Lifetime Reproductive Success, Selection, and the Opportunity for Selection in the White-Tailed Skimmer Plathemzs Lydza (Odonata: Libellulidae)

Evoluf~on,41(1), 1987, pp. 22-36

LIFETIME REPRODUCTIVE SUCCESS, SELECTION, AND THE OPPORTUNITY FOR SELECTION IN THE WHITE-TAILED SKIMMER PLATHEMZS LYDZA (ODONATA: LIBELLULIDAE)

WALTERD. KOENIG Hustings Reservation and Museum of Vertebrate Zoology, University of California, Carmel Valley, CA 93924

AND STEPHENS. ALBANO' Moore Laboratory of Zoology, Occidental College, Los Angeles, CA 90041

Abstract. -We present estimates of lifetime reproductive success in Plathemis lydia, a temtorial dragonfly. We partition the opportunity for selection into multiplicative episodes using the techniques of Arnold and Wade (1 984a, 1984b) and measure selection on several morphological and behavioral characters. For both sexes, variance in survivorship was the largest contribution to variance in lifetime reproductive success. Covariance effects are also strong for both sexes, sug- gesting considerable non-independence of episodes. Opportunity for selection calculated on a daily basis did not approximate analogous values determined from lifetime reproductive success. Phenotypic characters for which we investigated selection included body mass, hind wing length, first date of reproduction, and (for males) an index of temtorial aggressiveness. We failed to find any significant direct targets of selection in either males or females. However, the combined effects of direct and indirect selection on early reproduction were significant for males, acting primarily through increased su~ivorshipand increased time per day spent at the pond. Similarly, females present earlier in the season had shorter interclutch intervals. Partitioning of selection acting on male hind wing length and on aggressiveness reveals relationships between selective episodes, possibly indicative of phenotypic trade-offs between natural and sexual selection through male- male competition for females. Division of selection into episodes is a useful technique for identifying the source of selection. However, ordering effects can bias results, except when episodes occur in strictly chronological sequence. We present a method for circumventing this difficulty.

Received July 22, 1985. Revised December 16, 1985. Accepted September 23, 1986

Despite general agreement that lifetime cially if variance in mating success is de- reproductive success is the best measure of termined in part by age- or size-correlated fitness that can be collected (see Cavalli- characters (Arnold and Wade, 1984a). De- Sforza and Bodmer, 1961; Grafen, 1982; tecting such trade-offs is critical for deter- Clutton-Brock, 1983; Partridge and Halli- mining the relative strengths of natural and day, 1984), measures of this critical param- sexual selection. eter have been obtained to date in only a Here we present data on lifetime repro- handful of species (see, for example, Clut- ductive success in the white-tailed skimmer ton-Brock, 1987). Instead, analyses of se- Plathemis lydia (Odonata: Libellulidae). lection are typically performed on longitu- Adults of this species do not overwinter, dinal data from one or more breeding and survivorship of adults of both sexes is seasons, or on cross-sectional data gathered 83-89% per day (Koenig and Albano, 1985); on a single selection event. Such data can thus, a reasonable number of individuals provide important insights into the work- can be followed throughout their lifetimes. ings of selection, but may nonetheless be We examine the opportunity for selection misleading. In particular, phenotypic trade- (I = a2/R2,the variance in individual fitness offs between survivorship and other com- divided by the mean fitness squared) and ponents of fitness may eliminate or even selection itself on several characters. Z val- reverse trends found during a fraction of the ues are calculated following Arnold and lifespan of an organism (Bell, 1980), espe- Wade's (1 984a, 1984b) method of partitioning selection occurring in an arbitrary

