this consideration prohibits a direct comparison between frequencies per h (Fig. 1) and frequencies per min. Most importantly, however, no data were presented on the frequency of alarm calling per min. This is an interesting point for future investigation, namely the relation between calling probability and calling intensity. We believe that intensity reflects, for example, brood value and perceived risk, whereas the probability of calling may be more directly related to cues that elicit calling (e.g. chick behaviour, predator behaviour). Halupka & Halupka (1998) did detect a serious problem in our data analysis which we are aware of, and did explicitly describe. As stated in the methods and quoted by Halupka & Halupka (1998), we used three nests where parent and chick behaviour had changed during the observation period from ‘no parent calls/no chicks jump’ to ‘parent calls/chicks jump’ (in all other cases the categories did not change within a trial). For the
? 1998 The Association for the Study of Animal Behaviour
504
Commentaries three ‘mixed’ cases, these are of course paired data in a traditional sense; they are actually the most convincing cases, however, because the chicks jumped once the parents began to alarm-call. We were faced with (1) excluding them (which would not have changed the results, P<0.05), (2) testing them separately (the sample size was too small), (3) including them and using a paired test (impossible to do), or (4) including them and using an unpaired test (which we did and with which the referees agreed). Concerning the experimental data, we considered each predator event to be independent (i.e. one should never become habituated to a snake, nor take very high risks because the snake did not attack on a previous occasion), and this point is certainly worth discussing. This was an assumption we should have stated more clearly: we consider it adaptive, based on non-additive properties of predator events (except for cases when only one egg is taken by a predator, for example). It remains to be shown how experience of a predator will or should influence anti-predator behaviour. To control for pseudoreplication in the experimental set-up, we entered every nest at about the same level. We calculated the probabilities of alarm calling or of a reaction on a daily basis. Confounding variables were unlikely to influence the experiment because all experiments were done only in 1994 (Kleindorfer et al. 1996) and only in April and May (first clutches). To correct for developmental differences because of season or parental quality, the data were depicted with day 0 as the day of fledging and day "14 as the day of hatching (see Kleindorfer et al. 1996). Reasoning behind the Chick Reaction Hypothesis Concerning our reasoning behind the chick reaction hypothesis, we feel more strongly that Halupka & Halupka (1998) have misinterpreted the findings. First and foremost, we do not propose that the chick reaction hypothesis explains either the intensity, or the general occurrence, of alarm calls (for example, alarm calling occurs outside of the breeding phase!). We obviously did not clarify this point sufficiently in the formulation of the predictions or in the discussion. Second, the chick reaction hypothesis applies mainly to species with open-cupped nests. Specifically, it focuses on the starting point for sequential days with probabilities of alarm calling and chick
505
anti-predator responses. The point we wished to make is that in communication events, such as alarm calling, attention needs to be focused on the caller and intended receiver of calls. While alarm calls may serve to distract predators, why does the probability of adults giving alarm calls increase once chicks show anti-predator responses? We suggest that part of the answer lies in the adaptive explanation of increased predator avoidance behaviour by nestlings (ducking in the nest and jumping from the nest, based on chick age and predator type). References to changes in the rate and intensity of calls made in the introduction of Kleindorfer et al. (1996) are confusing: a distinction needs to be made for different time spans under investigation (for example 1 h versus 2 min). It is quite possible that the intensity of calls per 2 min does not vary, whereas the intensity per h does (as the result of changes in calling bout frequencies but not calls per bout). This has implications for the study of calling intensity when comparing different