Economic Investment by Ant Colonies in Searches for Better Homes. Authors Carolina Doran1, 2, Tom Pearce1, Aaron Connor1, Thomas Schlegel1, Elizabeth Franklin1, Ana B. Sendova-Franks3 and Nigel R. Franks1. 1 – School of Biological Sciences, University of Bristol, Woodland Road, Bristol BS81UG, UK 2 – Champalimaud Neuroscience Programme, Champalimaud Centre for the Unknown, Av. Brasília, Lisbon 1400-038, Portugal 3 - Department of Engineering Design and Mathematics, UWE Bristol, Coldharbour Lane, Bristol BS161QY, UK SUPPLEMENTARY MATERIAL 1 All graphical and statistical analyses were carried out in the statistical software package R (version 2.15.2)[1]. Figure S1A: The number of scouts, as a percentage of the total number of workers, counted every 10 min over 5 hours for 15 colonies each housed in each of five nests of different quality; a loess curve has been fitted to the data for each colony for each nest quality to facilitate comparisons. 2 Figure S1B: The same relationship as in figure S1A but with the data points removed, leaving only the fitted loess curves for clarity. 3 Model Selection Random effects Df AIC BIC logLik 1 (1|colony) 11 7844 7907 -3911 2 (nest|colony) 25 3500 3643 -1725 4373 14 <0.0001 3 (1|colony) + (-1+nest|colony) 26 3502 3651 -1725 0.00 1 1.0000 4 (time|colony) na na na na na na na 5 (1|colony) + (-1+time|colony) na na na na na na na 26 3359 3508 -1654 143 0 <0.0001 27 3361 3516 -1654 0.00 1 0.9998 41 2917 3153 -1417 472 14 <0.0001 42 2919 3160 -1417 0.00 1 0.9999 6 7 (nest|colony) + (-1+time|colony) (1|colony) + (-1+nest|colony) + (-1+time|colony) Chisq Chi Model Df Pr(>Chisq) (nest|colony) + 8 (-1+time|colony) + (-1+nest:time|colony) (1|colony) + (-1+nest|colony) 9 + (-1+time|colony) + (-1+nest:time|colony) Table S1: The nine models that were compared differ only in their random effects component. The response variable in each model is the log ratio between the number of workers scouting and the number of workers not scouting. The fixed component consists of the effect of nest quality, time and the interaction between nest quality and time. All nine models were fitted with the glmer() function in the R package ‘lme4’ [2] and compared with the anova() function, which uses a Likelihood Ratio Test (LRT) to compare models fitted with glmer(). A significant Chi-sq test statistic from the LRT indicates that the extra term in a subsequent model contributes significantly. A smaller value for AIC or BIC indicates a better fit of the model to the data. Model 1 is the null model, with the random factor ‘colony’ allowed to vary around the overall intercept (average value). Model 8 is the chosen, best, model. The ΔAIC between model 8 and 9 is 2 and hence there is nothing to distinguish between these two models either in terms of goodness-of-fit or in terms of parameter estimates. Values for models 4 and 5 are not included because in either case the algorithm for parameter estimation gave false convergence (warning message 8) at the second iteration. All other models reached convergence before the default 300 iterations. 4 Figure S2: A ‘caterpillar’ plot for the effect of the random factor ‘colony’ in the null model, 1 (Table S1). The 95% prediction intervals (PIs) for two of the colonies touch 0 but the PIs for none of the 15 colonies straddle 0. This means that colonies deviate significantly from the overall intercept and hence the inclusion of the random factor ‘colony’ in the model is necessary. Therefore, applying a Generalized Linear Mixed Model rather than a Generalized Linear Model is justified. 