I Present address: Department of Zoology, Univer- number of ''e~isodes." In order to masure sity of California, Berkeley, CA 94720. selection on correlated characters we use the 22 SELECTION IN DRAGONFLIES 23 partial correlational techniques of Lande and position sites (Koenig and Albano, 1985). Arnold (1 983). These techniques provide a Further details concerning the natural his- powerful program for analyzing and inter- tory of Plathemis lydia can be found in Ja- preting selection. cobs (1 959, McMillan (1 984), and Koenig and Albano (1985). Biology of Plathemis lydia Plathemis lydia is a medium-sized drag- MATERIALSAND METHODS onfly common throughout much of North The study was conducted at Blompond, America. Sexual dimorphism in size is slight a small (380 m2)man-made stock pond lo- (see below). However, the sexes are dichro- cated adjacent to Hastings Natural History matic; males have bluish-white abdomens Reservation, Monterey Co., California. The and wings with a dark median band, while perimeter of the pond was marked at 3-m females have dark brown abdomens and intervals. Beginning on 10 June 1984, we wings with median and terminal dark spots caught as many individuals as possible. The (see Jacobs, 1955). Males display their ab- wings of captured individuals were secured domens in territorial chases (Jacobs, 1955; with a hairpin, then we weighed individuals Campanella and Wolf, 1974) but not in to the nearest 0.01 gm, measured hind wing courtship. Thus, differing selective pres- length, and marked wings and abdomens sures on the sexes have apparently influ- with dots of enamel paint for individual rec- enced body and wing coloration to a greater ognition. extent than size. We marked 180 males and 84 females at Plathemis breeds in ponds; males defend Blompond during the study. Of these, 78 ephemeral territories 10 m in length males (43%) and 38 females (45%) were not around the perimeter of the ponds. At me- subsequently seen at Blompond. This level dium-to-high population densities, individ- of disappearance following marking is com- ual males spend only a few hours a day in parable to that reported earlier (Koenig and territorial behavior (Koenig and Albano, Albano, 1985), and is probably the conse- 1985). Consequently, territories are defend- quence of emigration following the trauma ed by a series of males during a day, al- of capture and handling. Virtually all indi- though the same male may return to a ter- viduals were marked as breeding adults. ritory (usually but not always the same one) Breeding activity at Blompond begins in the on successive days (Koenig and Albano, second week of May; thus, we conducted 1985). our study during the latter half of the breed- Females visit ponds solely to mate and ing season. lay eggs. Female arrival is relatively syn- Between 19 June and 5 August (no Plath- chronous, peaking in the early afternoon emis remained at Blompond after this date), (Campanella and Wolf, 1974; Koenig and we and two field assistants were present from Albano, 1985). Copulation lasts only a few 0700 (before any dragonflies arrived at the seconds; then the male disengages and hov- pond) until all individuals had left for the ers above the female while she flies over the day (usually by 1830). Every 20 minutes we water looking for an oviposition site ("non- recorded all males present as well as (if ter- contact" guarding; Parker, 1970; Alcock, ritorial) their temtorial boundaries. In ad- 1982; Sherman, 1983; Waage, 1983; dition, we continuously scanned the pond McMillan, 1984). The male continues to for females. Whenever a copulation or ovi- guard the female during oviposition (usu- position was observed we recorded its lo- ally somewhere in the male's territory); then cation, length, and the identity of partici- the female immediately leaves and the male pants. resumes territorial behavior. Eggs develop McVey (1984) found that Plathemis in unattended. New York oviposited at a mean rate of 22.8 This mating system can be considered to eggs per second. However, we do not have be one of resource defense polygyny (Emlen comparable data from our population, and and Oring, 1977; Thornhill and Alcock, thus we report reproductive success in units 1983), with males monopolizing females in- of "seconds of oviposition." For males, re- directly by defending spatially clumped ovi- productive success was estimated from the 24 W. D. KOENIG AND S. S. ALBANO seconds of oviposition of females, but only if the male was the female's last mate (see below). The three nearest breeding sites to Blom- pond are approximately 0.6, 1.2 and 1.7 km away. The two nearest of these sites were surveyed for marked individuals one or two times per day for 15 to 20 min each per visit. This allowed us to detect many males emigrating to and remaining at either of these ponds. However, because of the brief- ness of female visits, we were unlikely to see marked females ovipositing at these sites. Days between oviposi t ions To examine the opportunity for selection, FIG. 1. Distribution of the number of days between observed ovipositions for marked females at we compiled two sets of data. The first con- Blompond (R f SD). sisted of daily reproductive success. For each individual seen at least once on a day following marking, the number and length of when we began continuous watches; 3) their ovipositions were compiled for every day wings were fresh when captured; and 4) they alive following marking. We then analyzed were not observed at another pond. As ex- these "daily" data by: 1) including individ- ceptions to (4), we included 10 males last uals only on days they were observed at the seen at another ~ondon one or two occa- pond, and 2) including individuals on all sions after having been present at Blom- days, whether they came to the pond or not, pond for at least several days. Such indi- as long as they were seen on a later day and viduals appeared to be following a pattern thus known to have been alive. Reproduc- of post-reproductive wandering (Koenig and tive success for individuals on days they Albano, unpubl.) and are unlikely to have were not observed at the pond was set to reproduced after emigrating from Blom- zero. For males, this assumption is based pond. on the direct observation that males rarely A total of 46 males met the above criteria wandered temporarily to other ponds. With (45% of the 102 males marked at Blompond few exceptions, females also appeared not and observed at least once on a dav sub- to switch ponds (see Fig. 1 and below). sequent to capture). Because virtually all Females ovipositing at the pond did so at males at Blompond were marked during the intervals of 2.2 1 * 1.47 days (x * SD; N = study, we are confident of having captured 87 intervals from 3 1 individuals). Intervals these 46 individuals soon after they reached ranged from 1 to 5 days with a strong mode reproductive maturity. These data thus es- at 2 days, except for one interval each of 7 timate lifetime reproductive success. Un- and 12 days (Fig. 1). These two longest in- fortunately, because a high proportion of tervals were eliminated, as it is likely that matings (85.3% of 1,242) involved un- these females oviposited elsewhere during marked females, we were unable to obtain these periods. In total, we used data on 9 12 lifetime data for this sex. However. 42 female and 29 1 female dragonfly-days taken males were observed after marking for from 113 individual males and 50 individ- which we had data on all subsequent mat- ual females. Males were present on the pond ings and ovipositions. In order to avoid bias for 672 (73.7%) dragonfly-days and females arising from data from individuals seen only for 132 (45.4%; oviposition actually oc- once (Fincke, 1986), we used only data col- curred on 126 of these). lected after the date females were marked. The second data set consisted of total re- These data corres~ondto an unknown frac- productive success. For males, we included tion of lifetime reproductive success. (If, individuals meeting the following criteria: however, there is an even adult sex ratio, as 1) they were observed at least once on a day was found in Plathemis lydia by Campa- following marking; 2) they were marked at nella and Wolf [1974], this fraction can be Blompond on or subsequent to 19 June, roughly estimated by the ratio of the mean SELECTION IN DRAGONFLIES 25 total number of eggs laid by females divided unable to mark females the first time they by the mean lifetime number of eggs fertil- came to the pond. Nonetheless, although ized by males; this ratio [from the mean provisional, this measure still reflects sea- lifetime seconds of oviposition in Table 11 sonal trends in fecundity, and thus assesses is 0.4 1.) Assuming that the fraction of life- selection on adults acting to alter the sea- time reproductive success sampled is un- sonal timing of reproduction. biased, trends derived from these data will We also measured selection on body mass be unaffected. Thus, the "total" reproduc- and hind wing length. In order to eliminate tive success we obtained for females, al- measuring the mass of eggs carried by fe- though not as reliable, provides data which males coming to oviposit, we only included can at least be tentatively compared with body mass for females weighed after ovi- our more complete data from males. position. We do not, however, have data on We arranged and analyzed these mea- fluctuations in body mass over the lifespan sures as multiplicative "episodes" of selec- in individuals. Both characters were log tion (Arnold and Wade, 1984a, 1984b). For transformed prior to analysis. males, the episodes considered included: 1) We also measured the relative time spent days alive after marking, 2) proportion of in flight by territorial males as an index of days alive in which the individual was ob- territorial aggressiveness. We determined served at the pond, 3) number of hours per this character as follows. Focal watches day spent at the pond, 4) matings obtained (Altmann, 1974) were conducted on arbi- per hour present at the pond, 5) proportion trarily chosen territorial males. Watches of matings in which the female oviposited lasted 1 hr or until the male left the pond. for at least one second, and 6) mean dura- During the watch, the location and activity tion of ovipositions obtained by mates. of the focal male and the number of con- These episodes measure survivorship (I), specific males within 1 m of the focal in- reproductive effort (2 and 3), mating effi- dividual were recorded at 12-sec intervals. ciency (4), mate-guarding ability (5), and a For each watch, we determined the propor- combination of guarding ability and female tion of time the focal male spent in flight, fertility (6). For females, episodes were: 1) and the proportion of time one or more con- days alive after marking, 2) proportion of specifics were present within 1 m of the focal days alive in which the individual was ob- male. Both measures were arcsin trans- served to have oviposited at the pond, and formed for analysis. 3) duration of oviposition obtained per day Since the activity of territorial males in- on days the individual was observed. The creases with the density of intruding con- first of these episodes again measures sur- specifics (Koenig and Albano, 1985), we vivorship. Because females produce only one performed a regression of the time spent in clutch every several days (see above), the flight on the proportion of time that no con- second episode estimates the clutch inter- specifics were present within 1 m ofthe focal val, while the third estimates primarily male. As expected, the proportion of time clutch size and, to a lesser extent, the ability spent flying by territorial males increases to avoid or circumvent interference during with the density of intruding conspecifics oviposition. For both males and females, (F= 34.9, d.j = 1,46,R2= 0.43, P < 0.001). the episodes we considered do not occur We then calculated the standardized re- strictly sequentially, and thus the precise or- siduals of the time spent in flight from the dering of episodes is unavoidably arbitrary above regression for each focal watch. Next (see below). we averaged these residuals for all focal We tested for selection on four characters. watches conducted on the 13 males for which The first was the date individuals were we eventually acquired lifetime data. This marked. For males, this measure represents yielded an index ofterritorial aggressiveness seasonal effects and tests whether individ- for these males, controlling for differences uals emerging early in the season were at an resulting from population density. In all, we advantage compared to those emerging lat- performed 35 focal watches (1-7 per male, er. For females, this measure does not di- R = 3) lasting a total of 26 hours for these rectly indicate emergence date, as we were 13 individuals. 26 W. D. KOENIG AND S. S. ALBANO

