Dataset DOI: 10.5061/dryad.k98sf7mmj

Description of the data and file structure

The data were used to test whether stimulus type, female body condition, and background acoustic noise influenced female responses to the robot frog. We used model selection with generalized linear mixed models (GLMMs) to analyze the probability of female response and random-effect (frailty) Cox Proportional Hazards (CPH) models to evaluate latency to female response. Female identity was included as a random factor in all models to account for repeated exposures of the same individual to different stimuli. The two continuous predictors, body condition and background acoustic noise, were standardized (i.e., centered at zero and scaled by standard deviation). 

The response variables used in the GLMMs were: (1) the probability of emitting an acoustic signal, modeled with a binomial error distribution and a log-log (cloglog) link function; and (2) the probability of emitting visual signals, modeled with a binomial error distribution and a logit link function. The response variables used in the CPH models were: (3) latency to the first acoustic response, (4) latency to the first visual response, and (5) latency to move toward the robot. Females that did not exhibit a behavioral response during the 480-second trial period were treated as right-censored observations.

Files and variables

File: env_0_1_fixed.csv

Description: File used in model selection of probability of (1) acoustic signaling, and (2) visual signalling, by stimulus type, body condition, and background noise. Complete results are in Supplementary Material S5.

Variables

• ind: Female individual identity (numbering follows the capture–mark sequence; not all marked females were tested)
• stimulus: Stimulus type (categorical; either visual, acoustic, or multimodal)
• temp_raw: Air temperature (in ºC)
• temp: Air temperature (standardized, i.e., centered to zero and scaled by standard deviation)
• condition: Female body condition, estimated from the residuals from an ordinary least squares regression between log10-transformed body mass and log10-transformed snout-vent length (standardized, i.e., centered to zero and scaled by standard deviation)
• noise_raw: Acoustic background noise (in dB)
• noise: Acoustic background noise (standardized, i.e., centered to zero and scaled by standard deviation)
• call01: Probability of acoustic signaling (1 = call emission; 0 = no call emission)
• vsig01: Probability of visual signaling (1 = visual signal emission; 0 = no visual signal emission)

File: lat_surv_females_fixed.csv

Description: File used in model selection of latency to (1) acoustic signaling, (2) visual signaling, and (3) moving towards the robot, by stimulus type, body condition, and background noise. Complete results are in Supplementary Material S6.

Variables

• ind: Female individual identity (numbering follows the capture–mark sequence; not all marked females were tested)
• stim: Stimulus type (categorical; either visual, acoustic, or multimodal)
• temp_raw: Air temperature (in ºC)
• temp: Air temperature (standardized, i.e., centered to zero and scaled by standard deviation)
• condition: Female body condition, estimated from the residuals from an ordinary least squares regression between log10-transformed body mass and log10-transformed snout-vent length (standardized, i.e., centered to zero and scaled by standard deviation)
• noise_raw: Acoustic background noise (in dB)
• noise: Acoustic background noise (standardized, i.e., centered to zero and scaled by standard deviation)
• lat_call: Latency to acoustic signaling (time, in seconds, to the first call emission)
• call_status: Indicates whether a calling response occurred (1 = non-respondent females; 2 = respondent females)
• lat_vs: Latency to visual signaling (time, in seconds, to the first visual signal emission)
• vs_status: Indicates whether a visual signaling response occurred (1 = non-respondent females; 2 = respondent females)
• lat_mov: Latency to moving towards the robot  (time, in seconds, to the first move toward the robot)
• mov_status: Indicates whether a movement towards the robot occurred (1 = non-respondent females; 2 = respondent females)

Regarding the latency data at 480 s, it should be noted that these values do not represent actual response times. Rather, they correspond to right-censored observations in the context of Cox Proportional Hazards (CPH) models, which occur when a female did not exhibit the focal behavior within the 480 s trial window. These cases were not treated as if the female responded at 480 s but instead were coded as censored and handled explicitly in the CPH framework, in line with standard procedures in survival analysis (Therneau 2020; see also Chapter 3 in Klein & Moeschberger 2003). 

The survival function (i.e., the probability of “not responding” over time) and the cumulative hazard function (i.e., the accumulated risk of responding) are estimated using partial likelihood methods that incorporate both event times (respondents) and censoring information (non-respondents). The inclusion of censored data ensures that the estimation properly accounts for the fact that the exact time-to-response is unknown for some individuals, but that they were “at risk” for the entire observation window.

Crucially, censored observations do not bias the results, since survival models treat them as incomplete information rather than fixed endpoints. Instead, they contribute partial information about the likelihood of the hazard occurring, which is integral to the accurate estimation of both model coefficients and cumulative hazard functions (see Chapter 4 in Klein & Moeschberger, 2003). Therefore, including these non-responding females is necessary for an unbiased analysis of latency.

File: env_0_1.R

Description: R code of model selection of probability of (1) acoustic signaling, and (2) visual signalling, by stimulus type, body condition, and background noise

File: lat_surv_females.R

Description: R code of model selection of latency to (1) acoustic signaling, (2) visual signaling, and (3) moving towards the robot, by stimulus type, body condition, and background noise

References

Klein, J. P., & Moeschberger, M. L. (2003). Survival Analysis: Techniques for Censored and Truncated Data. Springer.

Therneau, T. (2020). A Package for Survival Analysis in R. R package version 3.5-7.