5 Goodness-of-fit for the chosen model Figure S3: The overall residuals for the chosen model 8 (Table S1) are approximately normal with just a little deviation from normality in the tails (as evident from the Q-Q plot on the right). A statistical test shows a significant deviation from normality (Shapiro-Wilk normality test: W = 0.9957, p-value = 2.96e-06) but this is to be expected to some extent for a sample size of 2325 (15 colonies x 5 nest qualities x 31 time points) and the model is robust to some deviation from normality. Figure S4: A Q-Q plot for the residuals per nest quality (1 to 5 represent increasing nest quality) reveals that most of the deviations from normality are in the extreme tails particularly for the highest nest qualities, 4 (Good) and 5 (Excellent). 6 Figure S5: An examination of residuals-versus-fitted-values plots per nest quality reveals that the residuals are approximately homogeneous (1 to 5 represent increasing nest quality). 7 overdisp_fun <- function(model) { ## number of variance parameters in ## an n-by-n variance-covariance matrix vpars <- function(m) {nrow(m)*(nrow(m)+1)/2} model.df <- sum(sapply(VarCorr(model),vpars))+length(fixef(model)) (rdf <- nrow(model@frame)-model.df) rp <- residuals(model) Pearson.chisq <- sum(rp^2) prat <- Pearson.chisq/rdf pval <- pchisq(Pearson.chisq, df=rdf, lower.tail=FALSE) c(chisq=Pearson.chisq,ratio=prat,rdf=rdf,p=pval)} >overdisp_fun(mmod8.sum) chisq ratio rdf 2083.8025402 0.9123479 2284.0000000 p 0.9988304 Table S2: We used Ben Bolker’s overdisp.fun() function to test for overdispersion in the chosen model 8, with sum contrasts (Table S1). Binomial (and Poisson) error terms assume that the variance and mean of the dependant variable are equal. Specifically, if the residual deviance (i.e. the residual sum of squares) is greater than the residual degrees of freedom (i.e. the variance is greater than expected for a standardised Binomial (Poisson) distribution), then some variance remains unexplained. This is called overdispersion ([3], pp. 59-61). Therefore, ideally, the ratio above (between the residual deviance and the residual degrees of freedom) should be 1. For model 8, the ratio is approximately 0.912, which suggests a slight underdispersion, but it is not statistically significant because the p-value is greater than 0.05. Parameter estimation Nest quality Lower limit for 95% CI Estimated probability Upper limit for 95%CI Excellent 0.02019156 0.02940141 0.04262932 Good 0.05350015 0.07757514 0.11121093 Medium 0.05088081 0.07240533 0.10205643 Satisfactory 0.08290268 0.11499615 0.15738163 Poor 0.10023777 0.13370814 0.17616642 Table S3: Estimated probability with 95% CIs for an ant scouting for the five different levels of nest quality. These are the values used in Fig. 2a in the paper. These values were calculated from the base level intercepts and their standard errors in the models in Tables S7 – S11 below using the z value of 1.96. Probability was estimated with plogis(base_intercept), the lower limit for the 95% CI with plogis(base_intercept+qnorm(0.025)*standard_error) and the upper limit for the 95% CI with plogis(base_intercept+qnorm(0.975)*standard_error). Nest quality Lower limit for 95% CI Estimated odds ratio