In the field experiments, 65 births were observed in 81 recaptured females (80% pregnancy rate). A total of 11 litters (17% of the births) were born late. In the full model, there was a tendency for an interaction effect of population composition with male turnover treatment on the probability of late births (turnover treatment: Χ
2 = 0.0, df = 1, p = 0.93, composition: Χ
2 = 6.2, df = 2, p = 0.045, interaction Χ
2 = 4.8, p = 0.093, Fig. 2, Table 2, for post hoc tests see respective H1 and H2 below).
In the experiment on captive pairs a total of 146 litters were born to 204 females (72% pregnancy proportion), of which 26% (38) were born late. Both male turnover treatment and cohort explained late litters rates, without interaction (turnover treatment: Χ
2 = 7.1, df = 1, p = 0.008, cohort: Χ
2 = 10.1, df = 2, p = 0.008, interaction: Χ
2 = 1.0, p = 0.56, Fig. 2; Table 2, discussed at respective hypotheses H1 and H4 below).
H1: The Bruce effect in single-male–single-female breeders [proportion of late litters: MF > MFFF and MF > MMMFFF)]
After a turnover of the breeding male, females in MF conditions had a higher proportion of late litters compared to females in multi-female conditions (simple-effect tests within treatment levels: within the replaced male treatment, the proportion of late litters was associated to population composition (Table 3): MF females produced a higher proportion of late litters (45%, 5 out of 11) than MFFF females (6%, 1/17, Fishers exact: p = 0.018) and tended to produce a higher proportion of late litters than MMMFFF females (0%, 0/7, p = 0.10). MFFF and MMMFFF females did not differ in the proportion of late litters (6 and 0%, p = 1.00). In the data set restricted to nulliparous females these results were confirmed: after male replacement the proportion of late litters in MF females (45%, 5/11) was higher than from (MFFF + MMMFFF) females (6%, 2/28, p = 0.018; all other within-factor comparisons p > 0.36). Dip tests indicated non-unimodal distribution of births in replaced male treatments (Fig. 3a, c), but while there were two modes (early births and late births) in the MF treatment (3c), there were two distinct modes of early births in the group treatments.
Within the returned male treatment in the full data set the proportion of late births (19%, 5/29) was not associated to population composition (p = 0.58). Within none of the population compositions, the proportion of late litters was associated with male turnover treatments (returned and replaced treatments in MF pairs: 20 and 45%, in MFFF: 8 and 6%, in MMMFFF: 0 and 29%, respective absolute numbers in Table 2, all p > 0.36). Dip tests in the returned treatment (Fig. 3b, d) did not support other than one (early birth) mode.
In the experiment on captive pairs, male replacement resulted in a higher probability of late births (30 late out of 86 births = 35%) compared to if the male was returned (8/60 = 13%, Fig. 2).
H2: The Bruce effect as an alternative to paternity confusion (proportion of late births: MFFF > MMMFFF)
Between the multi-female group experiments, there was no effect of single male versus multiple males on the proportion of late births (4 late out of 44 total, 9%, Figs. 2 and 3), and we found no interactive effect of male turnover and population composition (GLMM, number of males: Χ
2 = 0.0, df = 1, p = 0.92; male turnover treatment; Χ
2 = 0.0, df = 1, p = 0.83, female age: Χ
2 = 0.1, df = 1, p = 0.74; female breeding status: Χ
2 = 0.3, df = 1, p = 0.56, Table 2).
H3: Absolute animal density affects the Bruce effect (proportion of late births; in large enclosures: MMMFFF > MFFF > MF) and (MF small > MF large)
Density did not affect the proportion of late births (Χ
2 = 1.7, df = 2, p = 0.42, n = 54 females in large enclosures). Within the MF experiment we had used two enclosure sizes (5 births from small/5 from large in replace, 7/4 in return treatment). Pregnancy rate was 75% in both enclosure sizes, and the proportion of late births was 42% in smaller enclosures and 22% in larger enclosures. An interaction of enclosure size and turnover treatment did not explain the probability of late births to occur (interaction: Χ
2 < 0.4, p > 0.50).
H4: Cohort effects on proportion of late births (OW-n > YY-n > YY-p)
In the pairing experiment on captive bank voles, the proportion of late births was affected by both male turnover treatment and female cohort (Fig. 2; Table 2). OW-n and YY-n females did not differ in the proportion of late births (32 and 31%, Χ
2 = 0.0, df = 1, p = 1.00), but YY-p females had a lower proportion of late births than the other cohorts (4%, Χ
2 = 7.3, p = 0.007 and Χ
2 = 6.4, p = 0.012). Although we could not detect a cohort-specific treatment effect in the statistical model (non-significant interaction effect), we found that the distribution of birth dates deviated from unimodality only for OW-n females in the replaced male treatment. (D = 0.11, p < 0.001) with a distinct additional second peak from late births at day 26, and a first peak at day 19 (Fig. 3e) There was no indication of deviation from unimodality in this cohort in the male returned treatment, and in no other combinations of cohort and treatment (all D < 0.09, all p > 0.25, Fig. 3).
In the field experiments with groups of females (MMMFFF and MFFF), we had also used different cohorts of females. The proportion of late pregnancies in multi-female experiments was very small (9%) and not depend on female age or cohorts (Table 2, please note that the small sample size did not allow testing effects of male number (H2) and female cohorts (H4) in the same model. We therefore ran separate models for the two hypotheses).
Additional late pregnancies or replacement of early pregnancies?
With an imbalanced distribution of the response variable (gravid or not) among population compositions and the occurrence of empty cells, it was impossible to run a binary model across all groups. Using post hoc tests within levels, we found no simple effects of population composition within turnover treatments on pregnancy rate, and we found no simple effects of turnover treatments within population compositions on pregnancy rates (all Fisher’s exact tests, p > 0.49).
In the multi-female experiments (MFFF vs MMMFFF), 44 litters were born to 54 recaptured females (81%). Neither male numbers (Χ
2 = 0.9, df = 1, p = 0.40) nor male turnover (Χ
2 = 0.2, df = 1, p = 0.64) explained the rates of pregnancy in these experiments. Parous females had higher pregnancy rates (16/17) than nulliparous females (28/37, Χ
2 = 3.9, df = 1, p = 0.048) and year-born females tended to have higher pregnancy rates (28/33) than overwintered females (15/20, one female with unknown age, Χ
2 = 2.7, df = 1, p = 0.099, Table 4).
In the pairing experiment on captive voles, the replacement of the original male did not add additional pregnancies (Χ
2 = 2.1, df = 1, p = 0.15). The probability of pregnancy differed among female age cohorts though (Χ
2 = 14.9, df = 2, p < 0.001). Pregnancy rates were higher in both YY-n and YY-p (83 and 88%, Χ
2 test: Χ
2 = 0.1, df = 1, p = 0.80) as compared to OW-n females (61%, Χ
2 = 7.7, p = 0.006 and Χ
2 = 6.72, p = 0.01). We found no interaction of cohort and male turnover treatment (Table 4).