Fitness costs of female choosiness in a socially monogamous songbird

Female mate choice is thought to be responsible for the evolution of many extravagant male ornaments and displays, but the costs of being too selective may hinder the evolution of choosiness. Selection against choosiness should be strongest in socially monogamous mating systems, because females may end up without a partner and forego reproduction, especially when many females prefer the same few partners (frequency-dependent selection). Here we quantify the fitness costs of having mating preferences that are difficult to satisfy. We capitalise on the recent discovery that female zebra finches (Taeniopygia guttata) prefer males of familiar song dialect. We measured female fitness in captive breeding colonies in which one third of females were given ample opportunity to choose a mate of their preferred dialect (two thirds of all males; ‘relaxed competition’), while two thirds of the females had to compete over a limited pool of mates they preferred (one third of all males; ‘high competition’). As expected, social pairings were strongly assortative with regard to song dialect. In the high-competition group, 26% of the females remained unpaired, yet they still obtained relatively high fitness by using brood parasitism as an alternative reproductive tactic. Another 31% of high-competition females paired disassortatively for song dialect. These females showed increased levels of extra-pair paternity, mostly with same-dialect males as sires, suggesting that preferences were not abolished after social pairing. However, females that paired disassortatively for song dialect did not have lower reproductive success. Overall, females in the high-competition group reached equal fitness as those that experienced relaxed competition. Our study suggests that alternative reproductive tactics such as egg dumping can help overcome the frequency-dependent costs of being highly selective in a monogamous mating system, thereby facilitating the evolution of female choosiness.


Introduction
The degree of assortative mating, i.e. the proportion of assortative pairs, can range from zero to one 178 (see Figure 3). In our experimental setup, a value of zero can theoretically be reached if all pairs 179 mated disassortatively (zero assortative and up to 12 disassortative pairs in each aviary). A value of 180 one, corresponding to perfect assortative pairing, can only be reached if four females per aviary 181 remained unpaired (8 assortative and zero disassortative pairs per aviary). Under random pairing, 182 44.4% of pairings should be assortative (1/3 of females has a 2/3 chance of pairing assortatively, plus 183 2/3 of females has a 1/3 probability; 1/3 x 2/3 + 2/3 x 1/3 = 4/9). If all females would attempt to pair 184 assortatively, but no female would forego pairing, 66.7% of pairings should be assortative (8 185 assortative and 4 disassortative pairs per aviary). 186 Of the 106 social pairs that were observed, 72 (67.9%) were assortative. This significantly deviates 187 from the random expectation of 44.4% (exact goodness-of-fit test p < 0.0001). Considering the 188 number of eggs in the nests of those pairs (N = 1,022 in total), 730 eggs (71.4%) were cared for by 189 assortative pairs. At the genetic level, out of the 1,074 eggs fertilized and genotyped, 783 (72.9%) 190 had parents that mated assortatively for song dialect. Hence, both at the social and genetic level, we 191 found strong assortative mating, slightly exceeding the 66.7% "assortative if possible" threshold 192 ( Figure 3). 193 Females from the two treatment groups differed significantly in their pairing success, with only 4 194 females (10%) from the relaxed-competition treatment remaining unpaired, but 21 females (26%) 195 from the high-competition treatment not observed in a social bond (p = 0.03, Figure 4A, B, Table 1, 196 model 4, S4 Table). In the relaxed-competition group, 87.5% of females (N = 35) mated assortatively 197 with a male from their natal dialect, and one female (2.5%) was observed in two pair bonds (one 198 assortative, one disassortative; "mixed" in Figure 4A). In contrast, in the high-competition group, 199 only 37.5% of females (N = 30) mated exclusively assortatively, 5% (N = 4 females) participated in 200 both types of pairing, and 31% (N = 25 females) mated exclusively disassortatively ( Figure 4B, Table  201 1, models 5-7, S5 Table, S6 Table, S7 Table). 202 Females from the high-competition group took longer to start a social bond compared to females 203 from the relaxed-competition group (p = 0.008, Figure 4C, D, Table 1, model 8, S8 Table). For this 204 test, we assigned a maximum latency of 75 days to unpaired females (as in the pre-registered model 205 3), because we cannot exclude that such females would have paired after a longer period. If 206 unpaired females are excluded from the analysis, the difference between treatment groups in 207 latency to pair is no longer significant (t = 1.24, p = 0.22). Hence, the treatment prevented or delayed 208 social pairing (Table 1, models 4 and 8, S4 Table, S8 Table), but it did not prevent or delay egg-laying 209 (S2 Table, S3 Table). 