Upper limit for 95%CI Excellent 0.9974921 0.9991742 1.0008591 Good 0.9958724 0.9976319 0.9993945 Medium 0.9978243 0.9987898 0.9997562 Satisfactory 0.9979308 0.9985611 0.9994206 Poor 0.9986172 0.9992823 0.9999479 Table S4: Estimated odds ratio with 95% CIs for an ant scouting in successive 10 min for the five different levels of nest quality. These are the values used in Fig. 2b in the paper. These values were calculated from the base level coefficients for time and their standard errors in the models in Tables S7 – S11 below using the z value of 1.96. The odds ratio was estimated with exp(base_coefficient), the lower limit for the 95% CI with exp(base_coefficient+qnorm(0.025)*standard_error) and the upper limit for the 95% CI with exp(base_coefficient+qnorm(0.975)*standard_error). 8 Generalized linear mixed model fit by the Laplace approximation Formula: cbind(scouts$NoScouts, scouts$TotalAnts - scouts$NoScouts) scouts$Nest * scouts$Time + (scouts$Nest | scouts$Colony) + scouts$Time | scouts$Colony) + (-1 + scouts$Nest:scouts$Time | scouts$Colony) AIC BIC logLik deviance 2917 3153 -1417 2835 Random effects: Groups Name Variance Std.Dev. Corr scouts$Colony (Intercept) 1.9887e-01 0.4459495 scouts$Nest1 1.9964e-01 0.4468132 0.019 scouts$Nest2 1.7128e-01 0.4138631 0.360 scouts$Nest3 2.6454e-01 0.5143363 0.157 scouts$Nest4 3.0814e-01 0.5551047 0.187 scouts$Colony scouts$Time 4.9952e-08 0.0002235 scouts$Colony scouts$Nest1:scouts$Time 1.3763e-06 0.0011731 scouts$Nest2:scouts$Time 2.4504e-06 0.0015654 0.297 scouts$Nest3:scouts$Time 3.0467e-06 0.0017455 0.311 scouts$Nest4:scouts$Time 1.1420e-05 0.0033793 0.062 scouts$Nest5:scouts$Time 9.7144e-06 0.0031168 0.278 ~ (-1 + -0.286 -0.005 0.054 -0.408 -0.502 0.274 -0.406 0.098 -0.152 0.374 -0.367 0.069 Number of obs: 2325, groups: scouts$Colony, 15 Fixed effects: Estimate Std. Error z value Pr(>|z|) (Intercept) -2.4864269 0.1161956 -21.399 < 2e-16 scouts$Nest1 0.6178854 0.1179306 5.239 1.61e-07 scouts$Nest2 0.4457499 0.1100154 4.052 5.08e-05 scouts$Nest3 -0.0638854 0.1360621 -0.470 0.6387 scouts$Nest4 0.0106903 0.1464892 0.073 0.9418 scouts$Time -0.0013132 0.0003032 -4.332 1.48e-05 scouts$Nest1:scouts$Time 0.0005953 0.0003003 1.982 0.0474 scouts$Nest2:scouts$Time -0.0001267 0.0005025 -0.252 0.8009 scouts$Nest3:scouts$Time 0.0001023 0.0004896 0.209 0.8345 scouts$Nest4:scouts$Time -0.0010578 0.0007477 -1.415 0.1571 --Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1 Correlation of Fixed Effects: (Intr) sct$N1 sct$N2 sct$N3 sct$N4 scouts$Nst1 0.009 scouts$Nst2 0.339 -0.273 scouts$Nst3 0.150 -0.010 -0.393 scouts$Nst4 0.180 0.045 0.085 -0.155 scouts$Time -0.034 0.017 0.014 0.004 0.001 scts$Ns1:$T 0.017 -0.083 0.003 0.010 0.013 scts$Ns2:$T 0.008 0.002 -0.061 0.009 0.010 scts$Ns3:$T 0.003 0.007 0.011 -0.065 0.014 scts$Ns4:$T 0.000 0.007 0.009 0.010 -0.043 *** *** *** *** * scts$T s$N1:$ s$N2:$ s$N3:$ -0.366 -0.498 0.379 -0.297 0.215 -0.175 0.353 -0.529 -0.719 0.061 > plogis(-2.4864269) [1] 0.0768152 # Average probability of an ant scouting over all 5 nest qualities. > exp(-0.0013132) [1] 0.9986877 # Average odds ratio for an ant scouting per 10 min increase in time. 9 Table S5: The chosen model 8 run with sum contrasts. In this case the estimate for the highlighted intercept is the log odds for an ant scouting over all 5 nest qualities. Since it is also the base level, the odds ratio is equal to 1. Therefore, instead of