Following Arnold and Wade (1984a), the leading when episodes do not occur strictly selection differential (Falconer, 198 1) at ep- sequentially. In contrast, (3,' values are de- isode k, s,, was calculated as the shift in the fined so as to be independent of any ordering mean resulting from the action of selection effects, and thus provide an unbiased mea- during the current episode, starting from the sure of selection occurring due to the in- mean at the beginning of the episode. In dependent effects of a given episode. Fur- contrast, s,' values were calculated as the thermore, (3,' values can be directly tested shift in the original mean occurring through for statistical significance by examination of the action of each episode of selection in- the regression coefficient. Neither (3, nor (3,* dependent of any other episodes. Thus, s,' values can be readily derived from a regres- values provide a measure of the relative im- sion analysis, and thus cannot be directly portance of selection at each episode inde- tested for significance. pendent of the ordering of the episodes (see Several limitations of our data and anal- below). yses should be noted. First, we assumed that Selection intensity (Falconer, 198 1) at ep- all eggs were fertilized by the male mating isode k, i,, was calculated by dividing the with a female immediately prior to ovipo- selection differential s, by the standard de- sition. This is consistent with data on sperm viation ofthe original character (Arnold and competition in odonates, which indicate that Wade, 1984~).Similarly, i,' values are cal- the last male to copulate fertilizes the vast culated by dividing s,' by the standard de- majority (an average of 99.5% in Erythemis viation of the original character. In addi- simplicicollis) of the eggs she lays (McVey tion, we calculated a third measure of the and Smittle, 1984; see also Waage, 1979, selection intensity, i,*. This value is the shift 1983; Fincke, 1984). However, because of in the mean due to selection at an episode the extremely short copulations in Plath- in units of the standard deviation of the emis lydia (~3sec [McMillan, 19831 com- character at the start of the action of that pared with 12-1 8 sec in Erythemis [McVey, episode, rather than the original standard 198 I]), it is unlikely that sperm precedence deviation. Of these three measures of selec- is complete. A second problem involves fe- tion intensity, only s, values are additive males that did not mate immediately prior (Arnold and Wade, 1984~). to oviposition; such cases involved 13.9% Selection gradients ((3,) are the regression of ovipositions. Sperm precedence by the coefficient of standardized fitness on the last male to mate drops considerably on character at episode k; this is mathemati- subsequent days (McVey and Smittle, 1984). cally equivalent to the selection differential Combined with the low proportion of (s,) divided by the variance in the character marked females in our study, this means (Arnold and Wade, 1984~).As with selec- that paternity of eggs laid by lone females tion intensities, we calculated three (3 val- could not be determined. Third, a small ues. For the first (P,), we divided by the proportion of reproduction involved un- original variance in the character at each marked or unknown males; 2.2% of the episode. For Pk*,we divided by the changed 1,132 ovipositions in which the female mat- variance at the start of each episode rather ed shortly beforehand, and 1.6% of the 16.5 than the original variance. Finally, for (3,' hr of oviposition seen after an observed values, we independently regressed each ep- mating were not attributable to a known isode of fitness on the original distribution male. of the character. Although only (3, values We also did not correct for the temper- are additive, the other two selection gradi- ature dependence of oviposition rate ents provide a less biased measure of selec- (McVey, 1984; Fincke, 1985). However, be- tion at each episode. (3,* values take into cause of the thermoregulatory abilities of account the degree to which the character dragonflies (May, 1976; Heinrich and Ca- variance is changed at each episode, and sey, 1978), it is likely that abdomen tem- thus correspond to true selection gradients peratures vary much less than the ambient (Kalisz, 1986). However, both (3, and (3,* temperature. To the extent that this is true, values are potentially sensitive to ordering oviposition rates will be relatively uniform, effects of the episodes, and thus may be mis- and consequently the length of oviposition SELECTION IN DRAGONFLIES 27

TABLE1. Intersexual comparison of the opportunity for selection (U~/X~)in Plathemis lydia.

Mean c2 N I Idre Days alive Males 8.87 62.6 46 Females 6.04 25.0 50 : } 1.17 Seconds of oviposition Lifetime or total Males 727 1.58 x lo6 46 2.99 Females 299 1.13 x lo5 42 1.26 } 2'37 Daily (all days) Males 64 1.52 x lo4 912 3.65 Females 47 4.17 x lo3 29 1 1.87 } Daily (only days when present at pond) Males 8 8 1.86 x lo4 672 10.13 Females 104 3.28 x lo3 132 i::: bouts will provide a good index of number on a daily basis than on a lifetime basis of eggs fertilized or laid. Because the num- when all days are included, but much higher ber of males and females at the pond varies on a daily compared with lifetime basis when continuously, we have also not corrected only days in which individuals appeared at our I values for the operational sex ratio the pond are included. The distributions of (Emlen and Oring, 1977). However, the op- lifetime (for males) and total (for females) erational sex ratio is almost always biased reproductive success are presented in Figure towards males (at any one time, there are 2; data for daily reproductive success are usually no females at all at the pond), and presented in Figure 3. thus the effective intensity of selection may Partitioning of the opportunity for selec- be considerably greater on males than our tion on lifetime reproductive success in results indicate (Wade and Arnold, 1980). males is presented in Table 2. Listed are the Also, because of small sample sizes, we have opportunity for selection associated with in- not attempted to analyze for stabilizing or dividual episodes (w,), the running subtotals disruptive selection. Finally, we cannot be for the total opportunity for selection (wiw, certain that some of the characters we use, . . . w,), the opportunity for selection asso- such as the index of aggressiveness and body ciated with correlations between fitness mass, do not change ontogenetically. components (the "cointensity" terms), and Our results must thus be considered pro- the total opportunity for selection (W). Also visional until these shortcomings can be listed are the percentages of the total op- dealt with satisfactorily. portunity for selection represented by each episode and subtotal. RESULTS By far the largest single contribution to I (2'7% of the total) is provided by variance Opportunity for Selection in days alive (survivorship). Matings per Table 1 summarizes the opportunity for hour present at the pond (mating efficiency) selection (I)for survivorship and reproduc- is second with 11 O/o of the total. Hours pres- tive success. In all cases I values are higher ent per day on the pond contributes another for males than females, although those for 6%, while each of the other three episodes longevity are similar. When all days are in- accounts for only 2% or less of the total cluded, I values are higher for daily repro- opportunity for selection. The various coin- ductive success than for lifetime or total re- tensity terms, measuring in part covariances productive success. The opposite is true, among variables (Arnold and Wade, 1984a), however, when only days in which individ- are generally positive and often substantial, uals were actually present are included. The contributing a total of 52% to the total op- Z,lI, ratio for reproductive success is lower portunity for selection. 28 W. D. KOENIG AND S. S. ALBANO

