Electronic Supplementary Material Spontaneous female flash calls in an Asian firefly Hideo Takatsu, Mihoko Minami, Kei-ichi Tainaka, and Jin Yoshimura Statistical Inferences 1. Regression analysis of flash duration on temperature (Figure 2, Fig. S1) In this section, F and T denote flash duration and temperature, respectively. The detailed statistics were shown in Table S1 for all the regression models. For male (n = 298), the ordinary linear regression model was used for the relationship of F on T. The linear regression equation was: F = 0.190 – 0.0072T. The standard deviation of residuals was 0.025 and the adjusted R-square was 0.408. For female (n =225), the weighted linear regression model was used because the residual variance seemed to vary with temperature. We used the inverse of the squared estimated standard deviation (i.e. the estimated variance for given temperatures) as weights (Fig. S1). The linear regression equation for female was given by F = 0.753 – 0.0252T. The standard deviation of residuals was 0.1533 and the adjusted R-square was 0.337. 2. Discrimination between male and female by the logistic regression model (Figure 2) For the discrimination between male and female, we used the logistic regression model. We employed the model with flash duration and temperature as explanatory variables, because the AIC (Akaike’s information criteria) value was 21.2 with flash duration and temperature, while it was 74.5 with flash duration only. The estimated linear predictor was log (p/ (1-p)) = -39.05 + 1.38T + 103.15F, where p denotes the probability of being female. Equating the predictor to zero (i.e. p = 0.5) gives the discriminant boundary. The deviance explained by the model was 90.13%. The leave-one-out cross validation 1 estimate for the prediction error was 0.574% (3 out of 523). The densities of linear predictor scores (Inserted graph) were estimated with the kernel density estimation method with Gaussian kernel using function “density” in the statistical software R. 3. Discrimination between the success and the failure of the male attraction to the electronic firefly (Figure 3) To find the lower discriminant boundary of successful male attraction, we excluded the data in the area {F -0.066T + 1.91}, because we considered that the failures in this area were not due to confusion with the male flashes. We employed the logistic regression model with flash duration only as explanatory variables, because the AIC value was 104.5 with flash duration only, while it was 105.7 with flash duration and temperature. The estimated linear predictor was log (p/ (1-p)) = -4.77 + 29.79F, where p denotes the probability of successful male attraction. Equating the predictor to zero (i.e. p = 0.5) gives the lower discriminant boundary. To find the upper discriminant boundary, we excluded the data in the area {F 0.2}, because we considered that the failures in this area were due to confusion with the male flashes. We employed the logistic regression model with flash duration and temperature as explanatory variables, because the AIC value was 86.9 with flash duration and temperature, while it was 110.4 with flash duration only. The estimated linear predictor was log (p/ (1-p)) = 14.91 – 0.594T – 4.047F. Equating the predictor to zero (i.e. p = 0.5) gives the upper discriminant boundary of the successful male attraction. 2 Table S1. Some statistical values for the distributions of flash duration along ambient temperatures in L. parvula. For entries, Target, EC and SE denote a target factor, the estimated coefficients and the standard errors, respectively. Ob-M (Ob-F) denotes the (weighted) linear regression model for the observed males (females) in Fig.2. M-F-B denotes the logistic regression model for the discrimination between sexes in Fig.2. LS-B (U-S-B) denotes the logistic regression model for the lower (upper) discrimination boundary for successful male attractions in Fig.3. For targets, Ic, Tm and FD denote the intercept, temperature and flash duration, respectively. Model Target EC SE t-value Ob-M Ic 0.190 0.008 23.63 <2e-16 Tm -0.0072 0.0005 -14.35 <2e-16 Ic 0.753 0.041 18.29 <2e-16 Tm -0.0252 0.0024 -10.71 <2e-16 Ic -39.05 12.50 -3.124 0.001785 Tm 1.38 0.51 2.690 0.007152 FD 103.15 28.50 3.619 0.000295 Ic -4.77 0.88 -5.449 5.6e-08 FD 29.79 5.59 5.326 1.01e-07 Ic 14.91 3.05 4.885 1.03e-06 Tm -0.594 0.143 -4.169 3.06e-05 FD -4.047 0.795 -5.091 3.56e-07 Ob-F M-F-B L-S-B U-S-B 3 z-value Pr(>|t|) 0.25 0.15 0.05 moving standard deviation 12 14 16 18 20 temperature Fig. S1 The estimated residual variances for the weighted linear regression model. The circles are the 41-term moving standard deviations of flash duration with respect to temperature. The solid line is the estimated standard deviation of flash duration for given temperatures (max (0.83 – 0.048T, 0.083)). 4