converting the intercept into an odds ratio of probability of an ant scouting for a particular category versus the probability of an ant scouting for the base category, we convert directly into probability. The probability is 1/(1+exp(-intercept)), which is the same as the probability estimated with the probability function for the logistic distribution, plogis(). The highlighted coefficient against time is the log odds for an ant scouting with each successive 10 min. The odds ratio is exp(coefficient against time). 10 Generalized linear mixed model fit by the Laplace approximation Formula: cbind(scouts$NoScouts, scouts$TotalAnts - scouts$NoScouts) ~ scouts$Nest * scouts$Time + (scouts$Nest | scouts$Colony) + (-1 + scouts$Time | scouts$Colony) + (-1 + scouts$Nest:scouts$Time | scouts$Colony) AIC BIC logLik deviance 2917 3153 -1417 2835 Random effects: Groups Name Variance Std.Dev. Corr scouts$Colony (Intercept) 1.9887e-01 0.44594563 scouts$Nest.L 3.8462e-01 0.62018082 -0.380 scouts$Nest.Q 4.1698e-01 0.64574351 -0.447 0.500 scouts$Nest.C 4.1512e-01 0.64429638 -0.128 0.600 scouts$Nest^4 3.7455e-01 0.61200231 -0.168 0.136 scouts$Colony scouts$Time 1.6168e-08 0.00012715 scouts$Colony scouts$Nest1:scouts$Time 1.4101e-06 0.00118746 scouts$Nest2:scouts$Time 2.4843e-06 0.00157616 0.309 scouts$Nest3:scouts$Time 3.0804e-06 0.00175509 0.322 -0.390 scouts$Nest4:scouts$Time 1.1454e-05 0.00338433 0.070 -0.492 scouts$Nest5:scouts$Time 9.7481e-06 0.00312220 0.283 0.278 0.537 0.156 0.178 0.377 -0.358 0.072 Number of obs: 2325, groups: scouts$Colony, 15 Fixed effects: Estimate Std. Error z value Pr(>|z|) (Intercept) -2.4864324 0.1161954 -21.399 < 2e-16 scouts$Nest.L -1.1674045 0.1645446 -7.095 1.30e-12 scouts$Nest.Q -0.2976697 0.1706469 -1.744 0.0811 scouts$Nest.C -0.2397613 0.1695068 -1.414 0.1572 scouts$Nest^4 -0.3109517 0.1611995 -1.929 0.0537 scouts$Time -0.0013132 0.0003032 -4.332 1.48e-05 scouts$Nest.L:scouts$Time -0.0003630 0.0006417 -0.566 0.5716 scouts$Nest.Q:scouts$Time 0.0008404 0.0006429 1.307 0.1912 scouts$Nest.C:scouts$Time 0.0005546 0.0007873 0.704 0.4812 scouts$Nest^4:scouts$Time 0.0007687 0.0004441 1.731 0.0835 --Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1 Correlation of Fixed Effects: (Intr) sc$N.L sc$N.Q scots$Nst.L -0.353 scots$Nst.Q -0.427 0.497 scots$Nst.C -0.120 0.582 0.528 scots$Nst^4 -0.161 0.133 0.150 scouts$Time -0.034 -0.026 -0.011 scts$N.L:$T -0.018 -0.067 -0.027 scts$N.Q:$T -0.008 -0.028 -0.059 scts$N.C:$T -0.003 -0.009 -0.015 scts$N^4:$T -0.004 -0.005 -0.002 *** *** . . *** . sc$N.C sc$N^4 scts$T s$N.L: s$N.Q: s$N.C: 0.178 -0.006 -0.010 -0.018 -0.040 -0.013 -0.004 -0.003 0.600 -0.002 0.224 0.378 -0.008 -0.272 -0.242 0.647 -0.071 -0.214 -0.365 -0.092 0.177 Table S6: The chosen model 8 run with polynomial contrasts. As the highlighted values show, most of the variation in nest quality is explained by the linear contrasts. The contribution of quadratic, cubic or quartic contrasts is not statistically significant. 