210 In 18 cases, females attempted to rear offspring as single mothers (14 females attended one clutch 211 and two females each attended two consecutive clutches). Of those, 11 clutches (61%) were reared 212 by females that remained unpaired until the end of the experiment (overall, 25 out of 120 females 213 remained unpaired until the end, 21%). However, the average number of clutches attended to as 214 unpaired female did not differ significantly between the treatment groups (Table 1, model 9, S9  215   Table). Females from the high-competition group on average laid fewer eggs that they actively took 216 care of, although this was not significant (p = 0.21, Table 1, model 10, S10 Table). However, females 217 from the high-competition group laid significantly more eggs into clutches that were cared for by 218 other females (egg dumping in the strict sense; p = 0.038, Table 1, model 11, S11 Table) and into 219 nests of other females, nests attended by single males, or into unattended nest boxes (egg dumping 220 in the wide sense; p = 0.009, Table 1, model 12, S12 Table). Hence the proportion of parasitic eggs 221 among the total number of eggs laid was markedly higher in the high-competition than in the 222 relaxed-competition group (Table 1, models 13 and 14, S13 Table, S14 Table). 223 Splitting the females of each treatment group into subsets according to their social pairing status 224 ( Figure 4) shows that the parasitic egg-dumping tactic was used more often by the unpaired females 225 of the high-competition group (compared to relaxed-competition group; t-test with unequal 226 variances, t26.4 = 3.37, p = 0.002), followed by the disassortatively mated females of the high-227 competition group (t33.4 = 2.09, p = 0.044; Figure 5B). Overall, females from the high-competition 228 group achieved similar fitness as the females from the relaxed-competition group ( Figure 5A), 229 because rearing success (the proportion of fertile eggs that became independent offspring) did not 230 differ between the treatment groups (Table 1, model 15, S15 Table). 231 Finally, we examined levels of extra-pair paternity in the different treatment groups, focusing on the 232 84 females that were socially paired to only one partner. As expected, extra-pair paternity was more 233 frequent in the disassortatively paired females from the high-competition group (44%, 81 out of 183 234 eggs from 21 females) than in the assortatively paired females from the same treatment group (18%, 235 42 of 252 eggs from 27 females; t-test based on proportions for each female: t46 = 3.2, p = 0.002; 236 Figure 6). Assortatively paired females from the relaxed-competition group showed intermediate 237 levels of extra-pair paternity (36%, 112 of 312 eggs from 35 females; for additional details see S16 238 Table). In each of the three groups, the majority of extra-pair eggs were sired assortatively (70%, 239 65%, and 89% respectively) and all three numbers clearly exceed the corresponding random 240 expectations (36%, 27%, and 64%, respectively) calculated from the number of potential extra-pair 241 males in the aviary (4, 3, and 7 out of 11, respectively; see also S16 Table). 242

Discussion 244
Our study illustrates the importance of empirically quantifying the costs and benefits of choosiness 245 to predict selection on the level of choosiness. This can then inform about the expected intensity of 246 sexual selection through female choice. A recent theoretical study highlighted that choosiness in 247 monogamous systems may have high costs and hence will be selected against [27]. Based on this 248 study, we hypothesized that females in the high-competition group would suffer substantial fitness 249 costs compared to the relaxed-competition group (https://osf.io/8md3h). However, our empirical 250 findings strongly suggest that female zebra finches have evolved sufficient behavioural flexibility to 251 cope with the challenge of having preferences that are difficult to satisfy such that they did not 252 suffer lower fitness. This flexibility is not trivial, because zebra finches that were force-paired 253 suffered significant fitness costs compared to birds that were allowed to choose their mate [25]. 254 Females in the high-competition treatment on average achieved slightly higher relative fitness 255 compared to those in the relaxed-competition group. Thus, the best estimate for the fitness cost of 256 choosiness in our study equals zero. However, when considering reduced pairing success, delayed 257 pairing and reliance on conditional parasitism, one could argue that the biologically most likely 258 fitness cost is small, but positive. Females that relied on the parasitic tactic of egg dumping [31-33] 259 were surprisingly successful in terms of fitness ( Figure 5). However, our models on the use of this 260 tactic also suggest that this may be a form of 'making the best of a bad job', because the proportion 261 of a female's eggs that was dumped (in the wide sense) rather than actively cared for was higher in 262 the high-competition treatment group, and also increased with the inbreeding coefficient of the 263 female (p = 0.002, S14 Table). These results suggest that the parasitic tactic is associated with poor 264 pairing success and with poor female condition. Hence, overall, there likely is a small net cost of 265 having preferences that are hard to satisfy, but quantifying such a small cost is difficult because of 266 sampling noise. 