reproductive phases and functions of calls. Furthermore, as stated above, the relation between calling probability and calling intensity needs to be explored. Our paper did not address this issue. We focused on the probability of calling per day of the nesting phase (incubation and feeding) using the average calling probability for all nests and addressed one specific function of alarm calls. We furthermore agree that alarm calling is likely to have several functions. Halupka & Halupka (1998) argue that the relationship between intensity of alarm calls and age of the offspring suggests other functions, such as distracting intruders or alerting conspecifics. We agree. However, we never investigated the intensity of calls. All our data and graphs using the experimental data depict only the probability of calling calculated across all nests (no=0, yes=1). Halupka & Halupka (1998) further argue that the occurrence of alarm calls during incubation contradicts the chick reaction hypothesis. Since this hypothesis addresses the sequential probability of parental alarm calling as it correlates with chick anti-predator responses, it obviously cannot include the incubation phase. The correlation refers to the probability of calling and the probability of reacting; it does not refer to intensity. We observed alarm calls during incubation, but they were given infrequently, occurred unpredictably, and did not occur on sequential
Animal Behaviour, 55, 2
days with a high probability across nests (for example: reaction intensity to a snake was high at one nest and non-existent even after detection at another; stated in terms of probability, the number of nests reacting did not increase in a predictable way across the incubation phase). The main criticism of our predictions was that alarm-calling intensity should decrease (although we again did not refer to intensity) rather than increase with chick age because of the development of the chick’s neural system. Since we did not investigate intensity, one would need a detailed study of calls per bout to address this criticism. We are not convinced that this is the only prediction, however, since it could pay parents to alarm-call more once chicks have welldeveloped neural systems even when the risk is low (for example, predator is far away, a less dangerous predator, etc.). Halupka & Halupka (1998) argue that older offspring are more likely to detect an approaching predator without signals from their parents. This implies that chicks would have full knowledge about predators and what to do when a certain predator approaches. This raises several questions, such as how do chicks distinguish between potential predators, and should they always jump from the nest (which carries a high cost and risk)? How should chicks behave when a harmless bird species appears? To escape from the nest would be lethal if a marsh harrier were nearby, to remain would be lethal if a snake were attacking. Given all of these situations, we do think that alarm calls contain important relevant information (for example on the type and distance of the predator, and instructions on how to behave). Moustached warblers have different types of alarm calls (Leisler 1991), and the intensity of calls increases with risk such as the distance of the predator to the nest (unpublished data). The type and intensity of a call could thus provide chicks with information on what to do. The results in Kleindorfer et al. (1996) also indicated that, after parental calling, chick reaction is predator specific. Perhaps because the neural system does develop with time, communication (teaching) also becomes relatively more important. Finally, our own data on moustached warblers suggest that the intensity of alarm calls does not change with age. Comparing the intensity of alarm calls of early and late feeding moustached warblers to an experimentally placed plastic snake
12
Alarm calls/min
506
8
4
0
Early feeding
Late feeding
Figure 1. Alarm call intensity is given as the mean (+) number of alarm calls/min during our experiments (for a description see Kleindorfer et al. 1996) for early chick feeding (chicks 2–4 days old, N=6) and late chick feeding (chicks 11–13 days old, N=13). Every nest was entered only once. For comparative reasons we selected only experiments with a ground predator (snake). Included are only nests where the snake elicited alarm calls. Male and female data are pooled and only intensity within 10 m of the nest was considered.