11 Generalized linear mixed model fit by the Laplace approximation Formula: cbind(scouts$NoScouts, scouts$TotalAnts - scouts$NoScouts) ~ scouts$Nest * scouts$Time + (scouts$Nest | scouts$Colony) + (-1 + scouts$Time | scouts$Colony) + (-1 + scouts$Nest:scouts$Time | scouts$Colony) AIC BIC logLik deviance 2917 3153 -1417 2835 Random effects: Groups Name Variance Std.Dev. Corr scouts$Colony (Intercept) 5.4163e-01 0.73595626 scouts$Nest1 1.1652e+00 1.07943417 -0.819 scouts$Nest2 9.2814e-01 0.96339897 -0.682 0.777 scouts$Nest3 1.1792e+00 1.08591204 -0.742 0.801 scouts$Nest4 1.5571e+00 1.24785692 -0.816 0.832 scouts$Colony scouts$Time 1.0902e-07 0.00033019 scouts$Colony scouts$Nest1:scouts$Time 1.3172e-06 0.00114769 scouts$Nest2:scouts$Time 2.3914e-06 0.00154643 0.274 scouts$Nest3:scouts$Time 2.9876e-06 0.00172847 0.291 -0.439 scouts$Nest4:scouts$Time 1.1361e-05 0.00337056 0.049 -0.521 scouts$Nest5:scouts$Time 9.6555e-06 0.00310732 0.268 0.266 0.716 0.853 0.766 0.368 -0.383 0.064 Number of obs: 2325, groups: scouts$Colony, 15 Fixed effects: Estimate Std. Error z value Pr(>|z|) (Intercept) -3.4968703 0.1965454 -17.792 < 2e-16 scouts$Nest1 1.6282976 0.2842497 5.728 1.01e-08 scouts$Nest2 1.4561908 0.2551945 5.706 1.16e-08 scouts$Nest3 0.9465389 0.2866935 3.302 0.000961 scouts$Nest4 1.0211240 0.3277976 3.115 0.001839 scouts$Time -0.0008262 0.0008597 -0.961 0.336520 scouts$Nest1:scouts$Time 0.0001083 0.0008436 0.128 0.897855 scouts$Nest2:scouts$Time -0.0006137 0.0008641 -0.710 0.477539 scouts$Nest3:scouts$Time -0.0003848 0.0011145 -0.345 0.729914 scouts$Nest4:scouts$Time -0.0015448 0.0012022 -1.285 0.198788 --Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1 Correlation of Fixed Effects: (Intr) sct$N1 sct$N2 scouts$Nst1 -0.821 scouts$Nst2 -0.693 0.778 scouts$Nst3 -0.746 0.799 0.717 scouts$Nst4 -0.815 0.829 0.847 scouts$Time -0.073 0.050 0.056 scts$Ns1:$T 0.074 -0.063 -0.057 scts$Ns2:$T 0.072 -0.050 -0.072 scts$Ns3:$T 0.056 -0.039 -0.043 scts$Ns4:$T 0.052 -0.036 -0.040 *** *** *** *** ** sct$N3 sct$N4 scts$T s$N1:$ s$N2:$ s$N3:$ 0.764 0.050 -0.051 -0.049 -0.055 -0.036 0.044 -0.044 -0.043 -0.034 -0.046 -0.921 -0.870 -0.907 -0.664 0.844 0.897 0.626 0.732 0.404 0.722 Table S7: The chosen model 8 run with treatment contrasts using nest quality 5 (Excellent) as the base. The highlighted values were used to calculate the probabilities and odds ratios in row 1 of Tables S3 and S4 above. All other coefficients represent distances from the base. We used the zvalues against these non-base level coefficients to test for significant differences from the base level. To correct for the four comparisons per coefficient, we used α’ = 0.05/4 = 0.0125. 12 Generalized linear mixed model fit by the Laplace approximation Formula: cbind(scouts$NoScouts, scouts$TotalAnts - scouts$NoScouts) ~ scouts$Nest * scouts$Time + (scouts$Nest | scouts$Colony) + (-1 + scouts$Time | scouts$Colony) + (-1 + scouts$Nest:scouts$Time | scouts$Colony) AIC BIC logLik deviance 2917 3153 -1417 2835 Random effects: Groups Name Variance Std.Dev. Corr scouts$Colony (Intercept) 5.9963e-01 7.7436e-01 scouts$Nest1 4.8113e-01 6.9364e-01 -0.628 scouts$Nest2 4.3430e-01 6.5902e-01 -0.520 0.480 scouts$Nest3 6.5945e-01 8.1206e-01 -0.575 0.598 scouts$Nest5 1.5571e+00 1.2478e+00 -0.836 0.504 scouts$Colony scouts$Time 1.7013e-11 4.1247e-06 scouts$Colony scouts$Nest1:scouts$Time 1.4262e-06 1.1942e-03 scouts$Nest2:scouts$Time 2.5005e-06 1.5813e-03 0.315 scouts$Nest3:scouts$Time 3.0965e-06 1.7597e-03 0.327 -0.382 scouts$Nest4:scouts$Time 1.1470e-05 3.3867e-03 0.074 -0.487 scouts$Nest5:scouts$Time 9.7640e-06 3.1247e-03 0.286 0.280 0.453 0.646 0.512 0.378 -0.354 0.074 Number of obs: 2325, groups: scouts$Colony, 15 Fixed effects: Estimate Std. Error z value Pr(>|z|) (Intercept) -2.4757588 0.2027219 -12.213 < 2e-16 scouts$Nest1 0.6071938 0.1838439 