267 Our study suggests that an alternative reproductive tactic, namely egg dumping, may be important 268 to consider as a mechanism that effectively reduces the costs of choosiness and thereby favors the 269 evolution of choosiness even in monogamous mating systems. Alternative reproductive tactics can 270 thereby increase the intensity of sexual selection through female choice. Note that our analysis of 271 egg dumping is part of the post hoc data exploration rather than pre-registered hypothesis testing, 272 which implies that the probability that this result is a chance finding is higher ( Figure 5). 273 Nevertheless, we avoided extensive exploratory testing combined with selective reporting and post-274 hoc modification of analysis strategy to minimize the risks of false positive findings [49]. Accordingly, 275 This study also contributes to our understanding of zebra finch mating preferences with regard to 280 song dialects. Firstly, we confirmed that such preferences exist and that they are sufficiently strong 281 to result in a high degree of assortment even when bringing together unequal numbers of males and 282 females of each dialect. Secondly, assortative mating was present both at the genetic level (72.9% of 283 fertilized eggs had assortative genetic parents) and at the social level (71.4% of eggs were cared for 284 by assortative pairs). As 78.2% of all eggs sired by extra-pair males were assortative, we infer that 285 song dialect preferences affect both social pairing and extra-pair mate choice. 286 The possible adaptive function of these preferences in the wild is not known. They could function to 287 enhance local mating to obtain locally adapted genes, which would be adaptive in both the social 288 and extra-pair context. However, this possibility seems unlikely in light of the lack of genetic 289 differentiation even over a large geographic distance [50]. More widespread sampling of genotypes 290 throughout Australia would be required to rule out this possibility. Alternatively, song preferences 291 could function to find a mate that hatched locally [51,52]  Our study was designed to estimate the costs of female choosiness, whereby we predicted that 297 these costs would be high in a monogamous mating system [27]. However, this is not what we 298 found. Our study did not fail in the sense that our treatment was effective (Figs. 2 and 3), but 299 females were sufficiently flexible to not suffer substantial fitness costs under high competition for 300 preferred males, at least in our aviary setting. Thus, our study does not support the hypothesis that 301 female choosiness is costly in a socially monogamous system. Females from the high-competition 302 treatment were affected in terms of delayed pairing and reduced pairing success, but they made up 303 for this primarily by the alternative reproductive tactic of egg dumping and only rarely by caring for 304 clutches as a single mother. Females who ended up paired with a non-preferred partner, were more 305 likely to engage in extra-pair copulations, but this did not affect their fitness (see Fig. 4). This stands 306 in contrast to an earlier study that showed that force-paired females responded more negatively to in a significant reduction in fitness [25]. Overall, these results emphasize that models on the costs of 309 choosiness need to be informed by empirical research. 310

Methods 311
All methods closely adhere to the pre-registration document (https://osf.io/8md3h), except for the 312 exploratory post-hoc analyses (presented below). 313 Ethics 314 The study was carried out under the housing and breeding permit no. 311.4-si (by Landratsamt 315 Starnberg, Germany) which covers all procedures implemented (including the obtaining of one blood 316 sample per individual for parentage assignment). 