Condor, 82:4061116 @ The Cooper Ornithological Society 1980 SEX-SPECIFIC DIFFERENCES IN WINTER DISTRIBUTION PATTERNS OF CANVASBACKS JAMES D. NICHOLS AND G. MICHAEL HARAMIS ABSTRACT.-Winter band recovery distributions of North American Can- vasbacks (Aythya walisineria) suggested that males and females exhibit com- parable degrees of fidelity to general wintering areas. Of birds banded during the winter, the proportion of males was found to be higher in northern than in southern areas. Winter band recovery distributions of birds banded in particular areas during the summer were found to differ significantly between sexes, with females being recovered farther south. Factors that may have affected the evolution of sex-specific wintering distributions include: (1) pos- sible reproductive benefits derived by males who winter in the north and thus reach northerly breeding areas early; (2) sexual dimorphism in body size, which may render the smaller females especially susceptible to periods of inclement weather and food shortages; and (3) interactions between sexes in which males may control food supply when food is scarce. Two lines of evidence from field data on Canvasbacks in the Chesapeake Bay suggest the existence of competition between males and females. First, Canvasbacks trapped during winter in smaller bodies of water tended to have higher pro- portions of females and weigh less than birds trapped in large open bodies of water. Second, analysis of aerial photographs of wintering rafts of Canvas- backs showed patterns of intersexual segregation, with females being found more frequently on peripheral areas of rafts. Lack (1954, 1966) and Fretwell (1972) have ratios of North American waterfowl vary suggested that “limitation” of temperate geographically during the winter. Hoch- bird populations most likely occurs during baum (1944: 132) suggested that Canvas- the winter or nonbreeding season. It is backs (Aythya valisineria) may exhibit a therefore of interest to examine the distri- differential distribution of sexes on the win- bution and abundance of bird populations tering grounds. Allen (1931:354) and De- during this period of the year. In many Graff et al. (1961:79-80) noted the high pro- species of migratory birds in the Northern portion of males among Canvasbacks Hemisphere, males appear to winter farther wintering in New York, and suggested that north than females (see Nice 1933, Lack females were more numerous farther south. 1944, Selander 1966, Ketterson and Nolan The possibility that male and female Can- 1976 and references therein). It has been vasbacks winter in different areas is inter- suggested, (Witherby et al. 1939:219) that esting from both biological and manage- female and young diving ducks in Great ment points of view. The continental Britain generally migrate farther south than Canvasback population is thought to exhibit adult males. Evidence of such a sexual dif- the most disparate sex ratio (65-70% male) ference has been provided in North Amer- of all North American ducks. This ratio (and ica for Common Mergansers (Mergus mer- the low “reproductive potential” which it ganser; Anderson and Timkin 1972) and implies) has been labeled “a fundamental Buffleheads (Bucephala albeola; Erskine problem in the population dynamics of the 1971), in Europe for Common Pochards Canvasback” (Trauger 1974). Canvasbacks (Aythya ferina) and Tufted Ducks (Aythya hatch approximately equal numbers of fuligula; Salomonsen 1968), Mallards (Anus males and females (Hochbaum 1944:51, platyrhynchos; Lebret 1950, Nilsson 1976), Sowls 1955:164); hence, the disparate pop- European Wigeon (Anus Penelope), Green- ulation sex ratio must be due entirely to dif- winged Teal (Anus crecca), Northern Shov- ferential mortality. Indeed, in an analysis of elers (Anus clypeata), and Pintails (Anus Canvasback banding and recovery data, we acuta; Lebret 1950). Bellrose et al. (1961) found mortality rates to be significantly summarized evidence suggesting that sex greater in females than males in each of the WINTER DISTRIBUTION PATTERNS OF CANVASBACKS 407 three populations examined (Nichols and where ri is the linear rank of observation i, n is the Haramis 1980). The existence of this differ- number of observations in the first sample, and m de- notes the number of observations in the second sam- ence emphasizes the importance of exam- ple. We then compute the resultant or vector sum of ining any sexual differences that may occur the first sample as: in distribution or general characteristics and habits. If winter mortality is as important to RI = [(k co~/ji)~+ ($ SinU)]‘ ” migratory bird populations as has been sug- gested, then the investigation of possible The null hypothesis of no difference between the two differences in the winter distribution pat- bivariate samples is then rejected for large values of terns of the two sexes may help to explain R,. Mardia (1967) has shown that for n + m > 17: the higher female mortality rates and resul- U = 2R:(m + fi - 1)lmn tant disparate sex ratios. Our