3.303 0.000957 scouts$Nest2 0.4350706 0.1755494 2.478 0.013200 scouts$Nest3 -0.0745592 0.2148013 -0.347 0.728510 scouts$Nest5 -1.0210912 0.3277979 -3.115 0.001839 scouts$Time -0.0023709 0.0009007 -2.632 0.008477 scouts$Nest1:scouts$Time 0.0016531 0.0009417 1.755 0.079191 scouts$Nest2:scouts$Time 0.0009311 0.0011626 0.801 0.423238 scouts$Nest3:scouts$Time 0.0011600 0.0008682 1.336 0.181526 scouts$Nest5:scouts$Time 0.0015440 0.0012022 1.284 0.199020 --Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1 Correlation of Fixed Effects: (Intr) sct$N1 sct$N2 scouts$Nst1 -0.634 scouts$Nst2 -0.529 0.488 scouts$Nst3 -0.580 0.597 0.458 scouts$Nst5 -0.827 0.502 0.636 scouts$Time -0.032 0.035 0.037 scts$Ns1:$T 0.031 -0.051 -0.036 scts$Ns2:$T 0.025 -0.028 -0.047 scts$Ns3:$T 0.033 -0.037 -0.038 scts$Ns5:$T 0.024 -0.027 -0.028 *** *** * ** ** . sct$N3 sct$N5 scts$T s$N1:$ s$N2:$ s$N3:$ 0.507 0.030 -0.029 -0.023 -0.059 -0.023 0.020 -0.019 -0.015 -0.020 -0.046 -0.933 -0.941 -0.845 -0.701 0.918 0.840 0.716 0.756 0.733 0.458 Table S8: The chosen model 8 run with treatment contrasts using nest quality 4 (Good) as the base. The highlighted values were used to calculate the probabilities and odds ratios in row 2 of Tables S3 and S4 above. We used the z-values against the non-base level coefficients to test for significant differences from the base level. 13 Generalized linear mixed model fit by the Laplace approximation Formula: cbind(scouts$NoScouts, scouts$TotalAnts - scouts$NoScouts) ~ scouts$Nest * scouts$Time + (scouts$Nest | scouts$Colony) + (-1 + scouts$Time | scouts$Colony) + (-1 + scouts$Nest:scouts$Time | scouts$Colony) AIC BIC logLik deviance 2917 3153 -1417 2835 Random effects: Groups Name Variance Std.Dev. Corr scouts$Colony (Intercept) 5.3536e-01 7.3168e-01 scouts$Nest1 4.6664e-01 6.8311e-01 -0.596 scouts$Nest2 6.0869e-01 7.8019e-01 -0.561 0.562 scouts$Nest4 6.5945e-01 8.1206e-01 -0.501 0.581 scouts$Nest5 1.1792e+00 1.0859e+00 -0.738 0.324 scouts$Colony scouts$Time 4.5593e-12 2.1353e-06 scouts$Colony scouts$Nest1:scouts$Time 1.4262e-06 1.1942e-03 scouts$Nest2:scouts$Time 2.5004e-06 1.5813e-03 0.315 scouts$Nest3:scouts$Time 3.0966e-06 1.7597e-03 0.327 -0.382 scouts$Nest4:scouts$Time 1.1470e-05 3.3867e-03 0.074 -0.487 scouts$Nest5:scouts$Time 9.7643e-06 3.1248e-03 0.286 0.280 0.658 0.507 0.160 0.378 -0.354 0.074 Number of obs: 2325, groups: scouts$Colony, 15 Fixed effects: Estimate Std. Error z value Pr(>|z|) (Intercept) -2.5503150 0.1917044 -13.303 < 2e-16 scouts$Nest1 0.6817523 0.1809842 3.767 0.000165 scouts$Nest2 0.5096210 0.2058805 2.475 0.013312 scouts$Nest4 0.0745651 0.2148016 0.347 0.728490 scouts$Nest5 -0.9465439 0.2866954 -3.302 0.000961 scouts$Time -0.0012109 0.0004934 -2.454 0.014121 scouts$Nest1:scouts$Time 0.0004931 0.0005163 0.955 0.339586 scouts$Nest2:scouts$Time -0.0002290 0.0007605 -0.301 0.763343 scouts$Nest4:scouts$Time -0.0011601 0.0008682 -1.336 0.181497 scouts$Nest5:scouts$Time 0.0003840 0.0011146 0.345 0.730432 --Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1 Correlation of Fixed Effects: (Intr) sct$N1 sct$N2 scouts$Nst1 -0.603 scouts$Nst2 -0.568 0.564 scouts$Nst4 -0.507 0.580 0.652 scouts$Nst5 -0.731 0.329 0.504 scouts$Time -0.055 0.058 0.052 scts$Ns1:$T 0.052 -0.087 -0.049 scts$Ns2:$T 0.036 -0.038 -0.057 scts$Ns4:$T 0.031 -0.033 -0.029 scts$Ns5:$T 0.025 -0.026 -0.023 *** *** * *** * sct$N4 sct$N5 scts$T s$N1:$ s$N2:$ s$N4:$ 0.170 0.049 -0.047 -0.032 -0.059 -0.022 0.037 -0.035 -0.024 -0.021 -0.055 -0.774 -0.838 -0.217 -0.679 0.785 0.149 -0.015 0.693 0.634 0.285 Table S9: The chosen model 8 run with treatment contrasts using nest quality 3 (Medium) as the base. The highlighted values were used to calculate the probabilities and odds ratios in row 3 of Tables S3 and S4 above. We used the z-values against the non-base level coefficients to test for significant differences from the base level. 