7 0 40h Males Males

50 All days (R=64?123,N=912) cn Days when present 30 (P=87'137,N=672) 0 -0 r 10 0 60 Females

40 Days when present 20 (P=104?57, N.132)

O 0 '100 (200 ~400'800 (1600 Seconds of oviposit ion FIG. 3. Frequency distribution of daily reproductive success for male and female Plathemis lydia. The most successful male obtained a total of 890 sec (14.8 min) of oviposition in one day; the most successful female oviposited for 379 sec (6.3 min). Data (3 + SD) are given for all days and for only those days when Seconds of oviposit ion individuals were observed at the pond. FIG. 2. Frequency distribution of lifetime reproductive success (males) and total reproductive success (females) for Plathemis lydia, based on seconds of ovi- Correlations between the four characters position (3+- SD). used in the analyses and between the characters and the fitness components we measured are given for lifetime or total repro- The analogous partitioning of the total ductive successin Table 4. There is a strong opportunity for selection on females is pre- seasonal effect on lifetime reproductive suc- sented in Table 3. Survivorship contributes cess in males (due to effects on survivorship nearly half (48%) to the total I value. and time spent per day on the pond), and a Clutches per day alive (the interclutch in- weaker, but nonsignificant (P < 0.10), sea- terval) contributes 14% to I, while clutch sonal effect on total reproductive success in size contributes only 10%. Covariance ef- females; for both sexes, earlier individuals fects are again strong, contributing a total have a reproductive advantage. There was of 28% to the total I. no significant correlation between aggressiveness (time spent in flight) by males and Selection on Correlated Characters lifetime success; however, there was a sig- The two morphologic characters mea- nificant positive correlation between ag- sured (hind wing length and body mass) are gressiveness and the matings per hour males both indicative of body size. Mean + SD achieved while at the pond. There were also (N)hind wing length was 31.57 + 0.87 mm several significant correlations between (276)'for males and 32.6 1 + 0.95 mm (94) characters; for both sexes, body mass de- for females; mean body mass was 0.462 + clined seasonally (see Banks and Thompson 0.035 g (271) for males and 0.470 + 0.044 [I9851 for a similar result in the damselfly g (22) for females. Females have signifi- Coenagrion puella), and heavier individu- cantly longer hind wings than males (AN- als had longer wings. OVA, F = 95.2, d.J = 1,368, P < 0.001), The relative magnitude of direct selection but there is no sexual dimorphism in body on a trait, controlling for correlated effects mass (F = 0.9, d.J = 1,291, ns). Coeffi- on other measured traits, is indicated by the cients of variation for both characters are standardized partial regression coefficient all relatively small, between 2.0 and 2.9%. (p) resulting from a multiple regression of SELECTION IN DRAGONFLIES 2 9

TABLE2. Partitioning of the opportunity for selection on lifetime reproductive success in male Plathemis lydia (N = 46 individuals). Lines separate running subtotals.

Contnbutlon to total opportunlty for selection Source of varlatlon ~n relat~vefitness Symbol Value Percentage Days alive (wl) I1 0.81 27% Proportion of days present at pond (w2) I2 0.07 2% Cointensities (12) 0.00 0% Days present at pond (wl w2) 112 0.84 2 8O/o Hours per days on pond (w3) 13 0.17 6% Cointensities (1 23) 0.50 17% Hours present (wlw2wg) 1123 1.5 1 50% Matings per hour present (w4) 14 0.33 11O/o Cointensities (1 234) 0.53 18% Matings (w 1w2 w3 w4) I1234 2.37 79% Ovipositions per mating (w5) Cointensities (12345) Ovipositions (wl w2w3wqwg) Eggs fertilized/oviposition (~6) Cointensities (123456) Eggs fertilized (total selection, U3 * Colntensltles Include covariance terms between the current episode and all pnor episodes; e.g., "Colntensltles (12)" = COI(1.2) + COI(1212) + C01(12,21) - C01(12,2). For formulae, see Arnold and Wade (19840). fitness on the characters involved (Lande For females, those present earlier in the and Arnold, 1983). The results of such anal- season had higher fitness, and this effect act- yses using the three non-behavioral char- ed through all three episodes (Table 6). The acters are presented in Table 5. For males, total effect of selection was to figuratively the only significant effect is that between shift the mean date of marking 0.31 stan- survivorship and hind wing length; individ- dard deviation units earlier in the season. uals with wings one standard deviation Based on the P' values, the only significant shorter than the mean experienced relative effect is due to decreased interclutch interval survivorship 47% longer than average. All early in the season. three characters have comparable direct in- For males, selection again consistently fa- fluences on lifetime fitness; none are signif- vors seasonally early individuals, except for icant. For females, there are again no sig- a small tendency for later males to obtain nificant direct effects of the three characters more eggs per mating. In sum, selection fig- on either survivorship or total reproductive success. TABLE3. Partitioning of opportunity for selection on Episodes of Selection total reproductive success in female Plathemis lydia (N = 42 individuals). We calculated selection parameters for characters showing potentially significant ef- Contribution to total fects on some episode of fitness as indicated opportunlty for selection Source of varlatlon Sym- by the analyses in Tables 4 and 5. The char- ~nrelat~ve fitness bol Value Percentage acters included were first date on the pond Days alive (wl) I1 0.60 48% for both sexes, male hind wing length, and Clutches per day alive (w2) I2 0.18 14% the index of territorial aggressiveness (time Cointensities (12) * -0.03 -2% spent in flight). The results are presented in Clutches (wl w2) 112 0.75 60°/o Table 6. First we list the mean and variance Seconds of oviposition per clutch (w3) r3 0.13 10% of the character following the combined ac- Cointensities (1 23) * 0.38 30% tion of all prior episodes; then the selection Seconds of oviposition differentials, selection intensities, and the (total selection, W) I 1.26 100% selection gradients. * See footnote to Table 2. 30 W. D. KOENIG AND S. S. ALBANO

TABLE4. Correlations with episodes of lifetime reproductive success (males) and total reproductive success (females). Both body mass and hind wing length are log transformed. For explanation of the index of aggressiveness (percent time in flight) see text. Sample sizes are given in parentheses. Two-tailed test: *P < 0.05; **P < 0.01.