317 Background of study populations and assortative mating 318 The zebra finches used in this study originate from four captive populations maintained at the Max 319 Planck Institute for Ornithology: two domesticated (referred to as 'Seewiesen' (S) and 'Krakow' (K)) 320 and two recently wild-derived populations ('Bielefeld' (B) and 'Melbourne' (M)). For more 321 background and general housing conditions see [25,53,54]. The four populations have been 322 maintained in separate aviaries (without visual and with limited auditory contact). When birds from 323 two different populations (combining S with B, and K with M) were brought together in the same 324 breeding aviary, they formed social pairs that were predominantly assortative with regard to 325 population (87% assortative pairs), despite the fact that opposite-sex individuals were unfamiliar 326 with each other ([48], see also [55]). To find out whether this assortative mating took place because 327 of genetic (e.g. body size) or cultural (e.g. song) differences, we produced an offspring generation 328 ('F1') in which half of the birds were cross-fostered between populations (between S and B, or 329 between K and M) and half of the birds were cross-fostered within populations. For this purpose, we 330 used 16 aviaries (4 per population), each containing 8 males and 8 females of the same population 331 that were allowed to freely form pairs and breed. Cross-fostering was carried out at the aviary level, 332 such that 2 aviaries per population served for cross-fostering within population and the other 2 for 333 between-population cross-fostering. This resulted in 8 cultural lines (4 populations x 2 song dialects), 334 each maintained in 2 separate aviaries (16 aviaries). When unfamiliar individuals of the two song 335 dialects were brought together in equal numbers (50:50 sex ratio), they mated assortatively 336 regarding song (79% assortative pairs; Wang et al. 2020) but not regarding genetic population. To 337 disentangle the song effect of interest from possible side-effects of the cross-fostering per se, the 8 338 lines were bred for one more generation ('F2'). These F2 individuals are the focal subjects of this 339 study. Breeding took place in 16 aviaries (2 per song dialect within population), but without cross-340 fostering. The two replicate aviaries of each song dialect line each contained 8 males and 8 females 341 that produced the next generation. A subset of the resulting offspring (n = 144, not used in this 342 study, but see below) were used to test mate choice within each of the 4 genetic populations (here 343 referred to as 'F2 pilot experiment'). Again, we observed assortative pairing for song dialect (73% 344 assortative pairs). The remaining F2 offspring were used as candidates for the experiment, as 345 explained below. 346

Experimental setup 347
To quantify the female fitness consequences of having preferences for males that are either rare or 348 overabundant, we used 10 aviaries (3 for populations B and K, and 2 for populations S and M). Each 349 semi-outdoor aviary (measuring 4m x 5m x 2.5m) contained 12 males and 12 females of the same 350 genetic population, but from two different song dialects such that 4 females encountered 8 males of 351 the same dialect, while the remaining 8 females encountered only 4 males of their own dialect. 352 Hence, for each experimental aviary, we used individuals that were raised in 4 separate aviaries (2 of 353 each song dialect) to ensure that opposite-sex individuals were unfamiliar to each other. 354 The allocation of birds to the aviaries followed two principles. First, we listed for each of the 16 355 rearing aviaries the number of available female and male F2 offspring that had not been used 356 previously (in the 'F2 pilot experiment') and that were apparently healthy (374 birds). Depending on 357 the number of available birds, each rearing aviary was then designated to provide either 4 or 8 birds 358 of either sex, such that the total number of experimental breeding aviaries that could be set up was 359 maximized (10 aviaries). Second, the allocation of the available individuals within each rearing aviary 360 to the designated groups of 4 or 8 individuals of a given sex was decided by Excel-generated random 361 numbers. For instance, if a given rearing aviary had 17 candidate female offspring, individuals were 362 randomly allocated to a group of 4 for one experimental aviary, a group of 8 for another aviary, and 363 a group of 5 as leftover (not used). This allocation procedure may have introduced a bias, because 364 rearing aviaries that were highly productive (had more offspring) were more frequently designated 365 to send groups of 8 offspring to an experimental aviary, while those that produced fewer offspring 366 (in the extreme case fewer than 8 of one sex) were more likely used to send a group of 4 offspring to 367 an experimental aviary. This might bias our fitness estimates if offspring production was partly 368 heritable or if housing density prior to the experiment influenced the fitness in the experimental 369 aviaries. We therefore assessed these potential biases in the statistical analysis (see model 1b 370 below). 371 After allocating individuals to the 10 experimental aviaries, one female (designated for aviary 2) and 372 two males (designated for aviary 3) died before the start of the experiment. These individuals were then replaced by randomly choosing individuals of the same sex and rearing aviary, which however 374 had previously taken part in the 'F2 pilot experiment' (17-30 January 2019). These replacement birds 375 differed from the other individuals in the experiment, in that they had previous experience of nest-376 building and egg-laying >100 days before the start of experiment. 