14 Generalized linear mixed model fit by the Laplace approximation Formula: cbind(scouts$NoScouts, scouts$TotalAnts - scouts$NoScouts) ~ scouts$Nest * scouts$Time + (scouts$Nest | scouts$Colony) + (-1 + scouts$Time | scouts$Colony) + (-1 + scouts$Nest:scouts$Time | scouts$Colony) AIC BIC logLik deviance 2917 3153 -1417 2835 Random effects: Groups Name Variance Std.Dev. Corr scouts$Colony (Intercept) 5.0302e-01 0.70923641 scouts$Nest1 4.7654e-01 0.69031619 -0.586 scouts$Nest3 6.0869e-01 0.78018877 -0.521 0.574 scouts$Nest4 4.3431e-01 0.65902083 -0.361 0.472 scouts$Nest5 9.2812e-01 0.96338894 -0.651 0.180 scouts$Colony scouts$Time 1.4725e-07 0.00038373 scouts$Colony scouts$Nest1:scouts$Time 1.2789e-06 0.00113089 scouts$Nest2:scouts$Time 2.3531e-06 0.00153398 0.258 scouts$Nest3:scouts$Time 2.9492e-06 0.00171733 0.278 -0.460 scouts$Nest4:scouts$Time 1.1322e-05 0.00336489 0.039 -0.533 scouts$Nest5:scouts$Time 9.6170e-06 0.00310113 0.262 0.260 0.373 0.238 -0.153 0.365 -0.393 0.061 Number of obs: 2325, groups: scouts$Colony, 15 Fixed effects: Estimate Std. Error z value Pr(>|z|) (Intercept) -2.0406933 0.1851324 -11.023 < 2e-16 scouts$Nest1 0.1721221 0.1819353 0.946 0.34412 scouts$Nest3 -0.5096293 0.2058843 -2.475 0.01331 scouts$Nest4 -0.4350686 0.1755518 -2.478 0.01320 scouts$Nest5 -1.4561602 0.2551917 -5.706 1.16e-08 scouts$Time -0.0014400 0.0004390 -3.280 0.00104 scouts$Nest1:scouts$Time 0.0007220 0.0004777 1.512 0.13066 scouts$Nest3:scouts$Time 0.0002291 0.0007605 0.301 0.76327 scouts$Nest4:scouts$Time -0.0009309 0.0011626 -0.801 0.42329 scouts$Nest5:scouts$Time 0.0006136 0.0008641 0.710 0.47759 --Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1 Correlation of Fixed Effects: (Intr) sct$N1 sct$N3 scouts$Nst1 -0.589 scouts$Nst3 -0.524 0.570 scouts$Nst4 -0.369 0.472 0.374 scouts$Nst5 -0.643 0.188 0.241 scouts$Time -0.045 0.045 0.041 scts$Ns1:$T 0.041 -0.075 -0.038 scts$Ns3:$T 0.026 -0.027 -0.057 scts$Ns4:$T 0.017 -0.017 -0.016 scts$Ns5:$T 0.023 -0.023 -0.020 *** * * *** ** sct$N4 sct$N5 scts$T s$N1:$ s$N3:$ s$N4:$ -0.128 0.048 -0.043 -0.028 -0.047 -0.024 0.032 -0.030 -0.019 -0.012 -0.072 -0.728 -0.790 -0.718 -0.264 0.743 0.624 0.319 0.665 0.063 0.325 Table S10: The chosen model 8 run with treatment contrasts using nest quality 2 (Satisfactory) as the base. The highlighted values were used to calculate the probabilities and odds ratios in row 4 of Tables S3 and S4 above. We used the z-values against the non-base level coefficients to test for significant differences from the base level. 