Statistical analyses
We analysed differences in mating success between female-exposed and
control males using a one-sample Wilcoxon signed rank test with
continuity correction; each female was assigned a binary response
representing mate choice (“0” for control male, “1” for
female-exposed male), with mu set at 0.5 as females should have no
preference for either male under H0. We analysed mating
latency and duration of female-exposed versus control males using
Kruskal-Wallis rank sum tests in which treatment (female-exposed vs
control) was the sole categorical predictor. We analysed remating
latency of treatment vs control females using a Cox proportional hazard
model (Cox, 1972) with treatment as predictor, and we right-censored
females that did not remate after the 4 days. We graphically and
statistically verified the assumptions of the Cox proportional hazard
model using Schoenfeld residuals diagnostics (Schoenfeld, 1982). We
analysed female early-life reproductive success (daily offspring
production over 7 days) using a general linear mixed model ( “lme4” R
package; Bates et al. , 2015) with treatment, day and their
interaction as categorical fixed effects, and female ID as random
effect. We extracted the absolute values of the residuals-vs-fitted from
an initial heteroskedastic model and used them as weights in order to
meet the homoskedasticity assumption of the linear model (Midi, Rana and
Imon, 2009, 2013). We analysed the effects of sexual perception on
female lifetime reproductive success in a general linear model including
treatment as the sole categorical fixed effect. Finally, we analysed
sperm-offense data in a generalized linear mixed model with a
beta-binomial error distribution using the “glmmTMB” R package (to
deal with under-dispersion; Brooks et al. , 2017). We transformed
this data in order to meet the beta distribution range (i.e. y’= (y*(N-
1)+0.5)/N ; Smithson and Verkuilen, 2006). Treatment was the sole fixed
effect predictor included in this model, and batch was the only random
effect.
We ran all statistical tests, and produced all figures in R studio
1.1.456 (R Core Team, 2020). For all tests, we set α=0.05, ran type III
ANOVA and checked model assumptions using the “performance” R package
(Lüdecke et al. , 2021). We corrected for multiple comparison
using the Benjamini-Hochberg procedure (1995) for a false discovery rate
of 0.05; outcome of this procedure is detailed when relevant (cases of
false positive) in the result and discussion sections. We produced all
figures using the “ggplot2” R package (Wickham, 2016)