377 The 120 focal females had hatched in one of the 16 natal aviaries between 30 May and 25 378 September 2018 and remained in their natal aviaries initially together with their parents (which 379 were removed between 10 December 2018 and 16 January 2019). On 6 May 2019, all individuals 380 used in the experiment were transferred to the 10 aviaries, whereby the 12 males and 12 females in 381 each aviary were separated by an opaque divider. After one week, the divider was removed and the 382 experiment started. At this time, females were on average 313 days old (range: 230 -348 days). To 383 facilitate individual identification, each of the 12 males and 12 females within each aviary was 384 randomly assigned two coloured leg bands (using the following 12 combinations: blue-blue, black-385 black, orange-orange, orange-black, red-red, red-blue, red-black, white-white, white-black, white-386 orange, yellow-yellow, yellow-blue). 387

Breeding procedures 388
Each of the 10 experimental aviaries was equipped with 14 nest boxes. All nest boxes were checked 389 daily during week days (Mon-Fri) for the presence of eggs or offspring. Eggs and offspring were 390 individually marked and a note was made whether eggs were warm. For eggs laid on weekends, we 391 estimated the most likely laying date based on egg development. We collected a DNA sample from 392 all fertilized eggs (including naturally died embryos and nestlings), unless they disappeared before 393 sampling (see below) to determine parentage. Eggs containing naturally died embryos (N = 343) 394 were collected and replaced by plastic dummy eggs (on average 12 ± 4 (SD) days after laying and 7 ± 395 4 days after estimated embryo death). Eggs that remained cold (unincubated) for 10 days (N = 7 out 396 of 1,399 eggs) were removed without replacement and were incubated artificially to identify 397 parentage from embryonic tissue. During nest checks we noted the identity of the parent(s) that 398 attended the nest (based on colour bands) to clarify nest ownership for all clutches that were 399 incubated. 400 As the main response variable, we quantified the reproductive success ('fitness') of each female in 401 each experimental aviary as the total number of genetic offspring produced that reached the age of 402 35 days (typical age of independence). All eggs laid within a period of 70 days (between 13 May and 403 22 July 2019; N = 1,399) were allowed to be reared to independence; eggs laid after this period were 404 thrown away and replaced by plastic dummy eggs to terminate a breeding episode without too 405 much disturbance. 406 Out of 1,399 eggs, 319 eggs failed (180 appeared infertile, 101 disappeared, 30 broke, 4 had 407 insufficient DNA, 3 eggs showed only paternal alleles (androgenesis), 1 sample was lost). For the 408 remaining 1,080 eggs we unambiguously assigned maternity and paternity based on 15 409 microsatellite markers (for details see Wang et al. 2017). Of these 1,080 fertile eggs, 750 developed 410 into nestlings and 556 into offspring that reached 35 days of age. 411 The 1,399 eggs were distributed over 289 clutches (allowing for laying gaps of maximally 4 days), of 412 which 190 (1,022 eggs) were attended by a heterosexual pair (involving 106 unique pairs), 55 (120 413 eggs) remained unattended, 24 (120 eggs) were attended by a single female, 12 (41 eggs) were 414 attended by a single male, 3 (36 eggs) were attended by a female-female pair, 2 (30 eggs) were 415 attended by two males and two females, 2 (21 eggs) were attended by a trio with two females, and 1 416 (9 eggs) by a trio with two males. 417 We also quantified two additional response variables for every female, namely the latency to start 418 laying eggs (in days since the start of the experiment, counting to the first recorded fertile egg, and 419 ascribing a latency of 75 days to females without fertile eggs) and the total number of fertile eggs 420 laid within the 70-day experimental period (both based on the 1,080 eggs with genetically confirmed 421 maternity). The 319 failed eggs were not considered. 422 Over the course of the experiment (13 May to 22 July 22 for egg laying and until 9 September for 423 rearing young to independence) one male and two females (all of the more abundant type within 424 their aviaries) died of natural causes (a male in aviary 5 on 28 June, a female in aviary 6 on 24 July, 425 and a female in aviary 10 on 22 August). Thus, following the preregistered protocol, no bird was 426 excluded from data analysis. 427

Data analysis 428