primary objective was to test the gen- is approximately distributed as chi-square with two degrees of freedom. Under the hivariate normal distri- eral hypothesis that male Canvasbacks tend bution, the efficiency of Mardias’ test relative to Ho- to winter farther north than females. If this tellings’ T ’ (its parametric competitor) was found to lie hypothesis was corroborated, our secondary between 79% and 93% (Mardia 1968). Our own Monte objective was to explain the phenomenon Carlo simulation work with distributions that more or at least to obtain information on factors closely resemble band recovery patterns have resulted in approximated powers similar to those obtained us- that might have affected its evolution. ing hivariate normal distributions. “Ties” between ob- servations from the two samples occurred in some of METHODS our tests with empirical recovery distributions, and we computed approximate chi-square test statistics in the Our data were obtained from three sources: the com- manner suggested by Robson (1968). puter retrieval files of the U.S. Fish and Wildlife Ser- Aerial photographs of Canvasback rafts, taken at an vice Bird Banding Laboratory (BBL), Laurel, Mary- altitude of 150 m, were examined under a stereobinoc- land; aerial photographs of Canvasback rafts in the ular scope to identify males and females. Negatives Chesapeake Bay; and banding and weight records of were projected with an overhead viewer to map rafts Canvasbacks trapped in the Chesapeake Bay. The BBL for tests concerning the relative distribution of the data pertained to the period 1929-77, whereas the oth- sexes. er data were collected during the winter of 1977-78. The BBL data used here included 2,663 recoveries and We investigated the winter weight-loss of Canvas- backs by trapping them in mid-January and again in over 55,000 bandings. Only birds categorized as “nor- mal, wild” were used in our analyses. Recovery rec- early March 1978 in upper Chesapeake Bay. Welded- wire diving duck traps baited with corn were used. ords were restricted to birds shot or found dead. All Trapped birds were held for a minimum of 6 h before birds banded during the winter (January-February) weighing to permit the drying of plumage and the pas- were considered to be adults, whereas summer-banded sage of much of the ingested bait. This procedure re- birds (July-September) were aged as either young duced the variance of our data. Before weighing, we (first year) or adult (older than one year). inspected the gullet of each bird for corn content; only In two portions of our analysis we tested the null individuals containing little or no corn were consid- hypothesis that geographic distributions of band recov- ered in the analysis. Birds were weighed to the nearest eries from two samples came from the same popula- tion. Each observation (band recovery) is characterized 5 g. by latitude and longitude of recovery measured to the nearest lo-min coordinate. The statistical problem is RESULTS AND DISCUSSION thus one of testing whether two-sample distributions belong to the same bivariate population. Kolmogorov- TEMPORAL VARIATION IN WINTER DISTRIBU- Smirnov test statistics (e.g., Sokal and Rohlf 1969:571- TION PATTERN 575) were computed to test the null hypothesis that the empirical marginal distributions of latitude and lon- We wanted to learn whether Canvasbacks gitude (longitude was transformed via the relationship return to the same general wintering areas 1” longitude = (1” latitude) (cos A), year after year, because we believed that the presence or absence of such a tendency where A is the latitude; see Raisz 1962:147) were uni- variate normal. This null hypothesis was rejected (I ’ < would be important in interpreting any pat- 0.01) for both latitude and longitude in the first several tern of geographic variation in sex ratio that recovery distributions tested, and we concluded that might exist. We chose to address this ques- parametric tests based on the assumption of bivariate tion at the population level first by compar- normality would be inappropriate for these data. For testing pairs of empirical distributions of band ing direct (first year after banding) recovery recoveries we used the nonparametric test suggested distributions from particular winter banded by Mardia (1967, 1972: 197-201). Briefly, this test first samples with indirect (second and later involves computing the centroid or center of gravity of years after banding) recovery distributions the combined two-sample distribution. Vectors are (similar rationale was employed by Cowar- then considered from this centroid through each sam- ple point (band recovery), and the points are ranked, din 1977:6-7). For this analysis we used based on the vector directions. These directions or an- BBL data from Canvasbacks banded during gular observations are then replaced in the first sample January and February in particular states by: during specific years, and recovered during pi = 2?T r,/(n + m), i = 1, . n, the period from 15 December through 28 408 JAMES D.