15 Generalized linear mixed model fit by the Laplace approximation Formula: cbind(scouts$NoScouts, scouts$TotalAnts - scouts$NoScouts) ~ scouts$Nest * scouts$Time + (scouts$Nest | scouts$Colony) + (-1 + scouts$Time | scouts$Colony) + (-1 + scouts$Nest:scouts$Time | scouts$Colony) AIC BIC logLik deviance 2917 3153 -1417 2835 Random effects: Groups Name Variance Std.Dev. Corr scouts$Colony (Intercept) 4.0598e-01 0.63716213 scouts$Nest2 4.7654e-01 0.69032186 -0.431 scouts$Nest3 4.6662e-01 0.68309794 -0.387 0.355 scouts$Nest4 4.8113e-01 0.69363471 -0.325 0.547 scouts$Nest5 1.1652e+00 1.07942865 -0.748 0.479 scouts$Colony scouts$Time 9.3280e-08 0.00030542 scouts$Colony scouts$Nest1:scouts$Time 1.3329e-06 0.00115450 scouts$Nest2:scouts$Time 2.4072e-06 0.00155151 0.280 scouts$Nest3:scouts$Time 3.0033e-06 0.00173300 0.297 -0.430 scouts$Nest4:scouts$Time 1.1377e-05 0.00337291 0.052 -0.516 scouts$Nest5:scouts$Time 9.6710e-06 0.00310982 0.271 0.268 0.304 0.307 0.060 0.370 -0.378 0.065 Number of obs: 2325, groups: scouts$Colony, 15 Fixed effects: Estimate Std. Error z value Pr(>|z|) (Intercept) -1.8685625 0.1663413 -11.233 < 2e-16 scouts$Nest2 -0.1721292 0.1819347 -0.946 0.344095 scouts$Nest3 -0.6817563 0.1809810 -3.767 0.000165 scouts$Nest4 -0.6071866 0.1838430 -3.303 0.000957 scouts$Nest5 -1.6282975 0.2842470 -5.728 1.01e-08 scouts$Time -0.0007179 0.0003397 -2.113 0.034584 scouts$Nest2:scouts$Time -0.0007220 0.0004777 -1.511 0.130668 scouts$Nest3:scouts$Time -0.0004931 0.0005163 -0.955 0.339574 scouts$Nest4:scouts$Time -0.0016530 0.0009417 -1.755 0.079197 scouts$Nest5:scouts$Time -0.0001091 0.0008436 -0.129 0.897071 --Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1 Correlation of Fixed Effects: (Intr) sct$N2 sct$N3 scouts$Nst2 -0.438 scouts$Nst3 -0.393 0.356 scouts$Nst4 -0.333 0.539 0.307 scouts$Nst5 -0.739 0.472 0.305 scouts$Time -0.052 0.047 0.047 scts$Ns2:$T 0.037 -0.075 -0.033 scts$Ns3:$T 0.034 -0.030 -0.087 scts$Ns4:$T 0.019 -0.017 -0.017 scts$Ns5:$T 0.021 -0.019 -0.019 *** *** *** *** * . sct$N4 sct$N5 scts$T s$N2:$ s$N3:$ s$N4:$ 0.068 0.047 -0.033 -0.031 -0.051 -0.019 0.030 -0.021 -0.019 -0.011 -0.063 -0.465 -0.395 -0.169 -0.299 -0.263 0.411 -0.154 0.240 -0.303 0.096 Table S11: The chosen model 8 run with treatment contrasts using nest quality 1 (Poor) as the base. The highlighted values were used to calculate the probabilities and odds ratios in row 5 of Tables S3 and S4 above. We used the z-values against the non-base level coefficients to test for significant differences from the base level. 16 Model selection, goodness-of-fit tests for the chosen model and the interpretation of the model parameters were carried out according to [3–5]. For the tests of significance of the effects, we used the z-values and the associated p-values in the output from glmer(). Each such z-value is an approximation of the chi-squared value from the likelihood ratio test comparing the two models with the tested parameter present or missing, such that the square of z is approximately equal to the chi-squared value for 1 d.f. References: 1. Bates, D., Maechler, M. & Bolker, B. 2012 lme4: Linear mixed-effects models using S4 classes. R package version 0.999999-0. http://CRAN.R-project.org/package=lme4. 2. Faraway, J. J. 2005 Extending the Linear Model with R: Generalized Linear, Mixed Effects and Nonparametric Regression Models. CRC Press. 3. Maindonald, J. & Braun, W. J. 2010 Data Analysis and Graphics Using R: An Example-Based Approach (Cambridge Series in Statistical and Probabilistic Mathematics). Cambridge University Press. 4. R Core Team 2012 R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing, Vienna, Austria. ISBN 3-900051-07-0, URL http://www.R-project.org/. 5. Thomas, R., Vaughan, I. & Lello, J. 2011 Data Analysis with R Statistical Software. Cardiff: EcoExplore. 17