Following previously used methods [25, 56], we calculated 'relative fitness' for each female as 429 relative fitness of female i = N * number of offspring of female i / total number of offspring of all N 430 females in the aviary. This index has a mean of 1 for each aviary, and accounts for fitness differences 431 between the 4 genetic populations (note that all birds within an aviary come from the same genetic 432 population). Latency to egg laying was log10-transformed before analysis to approach normality. To 433 control for the effect of inbreeding on fitness, we calculated female inbreeding coefficients F from 434 existing genetic pedigree data (using the R package 'pedigree' V.  Table 1 lists all the statistical models that compare the two treatment groups. These comprise both 438 pre-registered models (1-3) and post-hoc exploratory models (4-15). All models have the same basic 439 structure comparing a fitness-related trait between the two treatment groups (120 rows of data 440 representing 80 high-competition and 40 relaxed-competition females). Thus, we used mixed-effect 441 models with Gaussian (models 1-12) or binomial (models 13-15) errors, with the fitness-related trait 442 as the dependent variable, with treatment as the fixed effect of interest, with a scaled inbreeding 443 coefficient of the female as a covariate, and with the experimental aviary (10 levels) and the natal 444 aviary (16 levels) as random effects. 445 For preregistered model 1 we ran two versions (1a and 1b). Because the dependent variable of this 446 model is relative fitness, which was scaled within experimental aviaries, the model was designed 447 without random effects (as a general linear model, 1a). To control for possible influences of the natal 448 environment and of the genetic F1 mother we added two fixed-effect covariates (in version 1b): (1)  449 the total number of F2 offspring in the natal aviary where the focal female was raised (ranging from 450 29 to 45 offspring across the 16 natal aviaries), (2) the number of independent F2 offspring produced 451 by the genetic mother (one year earlier, also within a 70-day window for egg-laying; mean: 5.7, 452 range: 0-12, N = 66 mothers of the 120 focal females). 453

Exploratory analyses 454
To quantify the extent of assortative mating with regard to song dialect at the behavioural level, we 455 relied on the 106 unique heterosexual pairs that were observed caring for at least one of 190 456 clutches (comprising 1,022 eggs). For the quantification of assortment on the genetic level, we relied 457 on the genetic parentage of the 1,080 successfully genotyped eggs, of which 6 eggs had to be 458 excluded because they were sired by males from the females' natal aviaries (due to sperm storage, N 459 = 4), because alleles from two males were detected (presumably due to polyspermy, N = 1), or 460 because no paternal alleles were detected (possible case of parthenogenesis, N = 1), leaving 1,074 461 informative eggs. 462 For each female, we scored their social pairing behaviour, i.e. we noted whether they had been 463 recorded as a member of one of the 106 heterosexual pairs engaging in brood care. We quantified 464 (a) the total number of social bonds (0, 1, or 2), (b) the number of assortative and disassortative 465 bonds and (c) the latency to their first social bond (i.e. the laying date of the first egg in a clutch they 466 attended as one of the 106 pairs, relative to the start of the experiment; ascribing a latency of 75 467 days to females with zero social bonds). Latency was log10-transformed before analysis.
For each female, we also counted the number of clutches (0, 1, or 2) attended as a single mother 469 and we quantified (a) the number of eggs (out of the 1,080 genetically assigned eggs) they actively 470 cared for themselves (in whatever social constellation), (b) the number of eggs dumped into nests 471 attended by other females (in whatever social constellation, "egg dumping in the strict sense") and 472 (c) the number of eggs dumped anywhere ("egg dumping in the wide sense", including in nests 473 attended by single males and in unattended nest boxes). All exploratory mixed-effect models (4-15) 474 closely follow the design of the preregistered models 2 and 3 (see above). 475 Models 13-15 deal with proportions of eggs, and hence we used binomial models with counts of 476 successes and failures, and controlling for overdispersion by fitting female identity (120 levels) as 477 another random effect. 478 We ran exploratory analyses on the levels of extra-pair paternity of 84 females that were socially 479 paired to only a single male (i.e. recorded with only one male, among the 106 nest-attending 480 heterosexual pairs). These 84 females produced a total of 795 eggs with parentage information. 481 However, we excluded 48 eggs (from 16 females) that were laid before the date of pairing of the 482 focal female (genetic mother). Overall, 239 of the remaining 747 eggs (32%) were sired by a male 483 that was not the social partner (the male with whom the female attended a nest), so these are 484 classified as 'extra-pair sired'. We calculated levels of extra-pair paternity for three groups of 485 females: (1) assortatively paired females of the relaxed-competition group (n = 35 females), (2) 486 assortatively paired females of the high-competition group (n = 28 females, one of which did not lay 487 any eggs with parentage information), (3) disassortatively paired females of the high-competition 488 group (n = 21 females). To compare levels of extra-pair