revealed no difference (Mann–Whitney U-test: Z= "0.48, P>0.6, see Fig. 1). This suggests that the intensity of alarm calls is independent of chick age, at least to certain predator types and for a certain time span. When alarm calls were given, intensity was quite high which we would expect if, for instance, predator distraction or alerting conspecifics is important. This is also supported by the fact that the intensity of alarm calls is more a function of distance of the predator to the nest (unpublished data), or generally the perceived risk. As chick anti-predator behaviour improves, we would expect the number of bouts of calling to the same level of potential risk to increase, while the intensity of alarm calls may vary less. Our data suggest that what changes across the nesting period is the probability of reacting to a predator; once a bird shows a reaction (to the same level of risk) there is less difference in the intensity of the reaction. Halupka & Halupka (1998) further argues that a significant correlation between alarm calling and behaviour of chicks (see Fig. 3 in Kleindorfer et al. 1996) need not necessarily reflect a causal
Commentaries relationship. Only after chick reactions were observed (irrespective of the intensity of reaction) did we find a pattern of increasing probabilities of parental alarm calling (irrespective of the intensity of calls). We agree that this is only a correlation and does not demonstrate any causality. However, the data discussed above on chicks jumping from the nest for two situations, namely with or without parental alarm calls, at least suggest that alarm calls do influence chick behaviour. Furthermore, the study showed that the proportion of chicks jumping from the nest tended to be higher for ground predators than for aerial predators, which we tentatively interpret as constituting an adaptive response. In playback experiments, we have already elicited different chick reactions, including jumping out of the nest, for bearded tits, Panurus biarmicus, and moustached warblers (H. Hoi, unpublished data). We agree with Halupka & Halupka (1998) that further experimental evidence is needed to understand more fully the social and environmental context of alarm calling. Our paper explicitly concluded with five areas of research in this regard. We strongly recommend disentangling alarm calls from the nest defence package to assess their relative contribution for predator avoidance strategies. One basic issue which Kleindorfer et al. (1996) addressed is the need to differentiate between the onset of sequential days of calling versus the intensity of calls since the underlying explanations for these may differ.
507
Halupka & Halupka (1998) have clearly shown the need to determine the relationship between the probability and the intensity of alarm calling. Another issue this study raises is the need to separate the levels of explanation for the occurrence of alarm calls, which may ultimately be related to the brood value hypothesis. Since anti-predator behaviour by chicks increases their brood value (measured in terms of survival probability), the question remains: do parents attend to rule-of-thumb measures of reproductive value or to more proximate cues such as chick behaviour?
REFERENCES Halupka, K. & Halupka, L. 1998. Alarm calls and chick reaction: comments on paper by Kleindorfer et al. Anim. Behav., 55, 502–503. Kleindorfer, S., Hoi, H. & Fessl, B. 1996. Alarm calls and chick reactions in the moustached warbler, Acrocephalus melanopogon. Anim. Behav., 51, 1199–1206. Leisler, B. 1991. Acrocephalus melanopogon. In: Handbuch der Vo¨gel Mitteleuropas (Ed. by K. M. Glutz von Blotzheim & K. M. Bauer), pp. 217–251. Wiesbaden: Aula Verlag. Lienert, G. A. 1986. Verteilungsfreie Methoden in der Biostatistik. Vol. 1. Meisenheim: Anton Hain. Montgomerie, R. D. & Weatherhead, P. J. 1988. Risks and rewards for nest defence by parent birds. Q. Rev. Biol., 63, 167–187. Weatherhead, P. J. 1989. Nest defence by song sparrows: methodological and life history considerations. Behav. Ecol. Sociobiol., 25, 129–136.
Clarification of the chick reaction hypothesis SONIA KLEINDORFER*, HERBERT HOI† & BIRGIT FESSL† *University of Washington, Seattle †Konrad Lorenz Institut fu¨r Vergleichende Verhaltensforschung (KLIVV) (Received 20 February 1997; initial acceptance 23 April 1997; final acceptance 19 September 1997; MS. number: -1100)
Halupka & Halupka (1998) cite two major errors in Kleindorfer et al. (1996): sampling error and the reasoning behind the chick reaction hypothesis. Sampling Error Although Halupka & Halupka (1998) use the term sampling error, it seems that they are actually addressing analysis error. By definition, sampling error refers to any bias in sampling that influences the reliability, validity or objectivity of a sample (see Lienert 1986), but there is no bias in data gathering that we could establish. We assume that Halupka & Halupka are referring to sampling error resulting from repeated trials at each nest, which could lead to habituation or increased risk taking given prior predator experiences. We based our methods on the results of a methodological review by Weatherhead (1989), who determined that the method of repeated observer approaches is appropriate for studying nest defence. Clearly, this topic deserves continued and detailed scrutiny given the widespread use of the observer