STAT3926/STAT4026: Statistical Consulting Project
📅 Published: May 20, 2024
👨💻 Authors: 520432255, 500480816, 500573428
🎯 Client: Magic Mei-Ting Kao
This consulting project focused on evaluating and enhancing the client’s current statistical modelling workflow for analysing the vocalisation behaviour of Chalinolobus gouldii (Gould’s wattled bats). We confirmed that a Poisson Generalised Linear Mixed Model (GLMM) is appropriate, implemented enhancements, validated assumptions, and recommended a reproducible and adaptable framework to support the client’s PhD thesis.
- Determine factors affecting vocalisation activity at foraging sites.
- Evaluate the suitability of existing models and implement GLMMs with nested random effects.
- Provide a reproducible workflow for analysing general activity and social call activity.
- Bat activity and environmental covariates collected across multiple sites and dates.
- Variables include bat activity counts, vegetation, temperature, rainfall, and anthropogenic factors.
- Focused on three key reproductive periods: mating, breeding, and pregnancy.
- Data preprocessing and filtering.
- EDA: Histograms, skewness checks, correlation matrix.
- Model fitted: Poisson GLMM, with date nested within location.
- Diagnostics via DHARMa: residuals, dispersion, outlier and zero-inflation tests.
- Visualised fixed and random effects to interpret ecological relevance.
- 🦇 Activity is significantly lower during mating (67%) and pregnancy (65%).
- 🌡️ Each 1°C rise in temperature increases activity by 9% (p < 0.01).
- 🏙️ Anthropogenic features significantly reduce activity by 27% (p < 0.05).
- 🌧️ Rainfall and water areas had non-significant effects.
- 📍 Random effects showed major location-specific variability:
- High activity at BLH_w4 (+62%) and SOP_brickpit (+34%).
- Low activity at CTNP_duck (22%) and CSF_w (71%).
| Diagnostic Test | Result |
|---|---|
| Kolmogorov-Smirnov | p = 0.156 ✅ |
| Dispersion | p = 0.216 ✅ |
| Outlier Detection | p = 0.9 ✅ |
| Zero-Inflation | p = 0.97 ✅ (well-fit) |
These results confirm the Poisson GLMM as a robust model for the data.
- Figure 1: Correlation heatmap
- Figure 2: Distribution of
overall_activity - Figure 12: Residual diagnostic plots
- Figure 13: Fixed effects estimates with significance indicators
- Figure 14: Random effects per location visualisation
- Continue using Poisson GLMM, enhanced with:
- Robust residual checks
- Clear fixed/random effects interpretation
- Apply same workflow to social call activity analysis
- Validate models in
brmsandMCMCglmmfor publication-ready analysis - Maintain focus on reproducibility and scientific rigour