-- - - Indrx of Variable Date marked Body mass H~ndwlng length aggressrveness Males Body mass Hind wing length Survivorship Hours present per day alive Matings per hour on pond Seconds of oviposition per mating Total seconds of oviposition Females Body mass Hind wing length Survivorship Interclutch interval Clutch size Total seconds of oviposition uratively shifts the date of marlung for males part to a trade-off between episodes. In par- 0.64 standard deviations earlier in the sea- ticular, more aggressive territory holders son. The majority of this effect results from obtained significantly greater mating effi- significantly increased survivorship and in- ciency, but this effect was largely cancelled creased time per day males spend on the by the shorter number of hours per day such pond early in the season. individuals spent at the pond. There is little overall selection on male DISCUSSION hind wing length. However, this probably results from a trade-off between survivor- Opportunity for Selection ship and the other episodes of selection, par- The dominant contribution to the op- ticularly mating efficiency; individuals with portunity for selection for male Plathemis longer wings obtain more matings per hour lydia is through differential survivorship at the pond but do not live as long as short- (Table 2). The second most important con- winged individuals. This trend is also sug- tribution is from mating efficiency, followed gested by the correlations in Table 4, al- by hours present per day at the pond. Ad- though they are not significant. ditional components, except for the covari- For the 13 males for which we had both ance terms, are negligible. Unfortunately, behavioral and lifetime data there was little the covariance terms are not readily inter- overall selection on time spent in flight (the pretable (Arnold and Wade, 1984a); how- index of territorial aggressiveness), due in ever,'they reflect at least in part, correlations

TABLE5. Selection gradients (standardized partial P i SE) for date marked and body mass with survivorship and reproductive success. N = 40 males and 15 females.

Episode Date marked Body mass Hind wlng length Multiple ~2 Males Survivorship -0.31 f 0.16 0.22 f 0.15 -0.47 f 0.15** 0.28 Lifetime eggs fertilized -0.29 f 0.19 0.16 f 0.17 -0.21 f 0.17 0.15 Females Survivorship 0.06 i 0.36 0.07 i 0.40 -0.09 -t 0.37 0.0 1 Total eggs laid -0.07 i 0.36 0.10 ? 0.40 -0.06 i 0.37 0.02 **P < 001. TABLE6. Analysis of directional selection on seasonal timing, body size, and aggressiveness (measured as total time in flight, see text) in Plathemis lydia based on lifetime reproductive success (males) and total reproductive success (females). Means * c2 are calculated at the start of each episode, except for the values given for "total selection," which are the final values. For date on pond, mean values are days since 1 June; those for total time in flight are residuals from the predicted regression value (see text). For discussion of the selection gradients, see text.

Selection dlfferent~als Selectron intensities Selection gradients Character Selectron ep~sode z ? 2 4 st' lk ~k* llr' 0 8* 8'

Females: date on pond (days, N = 42) Survivorship 27.3 ? 156.4 -1.29 -1.29 -0.10 0.10 0.10 Clutch interval 26.0 + 139.2 -1.60 -2.26 -0.13 -0.14 -0.18 Clutch size 24.4 + 122.5 -1.02 -1.04 -0.08 -0.09 -0.08 Total selection 23.4 -C 118.8 -3.91 - -0.31 - -

Males: date on pond (days, N = 46) Survivorship 33.8 + 127.5 -4.23 -4.23 -0.37 -0.37 -0.37 Hours present per day 29.6 + 79.7 -2.15 -3.38 -0.19 -0.24 -0.30 Matings per hour present 27.4 -C 59.9 -0.83 -2.64 -0.07 -0.12 -0.23 Seconds of oviposition per mating 26.6 + 47.6 0.05 1.23 0.00 0.01 0.1 1 Total selection 26.6 + 56.5 -7.17 - -0.64 - - Males: hind wing length (N = 38) Survivorship 3.461 k 5.38 x -0.0054 -0.0054 -0.23 -0.23 -0.23 Hours present per day 3.456 ? 7.44 x 0.0008 -0.001 3 0.03 0.03 -0.06 Matings per hour present 3.457 ? 6.44 x 0.0025 0.0027 0.1 1 0.10 0.12 Seconds of oviposition per mating 3.459 k 8.41 x 0.0005 0.0029 0.02 0.02 0.13 Total selection 3.460 2 7.31 x -0.0016 - -0.07 - -