approach method. Our data are insufficient to provide a rigorous analysis of this potential source of bias. Concerning Halupka & Halupka’s criticism of Fig. 1 in Kleindorfer et al. (1996), which shows the frequency of alarm calls for all nests, we agree that it is generally more prudent to control for potentially confounding factors such as nest, time of year and parental condition if the aim is to explore statistically which factors influence the frequency of alarm calls. However, we never offered any statistics with these results (the statistics were in relation to averages per nest). Our Correspondence: H. Hoi, KLIVV, Savoyenstrasse 1a, A-1160, Vienna, Austria (email: [email protected]. ac.at). S. Kleindorfer is at the University of Washington School of Medicine, Seattle, WA 98195, U.S.A. 0003–3472/98/020504+04 $25.00/0/ar970609
aim in Fig. 1 was to present the general pattern of alarm calling, which is similar for many passerines (reviewed in Montgomerie & Weatherhead 1988). The main aim of the paper was to investigate the day of onset of sequential alarm calling, calculated as a percentage for all nests, not the absolute frequency of calls. In fact, we never explored the frequency of alarm calling in the experimental data analysis. The emphasis of the paper shifted from the general pattern observed (Fig. 1) to a more detailed analysis of the time-dependent probability of both parental alarm calling and chick reactions during controlled predator placements near the nest. The time-frame for analysis also shifted: whereas Fig. 1 showed the results for 1-h behavioural observations, the remainder of the paper used a 2-min time-frame for an experimental set-up. As a general note, any discussion of intensity needs a specific reference to the timeframe used in the analysis (i.e. number of calls per time unit). In our case, this consideration prohibits a direct comparison between frequencies per h (Fig. 1) and frequencies per min. Most importantly, however, no data were presented on the frequency of alarm calling per min. This is an interesting point for future investigation, namely the relation between calling probability and calling intensity. We believe that intensity reflects, for example, brood value and perceived risk, whereas the probability of calling may be more directly related to cues that elicit calling (e.g. chick behaviour, predator behaviour). Halupka & Halupka (1998) did detect a serious problem in our data analysis which we are aware of, and did explicitly describe. As stated in the methods and quoted by Halupka & Halupka (1998), we used three nests where parent and chick behaviour had changed during the observation period from ‘no parent calls/no chicks jump’ to ‘parent calls/chicks jump’ (in all other cases the categories did not change within a trial). For the
? 1998 The Association for the Study of Animal Behaviour
504
Commentaries three ‘mixed’ cases, these are of course paired data in a traditional sense; they are actually the most convincing cases, however, because the chicks jumped once the parents began to alarm-call. We were faced with (1) excluding them (which would not have changed the results, P<0.05), (2) testing them separately (the sample size was too small), (3) including them and using a paired test (impossible to do), or (4) including them and using an unpaired test (which we did and with which the referees agreed). Concerning the experimental data, we considered each predator event to be independent (i.e. one should never become habituated to a snake, nor take very high risks because the snake did not attack on a previous occasion), and this point is certainly worth discussing. This was an assumption we should have stated more clearly: we consider it adaptive, based on non-additive properties of predator events (except for cases when only one egg is taken by a predator, for example). It remains to be shown how experience of a predator will or should influence anti-predator behaviour. To control for pseudoreplication in the experimental set-up, we entered every nest at about the same level. We calculated the probabilities of alarm calling or of a reaction on a daily basis. Confounding variables were unlikely to influence the experiment because all experiments were done only in 1994 (Kleindorfer et al. 1996) and only in April and May (first clutches). To correct for developmental differences because of season or parental quality, the data were depicted with day 0 as the day of fledging and day "14 as the day of hatching (see Kleindorfer et al. 1996). Reasoning behind the Chick Reaction Hypothesis Concerning our reasoning behind the chick reaction hypothesis, we feel more strongly that Halupka & Halupka (1998) have misinterpreted the findings. First and foremost, we do not propose that the chick reaction hypothesis explains either the intensity, or the general occurrence, of alarm calls (for example, alarm calling occurs outside of the breeding phase!). We obviously did not clarify this point sufficiently in the formulation of the predictions or in the discussion. Second, the chick reaction hypothesis applies mainly to species with open-cupped nests. Specifically, it focuses on the starting point for sequential days with probabilities of alarm calling and chick
505