paternity between the latter two groups of 489  Table 1. Comparisons between females of the 'high-competition' (n = 80) and 'relaxed competition' 637 (n = 40) treatment. Overview of planned tests (models 1-3, as outlined in the pre-registration 638 document before data collection; https://osf.io/8md3h) and post-hoc tests that were conducted 639 after knowing the results of the planned tests (data exploration, models 4-15). All conducted tests 640 are reported in their initial form (no selective reporting, no post-hoc modification). Indicated are 641 average values for the two treatment groups for each dependent variable. Proportions of eggs refer 642 to means of individual mean proportions. For latencies, back-transformed values after averaging 643 log10-transformed values are shown. P-values refer to group differences based on glms or glmms. 644 Covariates are the female's inbreeding coefficient (F), the size of the peer group in the female's natal 645 aviary (peersize), and the fitness of the female's mother. Random effects are the experimental aviary 646 (exp AV, 10 levels), the female's natal aviary (natal AV, 16 levels), and -in binomial models of counts 647 with overdispersion -female identity (FID, 120 levels) (see S1-S15 Tables for details). Note that the 648 high significance of the treatment effect in models 6 and 7 is partly caused by the experimental 649 design. 650  limited by the availability of preferred mates (red, high competition) compared to females that are 652 not limited by their choosiness or by the availability of preferred mates (blue, relaxed competition). 653 For simplicity, we assume that preferred and non-preferred mates do not differentially affect female 654 fitness. Diamonds illustrate variation in individual fitness around the mean fitness of a group of 655 females (centre of diamonds, horizontal lines). Black arrows stand for various aspects of costs of 656 choosiness under competition for mates. The grey arrow represents fitness gains via alternative 657 reproductive tactics of females that remain socially unpaired, including reproduction as single 658 female or via parasitic egg dumping. The cost of socially pairing with a non-preferred male may e.g. 659 result from reduced willingness to copulate leading to infertility, aggression, and reduced male 660 brood care. Note that in empirical studies the apparent costs of pairing with a non-preferred male 661 (cost of "dissatisfaction") and of remaining unpaired might be confounded by effects of intrinsic 662 quality differences between the three groups of females shown in red. Also note that all choosy 663 females (red) pay a cost of competition, which might also vary between groups, for example if some 664 females avoid to compete. charts showing the proportion of females in each of the two treatment groups that were either not 688 observed as a pair (unpaired), or were seen in assortative, disassortative or both type of pair bonds 689 (mixed). Numbers indicate the count of females in each group. (C, D) Histograms illustrating the 690 temporal patterns of emergence of social bonds (either assortative or disassortative). Shown is the 691 day after the start of the experiment (potentially ranging from 1 to 70) on which the first egg was 692 recorded in a nest taken care of by one of the 106 breeding pairs (note that this may include 693 parasitic eggs not laid by the focal female). The y-axes are scaled to compensate for the two-times 694 larger number of females in the high-competition treatment relative to the relaxed-competition 695 treatment. Note that assortative bonds (N = 72) formed significantly earlier than disassortative 696 bonds (N = 34; back-transformed estimates 9.3 vs. 17.8 days, t-test on log-transformed latency: t104 = 697 3.67, p = 0.0004). 698 699 Figure 5. Relative fitness (A, as described in Figure 2) and number of "dumped eggs" laid (B, wide  701 definition of parasitic eggs, see Results) for females of different pairing status (unpaired, or mated 702 assortatively, disassortatively or both, as in Figure 4A, B) in each of the two treatment groups 703 (relaxed competition versus high competition). Horizontal lines indicate group averages. 704 705 706 Figure 6. Proportion of eggs sired outside the monogamous pair bond (extra-pair paternity, EPP, 708 grey bars) versus within-pair paternity (WPP, white bars) for three groups of females with a single 709 social pair bond. These are (1) assortatively paired females (n = 35) from the relaxed competition 710 treatment (blue), (2) assortatively paired females (n = 28, one of which did not lay any eggs) from 711 the high competition treatment (red), and disassortatively paired females (n = 21) from the high 712 competition treatment. For each category of eggs, we indicate the proportion that is sired 713 assortatively ('assort') for song dialect and in parentheses the random expectations ('exp') for this 714 proportion of assortative mating based on the number of available extra-pair males of each song 715 dialect. For more details see also S16 Table.