Males: time spent in flight (N = 13) Survivorship -0.93 k 162.2 0.689 0.689 0.05 0.05 0.05 Hours present per day -0.24 k 157.7 -1.762 -1.891 -0.14 -0.14 -0.15 Matings per hour present -2.01 ? 181.4 4.177 3.733 0.33 0.31 0.29 Seconds of oviposition per mating 2.17 k 186.6 -0.893 0.529 -0.07 -0.07 0.04 Total selection 1.28 4 216.1 2.210 - 0.17 - - Statrstical tests were performed on the 8 value for total select~ononly and on all 13' values, * P < 0.05, ** P < 0.001. 32 W. D. KOENIG AND S. S. ALBANO between selective episodes. Thus, in Table fluencing survivorship will have consider- 2, the large positive COZ(123) and able opportunity to exert strong positive ef- COZ(1234) terms suggest that males obtain- fects on lifetime fitness regardless of their ing more matings per hour also tend to be influence on other episodes of selection. For at the pond longer per day and to live longer example, a reduction in reproductive effort as well. In general, there is no indication of which serves to increase survivorship trade-offs between performance at different nevertheless may increase fitness. Con- selective episodes for males. versely, characters influencing short-term For females, differential survivorship reproductive success may have little overall again dominates, contributing 48% to the effect on fitness. total opportunity for selection (Table 3). Of For males, our analysis differs from sim- secondary importance is variance in the in- ilar analyses presented earlier (e.g., Arnold terclutch interval, followed by variance in and Wade, 1984b) in that we divided fitness clutch size. into a relatively large number of episodes. Similar to our findings here, studies on This procedure has the potential advantage other odonates have also indicated that sur- of identifying very specific episodes of im- vivorship contributes substantially to vari- portance out of the wide spectrum of selec- ance in lifetime reproductive success. Fincke tive episodes experienced during the life- (1982) reported relative variance in male time of an individual. A potential survivorship of the damselfly Enallagma disadvantage, however, is the production of hageni to be 78% of the variance in lifetime relatively large "cointensity" terms which mating success; for the same species, Fincke are not readily interpretable, although they (1986) found that Z values for male survi- represent, at least in part, covariance be- vorship contributed 39% to the total life- tween episodes (Arnold and Wade, 1984a). time fertilizations achieved by males, Thus, it may be desirable in some circum- whereas for females, Zvalues were 2.4 times stances to minimize the number of episodes as great for survivorship than for the esti- used. mated total number of eggs laid. Banks and Thompson (1985) draw a similar conclu- Short-term versus Lifetime sion; in their study of the damselfly Coena- Reproductive Success grion puella, differences in lifespan account- Our data confirm some of the difficulties ed for 70% of the variance in male mating of extrapolating lifetime data from short- success. Finally, McVey (1987), studying the term information. I values calculated from dragonfly Erythemis simplicicollis, found daily reproductive success data are consid- that variance in survivorship accounted for erably different from those calculated using 57% of the relative variance in estimated the lifetimekotal data (Table 1). Using only male lifetime reproductive success and 68% days when individuals came to the pond, of the relative variance in estimated female the opportunity for selection is 8 1% of the lifetime reproductive success. Taken to- lifetime value for males and only 19% of gether, these studies suggest that, for many the total value for females. I values calcu- species of odonates, variance in longevity lated by including days when individuals provides an important, if not the dominant, known to be alive did not come to Blom- contribution to the total opportunity for se- pond overestimated the lifetimehotal val- lection among both sexes, whether males ues. Estimates of the relative opportunity are territorial (as in the dragonflies Plath- for selection on males compared with fe- emis and Erythemis) or not (as in the dam- males are similarly misleading: using only selflies Enallagma and Coenagrion). days when individuals were present, the I,/ One consequence of this pattern is that I, ratio is 4 times greater than the lifetime/ variance in lifetime mating success will be total estimate of Z,/Z,. The Z,lZ, ratio derived largely due to variance in survivorship, and from using all days is, however, close to the thus mating success will not represent the lifetimehotal estimate (I = 1.95 vs. 2.37, opportunity for sexual selection (McCauley, Table 1). Data using all days comes closer 1983; Koenig and Albano, 1986; see below). to approximating lifetime data, as it as- Another consequence is that characters in- sumes knowledge of survivorship of marked SELECTION IN DRAGONFLIES 33 individuals even when they are not present time reproductive success (Fincke, 1987). at the pond. However, both measures of I Among vertebrates, Clutton-Brock (1983) using daily data suffer the additional diffi- reports an I,/I, ratio of 3.4 for the polygy- culty of non-independence among different nous red deer Cervus elaphus. However, an days of data taken from the same individ- exception to this trend is reported by McVey uals. (1 987) for E ythemis simplicicollis, in which Similar comparisons of daily and lifetime the I,/I, ratio was 0.94. Unfortunately, there data for the territorial dragonfly Erythemis are data from few monogamous species for simplicicollis are given by McVey (1 987). comparison; the data from kittiwakes Rissa For males, McVey's estimates of relative tridactyla, studied by Coulson and Thomas variance in male daily reproductive success (1985), yield an I,/I, ratio of 1.2, whereas exceed I values for lifetime reproductive Fitzpatrick and Woolfenden (1 987) indicate success (daily values ranged from 2.13 to that the I,/I, ratio for the Florida scrub jay 6.22 depending on the particular data set Aphelocoma coerulescens, is very close to used, lifetime values ranged from 1.29 to one. Thus, although the opportunity for se- 1.90). In contrast, relative variance in fe- lection may often be greater for males than male daily reproductive success was con- females in polygynous species and greater siderably lower than estimated lifetime val- in polygynous compared with monogamous ues (daily values were 0.44 and 0.83 species (see Clutton-Brock, 1983), McVey's [excluding days when females were not seen results indicate that this may not hold gen- and including such days, respectively], the erally. lifetime estimate was 1.35). Fincke (1986) similarly shows that daily data are not nec- Patterns of Selection essarily a good indication of the relative op- For female Plathemis, selection favored portunity for selection present on a lifetime individuals reproducing earlier in the sea- scale for Enallagma hageni. son; overall, the mean date females were Neither McVey's nor our data are as re- first on the pond figuratively decreased 3.9 1 liable for females as they are for males. days or 0.3 1 standard deviations (Table 6). Nevertheless, it clearly seems inadvisable Based on the 0'values, however, the only to infer lifetime patterns of opportunity for episode during which selection was signifi- selection, either within or between sexes, cant was the interclutch interval, which in- from short-term data (see also Clutton- creased later in the season. No significant Brock, 1983; Banks and Thompson, 1985). direct selection was detected on any of the In both our study and McVey's (1987), es- three characters measured for females (Ta- timates of the opportunity for selection ble 5). based on daily success better approximated Of the four characters measured for males, lifetime estimates when all days were in- only date of first reproduction is signifi- cluded, whether or not individuals actually cantly correlated with lifetime fitness (Table came to the pond. 4). A univariate analysis indicates that selection on this character is significant at two Intersexual Comparisons life. history stages; males living earlier had Our estimate of relative variance in life- higher survivorship and were present at the time/total reproductive success of male pond more hours per day than those living Plathemis lydia is 2.4 times greater than that later, again as indicated by the 0'values for females (Table I). I values for lifetime (Table 6). reproductive success of other polygynous The selection for seasonally earlier repro- species have usually, though not always, fol- duction is interesting in view of the seasonal lowed a similar pattern. Banks and Thomp- decline in body mass for both sexes (Table son (1985), studying mating success in the 4). These correlations suggest that the de- damselfly Coenagrion puella, report an I,/ cline in body mass may be the indirect result I, ratio of 2.1, while estimates of the I,/I, of selection to speed up development as the for the damselfly Enallagma hageni range season progresses so as to mitigate the lower between 2.9, based on lifetime mating suc- success of later individuals. This could be cess (Fincke, 1982), and 1.7, based on life- accomplished by a reduction in the variable 34 W. D. KOENIG AND S. S. ALBANO number of instars through which the larvae on the extent to which episodes occur in an develop (Benke and Benke, 1975). As we unambiguous chronological sequence. For found no selection for larger body size, this example, the three episodes of hatchability, trend would not be countered by selection followed by juvenile survivorship, followed for body mass per se. However, an alter- by adult reproductive success, follow each native hypothesis that mass declines late in other in chronological order with no tem- the season due to lower food availability poral overlap. This is not the case for the (Harvey and Corbet, 1985) cannot be re- episodes we use here. Although we arranged jected. episodes in a logical sequence, all occur si- None of the three non-behavioral char- multaneously during the lifetime of indi- acters has a significant direct effect on life- viduals, and other orderings are possible. In time fitness when controlling for the other order to circumvent this difficulty, we cal- two characters (Table 5). Although the low culated selection gradients for individual variance in the two morphological charac- episodes independent of the action of any ters undoubtedly is partly responsible, this other episodes (P,' values). lack of significance is also due to trade-offs between selective effects at different life-his- Mating Success and Sexual Selection tory stages. For example, males with shorter Several recent authors have equated vari- hind wings survive longer but have lower ance in mating success with the opportunity mating efficiency; the result is almost no for sexual selection (Wade and Arnold, 1980; overall selection on hind wing length (Table Fincke, 1982; Price, 1984; Arnold, 1983; 6). A similar trade-off is suggested for male Arnold and Wade, 1984b). Our results in- territorial flight; males that fly a greater dicate some of the shortcomings of this op- amount of time experience greater mating erational definition, especially when applied efficiency, but are on the pond a shorter to lifetime data (see also Banks and Thomp- period each day than less aggressive indi- son, 1985; Sutherland, 1985; Koenig and viduals (Table 6). Possibly males that fly Albano, 1986). From Table 2, for example, more contact more females while at the pond the proportion of the total opportunity for or are more successful at competing for selection on male lifetime reproductive suc- mates. The decreased duration of time more cess due to variance in mating success (I,,,,) active males spend at the pond follows from is very high (79%). However, to then con- the apparent energetic constraints of main- clude that the opportunity for sexual selec- taining a territory in Plathemis (Koenig and tion is large obscures the fact that the largest Albano, 1985). single contribution to the total Ivalue is the variance in survivorship-an episode of se- Measuring Episodic Selection lection which cannot be easily interpreted Here we have used three different selec- as sexual selection. tion gradients to measure the strength of This problem arises because mating suc- selection occurring at particular episodes cess is the collapsed result of several selec- (Table 6). The first two of these (P, and Pk* tive episodes (see Arnold and Wade, 1984a), values) are discussed in Arnold and Wade only some of which may represent Danvin- (1984a). P, values have the advantage of ian sexual selection (Koenig and Albano, being additive, but are not true selection 1986). In Plathemis, for example, variance gradients, as they are calculated using the in mating efficiency (w,) could result in part original variance rather than the adjusted from female choice or male-male compe- variance existing at the start ofeach episode. tition, and thus may represent sexual selec- p,* values are calculated using adjusted tion. Similarly, variance in hours present variances, and thus yield a more realistic per day on the pond (w,) may in part be due approximation of selection occurring at a to variance among males in territory-hold- specific episode (Kalisz, 1986). However, ing ability due to male-male competition; both these selection gradients are sensitive to this extent it may be appropriate to view to the ordering of individual episodes (Al- this episode as one of sexual selection. How- ban0 and Koenig, unpubl.). Thus, the con- ever, at least some of the variance in hours fidence that can be placed in them depends per day males spend at the pond, as well as SELECTION IN DRAGONFLIES 3 5