anti-predator responses. The point we wished to make is that in communication events, such as alarm calling, attention needs to be focused on the caller and intended receiver of calls. While alarm calls may serve to distract predators, why does the probability of adults giving alarm calls increase once chicks show anti-predator responses? We suggest that part of the answer lies in the adaptive explanation of increased predator avoidance behaviour by nestlings (ducking in the nest and jumping from the nest, based on chick age and predator type). References to changes in the rate and intensity of calls made in the introduction of Kleindorfer et al. (1996) are confusing: a distinction needs to be made for different time spans under investigation (for example 1 h versus 2 min). It is quite possible that the intensity of calls per 2 min does not vary, whereas the intensity per h does (as the result of changes in calling bout frequencies but not calls per bout). This has implications for the study of calling intensity when comparing different reproductive phases and functions of calls. Furthermore, as stated above, the relation between calling probability and calling intensity needs to be explored. Our paper did not address this issue. We focused on the probability of calling per day of the nesting phase (incubation and feeding) using the average calling probability for all nests and addressed one specific function of alarm calls. We furthermore agree that alarm calling is likely to have several functions. Halupka & Halupka (1998) argue that the relationship between intensity of alarm calls and age of the offspring suggests other functions, such as distracting intruders or alerting conspecifics. We agree. However, we never investigated the intensity of calls. All our data and graphs using the experimental data depict only the probability of calling calculated across all nests (no=0, yes=1). Halupka & Halupka (1998) further argue that the occurrence of alarm calls during incubation contradicts the chick reaction hypothesis. Since this hypothesis addresses the sequential probability of parental alarm calling as it correlates with chick anti-predator responses, it obviously cannot include the incubation phase. The correlation refers to the probability of calling and the probability of reacting; it does not refer to intensity. We observed alarm calls during incubation, but they were given infrequently, occurred unpredictably, and did not occur on sequential
Animal Behaviour, 55, 2
days with a high probability across nests (for example: reaction intensity to a snake was high at one nest and non-existent even after detection at another; stated in terms of probability, the number of nests reacting did not increase in a predictable way across the incubation phase). The main criticism of our predictions was that alarm-calling intensity should decrease (although we again did not refer to intensity) rather than increase with chick age because of the development of the chick’s neural system. Since we did not investigate intensity, one would need a detailed study of calls per bout to address this criticism. We are not convinced that this is the only prediction, however, since it could pay parents to alarm-call more once chicks have welldeveloped neural systems even when the risk is low (for example, predator is far away, a less dangerous predator, etc.). Halupka & Halupka (1998) argue that older offspring are more likely to detect an approaching predator without signals from their parents. This implies that chicks would have full knowledge about predators and what to do when a certain predator approaches. This raises several questions, such as how do chicks distinguish between potential predators, and should they always jump from the nest (which carries a high cost and risk)? How should chicks behave when a harmless bird species appears? To escape from the nest would be lethal if a marsh harrier were nearby, to remain would be lethal if a snake were attacking. Given all of these situations, we do think that alarm calls contain important relevant information (for example on the type and distance of the predator, and instructions on how to behave). Moustached warblers have different types of alarm calls (Leisler 1991), and the intensity of calls increases with risk such as the distance of the predator to the nest (unpublished data). The type and intensity of a call could thus provide chicks with information on what to do. The results in Kleindorfer et al. (1996) also indicated that, after parental calling, chick reaction is predator specific. Perhaps because the neural system does develop with time, communication (teaching) also becomes relatively more important. Finally, our own data on moustached warblers suggest that the intensity of alarm calls does not change with age. Comparing the intensity of alarm calls of early and late feeding moustached warblers to an experimentally placed plastic snake
12
Alarm calls/min
506
8
4
0
Early feeding
Late feeding
Figure 1. Alarm call intensity is given as the mean (+) number of alarm calls/min during our experiments (for a description see Kleindorfer et al. 1996) for early chick feeding (chicks 2–4 days old, N=6) and late chick feeding (chicks 11–13 days old, N=13). Every nest was entered only once. For comparative reasons we selected only experiments with a ground predator (snake). Included are only nests where the snake elicited alarm calls. Male and female data are pooled and only intensity within 10 m of the nest was considered.