the proportion of days males come to the tion. We also thank J. Dickinson, 0.Fincke, pond (w,), is probably due to energetic con- M. McVey, K. Sullivan, J. Waage, E. Waltz, straints imposed by variance in male for- L. Wolf, P. Williams, and the reviewers for aging ability; hence, much or all of these their extensive comments and hel~on the two episodes may be more appropriately manuscript. This paper is dedicatedio Frank viewed as episodes of natural selection, as Pitelka in honor of his 70th birthday. is, of course, variance in survivorship (w,). Thus, the actual proportion of the opportunity for selection in lifetime reproductive ALCOCK,J. 1982. Post copulatory mate guarding by success for male Plathemis due to sexual males of the damselfly Hetaerina vulnerata Selys selection is probably somewhere between (Odonata: Calopterygidae). Anim. Behav. 30:99- 107. 11% (I,) and 17% (I3 + I,) rather than the ALTMANN,J. 1974. Observational study of behavior: 79% estimated from the total variance in Sampling methods. Behaviour 49:227-267. mating success (Il2,,). ARNOLD,S. J. 1983. Sexual selection: The interface For Plathemis, selection acting through of theory and empiricism, pp. 67-107. In P. P. G. both mating efficiency and time devoted to Bateson (ed.), Mate Choice. Cambridge Univ. Press, Cambridge, U.K. reproduction may be strong, and in some ARNOLD,S. J., AND M. J. WADE. 1984a. On the mea- cases may reflect sexual selection. For ex- surement of natural and sexual selection: Theory. ample, selection acting to increase time in Evolution 38:709-7 19. flight of territorial males through increased -. 1984b. On the measurement of natural and sexual selection: Applications. Evolution 38:720- mating efficiency seems particularly likely 734. to reflect male-male competition, and thus BANKS,M. J., AND D. J. THOMPSON.1985. Lifetime sexual selection. As shown in Table 6 and mating success in the damselfly Coenagrion puella. discussed above, such selection was coun- Anim. Behav. 33:1175-1183. tered by the tendency for less aggressive BELL,G. 1980. The costs of reproduction and their consequences. Amer. Natur. 116:45-76. males to spend a longer period of time per BENKE,A. C., AND S. S. BENKE. 1975. Comparative day on the pond, a result possibly due to a dynamics and life histories of coexisting dragonfly combination of natural and sexual selec- populations. Ecology 56:302-3 17. tion. Similarly, selection acting on male hind CAMPANELLA,P. J., AND L. I-. WOLF. 1974. Temporal leks as a mating system in a temperate zone dragon- wing length through increased mating effi- fly (Odonata: Anisoptera). I. Plathemis lydia (Dru- ciency may also be a consequence of direct ry). Behaviour 5 1:49-87. male-male competition for mates (hence CAVALLI-SFORZA,L. L., AND W. F. BODMER. 1961. sexual selection), and was countered by low- The Genetics of Human Populations. Freeman, San er survival (hence natural selection). Francisco, CA. CLUTTON-BROCK,T. H. 1983. Selection in relation Thus, trade-offs between natural and sex- to sex, pp. 457-481. In D. S. Bendell (ed.), Evo- ual selection via male-male competition are lution from Molecules to Man. Cambridge Univ. suggested by our data. Whether any of the Press, Cambridge, U.K. selective effects we document acting through -, editor. 1987. Reproductive Success. Univ. Chicago Press, Chicago, IL. increased mating efficiency result from fe- COULSON,J. C., AND C. THOMAS. 1985. Differences male choice, however, is unresolved with in the breeding performance of individual kittiwake the present analyses. Indeed, mate choice gulls, Rissa tridactyla (L.), pp. 489-503. In R. M. independent of resources (e.g., oviposition Sibly and R. H. Smith (eds.), Behavioural Ecology: sites) has yet to be documented in any odo- Ecological Consequences of Adaptive Behaviour. Blackwell, London, U.K. nate. EMLEN,S. T., AND L. W. ORING. 1977. Ecology, sexual selection and the evolution of mating systems. Science 197:215-223. We thank our field assistants, A. Peters FALCONER,D. S. 198 1. Introduction to Quantitative Genetics, 2nd Ed. Longman, London, U.K. and M. Peterson, for helping with the proj- FINCKE,0. M. 1982. Lifetime mating success in a ect. R. Noyce kindly gave us permission to natural population of the damselfly, Enallagma work at Blompond. 0. Fincke, S. Kalisz, hageni (Walsh) (Odonata: Coenagrionidae). Behav. and M. McVey kindly provided us with pre- Ecol. Sociobiol. 10:293-302. -. 1984. Sperm competition in the damselfly prints of papers; M. Wade helped us cal- Enalagma hageni Walsh (Odonata: Coenagrioni- culate the opportunity for selection; and S. dae): Benefits of multiple mating to males and fre- Arnold helped us understand sexual selec- males. Behav. Ecol. Sociobiol. 14:235-240. 36 W. D. KOENIG AND S. S. ALBANO

-. 1985. Alternative mate-finding tactics in a -. 1984. Dragonfly monopoly. Natur. Hist. 93: non-temtorial damselfly (Odonata: Coenagrioni- 32-39. dae). Anim. Behav. 33:1124-1137. MCVEY,M. E. 198 1. Lifetime reproductive tactics in -. 1986. Lifetime reproductive success and the a temtorial dragonfly Erythemis simplicicollis. Ph.D. opportunity for selection in a nontemtorial dam- Diss. Rockefeller Univ., N.Y. selfly (Odonata: Coenagrionidae). Evolution 40:79 1- -. 1984. Egg release rates with temperature and 803. body size in Libellulid dragonflies (Anisoptera). -. 1987. Sources of variation in lifetime repro- Odonatologica 13:377-385. ductive success in a nontemtorial damselfly (Odo- -. 1987. The opportunity for sexual selection nata: Coenagrionidae). In T. H. Clutton-Brock (ed.), in a territorial dragonfly Erythemis simplicicollis. Reproductive Success. Univ. Chicago Press, Chi- In T. H. Clutton-Brock(ed.), Reproductive Success. cago, IL. In press. Univ. Chicago Press, Chicago, IL. In press. FITZPATRICK,J. W., AND G. E. WOOLFENDEN.1987. MCVEY,M. E., AND B. J. SMITTLE.1984. Sperm pre- Components of lifetime reproductive success in the cedence in the dragonfly Erythemis simplicicollis. Florida scrub jay. In T. H. Clutton-Brock (ed.), J. Insect Physiol. 30:619-628. Reproductive Success. Univ. Chicago Press, Chi- PARKER,G. A. 1970. Sperm competition and its evo- cago, IL. In press. lutionary consequences in the insects. Biol. Rev. GRAFEN,A. 1982. HOWnot to measure inclusive fit- 45:525-567. ness. Nature 298:425-426. PARTRIDGE,L., AND T. HALLIDAY.1984. Mating pat- HARVEY,I. F., AND P. S. CORBET. 1985. Temtorial terns and mate choice, pp. 222-250. In J. R. Krebs behaviour of larvae enhances mating success of male and N. B. Davies (eds.), Behavioural Ecology: An dragonflies. Anim. Behav. 33:56 1-565. Evolutionary Approach. Sinauer, Sunderland, MA. HEINRICH,B., AND T. M. CASEY. 1978. Heat transfer PRICE,T. D. 1984. Sexual selection of body size, in dragonflies: 'Fliers' and 'perchers.' J. Exp. Biol. territory and plumage variables in a population of 74: 17-36. Darwin's finches. Evolution 38:327-34 1. JACOBS,M. E. 1955. Studies on sexual selection and SHERMAN,K. J. 1983. The adaptive significance of temtorialism in dragonflies. Ecology 36:566-586. postcopulatory mate guarding in a dragonfly, Pachi- KALISZ,S. 1986. Variable selection on the timing of diplax longipennis. Anim. Behav. 3 1: 1 107-1 1 15. germination in Collinsia verna (Scrophulariaceae). SUTHERLAND,W. J. 1985. Chance can produce a sex Evolution 40:479-49 1. difference in variance in mating success and explain KOENIG,W. D., AND S. S. ALBANO.1985. Patterns of Bateman's data. Anim. Behav. 33: 1349-1 352. territoriality and mating success in the white-tailed THORNHILL,R., AND J. ALCOCK.1983. The Evolution skimmer Plathemis lydia (Odonata: Anisoptera). of Insect Mating Systems. Harvard Univ. Press, Amer. Midl. Natur. 114: 1-12. Cambridge, MA. -. 1986. On the measurement of sexual selec- WAAGE,J. K. 1979. Dual function of the damselfly tion. Amer. Natur. 127:403-409. penis: Sperm removal and transfer. Science 203: LANDE,R., AND S. J. ARNOLD. 1983. The measure- 916-918. ment of selection on correlated characters. Evolu- -. 1983. Sperm competition and the evolution tion 37:1210-1226. of odonate mating systems, pp. 25 1-290. In R. L. MAY,M. L. 1976. Thermoregulation and adaptation Smith (ed.), Sperm Competition and the Evolution to temperature in dragonflies (Odonata: Anisop- of Animal Mating Systems. Academic Press, N.Y. tera). Ecol. Monogr. 46: 1-32. WADE,M. J., AND S. J. ARNOLD. 1980. The intensity MCCAULEY,D. W. 1983. An estimate of the relative of sexual selection in relation to male sexual be- opportunities for natural and sexual selection in a havior, female choice, and sperm precedence. Anim. population of milkweed beetles. Evolution 37:70 1- Behav. 28:446-46 1. 707. MCMILLAN,V. 1983. Mate guarding in Plathemis lydia. Ph.D. Diss. Syracuse University, Syracuse, N.Y. Corresponding Editor: D. W. Schemske