revealed no difference (Mann–Whitney U-test: Z= "0.48, P>0.6, see Fig. 1). This suggests that the intensity of alarm calls is independent of chick age, at least to certain predator types and for a certain time span. When alarm calls were given, intensity was quite high which we would expect if, for instance, predator distraction or alerting conspecifics is important. This is also supported by the fact that the intensity of alarm calls is more a function of distance of the predator to the nest (unpublished data), or generally the perceived risk. As chick anti-predator behaviour improves, we would expect the number of bouts of calling to the same level of potential risk to increase, while the intensity of alarm calls may vary less. Our data suggest that what changes across the nesting period is the probability of reacting to a predator; once a bird shows a reaction (to the same level of risk) there is less difference in the intensity of the reaction. Halupka & Halupka (1998) further argues that a significant correlation between alarm calling and behaviour of chicks (see Fig. 3 in Kleindorfer et al. 1996) need not necessarily reflect a causal
Commentaries relationship. Only after chick reactions were observed (irrespective of the intensity of reaction) did we find a pattern of increasing probabilities of parental alarm calling (irrespective of the intensity of calls). We agree that this is only a correlation and does not demonstrate any causality. However, the data discussed above on chicks jumping from the nest for two situations, namely with or without parental alarm calls, at least suggest that alarm calls do influence chick behaviour. Furthermore, the study showed that the proportion of chicks jumping from the nest tended to be higher for ground predators than for aerial predators, which we tentatively interpret as constituting an adaptive response. In playback experiments, we have already elicited different chick reactions, including jumping out of the nest, for bearded tits, Panurus biarmicus, and moustached warblers (H. Hoi, unpublished data). We agree with Halupka & Halupka (1998) that further experimental evidence is needed to understand more fully the social and environmental context of alarm calling. Our paper explicitly concluded with five areas of research in this regard. We strongly recommend disentangling alarm calls from the nest defence package to assess their relative contribution for predator avoidance strategies. One basic issue which Kleindorfer et al. (1996) addressed is the need to differentiate between the onset of sequential days of calling versus the intensity of calls since the underlying explanations for these may differ.
507
Halupka & Halupka (1998) have clearly shown the need to determine the relationship between the probability and the intensity of alarm calling. Another issue this study raises is the need to separate the levels of explanation for the occurrence of alarm calls, which may ultimately be related to the brood value hypothesis. Since anti-predator behaviour by chicks increases their brood value (measured in terms of survival probability), the question remains: do parents attend to rule-of-thumb measures of reproductive value or to more proximate cues such as chick behaviour?
REFERENCES Halupka, K. & Halupka, L. 1998. Alarm calls and chick reaction: comments on paper by Kleindorfer et al. Anim. Behav., 55, 502–503. Kleindorfer, S., Hoi, H. & Fessl, B. 1996. Alarm calls and chick reactions in the moustached warbler, Acrocephalus melanopogon. Anim. Behav., 51, 1199–1206. Leisler, B. 1991. Acrocephalus melanopogon. In: Handbuch der Vo¨gel Mitteleuropas (Ed. by K. M. Glutz von Blotzheim & K. M. Bauer), pp. 217–251. Wiesbaden: Aula Verlag. Lienert, G. A. 1986. Verteilungsfreie Methoden in der Biostatistik. Vol. 1. Meisenheim: Anton Hain. Montgomerie, R. D. & Weatherhead, P. J. 1988. Risks and rewards for nest defence by parent birds. Q. Rev. Biol., 63, 167–187. Weatherhead, P. J. 1989. Nest defence by song sparrows: methodological and life history considerations. Behav. Ecol. Sociobiol., 25, 129–136.