Chapter 3: Data Analysis
Bernadette Rigley
Summary
Utilizing nest survival data gathered over multiple years through an experimental design setup, this project aims to compare variables using a logistic exposure model approach. Through this analysis, I intend to pinpoint the factors influencing nest survival, ultimately informing the development of effective conservation strategies for supporting nesting grassland birds.
Objectives
We aim to understand how regenerative grazing and haying regimes influence grassland bird communities on working farms in Virginia.
Questions:
- Does the timing of management practices influence the reproductive success of grassland birds in fields under regenerative haying and grazing practices.
Hypotheses:
Nest concealment hypothesis: nest success will be contingent on the timing and intensity of management, such as grazing intensity and the timing of haying. The basis for this hypothesis lies in our understanding that both grazing and haying directly modify vegetation structure. Reduced vegetation density around nests may expose nests, making them more detectable to predators, while denser vegetation may provide better concealment and protection.
Edge habitat hypothesis: landscape variables may better account for the variation in nest survival. Forest edges negatively influence nest survival as predators are more active around edges.
Methods
Here, I model nest survival as a function of nest site and treatment level habitat covariates and linear and quadratic time trends. I’ve used a logistic exposure model to calculate species specific daily survival rates (Shaffer 2004). The logistic exposure model uses a logit link function (loge[p/(1 - p)], where p is the probability of a success) and utilizes encounter history to model daily survival rates as a function of the explanatory variables. Given that nests are found at various stages, this approach accounts for the bias encountered with varying exposure.
Data Overview
- Response Variable
- Nest monitoring data collected between 2020-2023
- Covariate Data
- Random effects
- Temporal effects
- Nest site level
- Treatment level
- Landscape level
Click here for covariate definition at each scale
Nest Data
Let’s take a peek at our nest data. We have 605 nests with complete data for modeling purposes. Nests were found through both behavioral cues and systematic searching (rope dragging/ stick swishing), with some found incidentally. Nests were monitored approximately every 4-7 days with minimal disturbance until the nest either fledged or failed. Cues such as observed flightless young, parents alarm calling, adults delivering food, and undiscarded fecal sacs were used along with the age of the brood to determine if a nest was successful.
| Site | Count |
|---|---|
| Chancellors Rock | 76 |
| Glenmore | 29 |
| Hidden Creek | 3 |
| Little Milan | 19 |
| Oak Grove | 113 |
| Oak Spring | 67 |
| Over Jordan | 6 |
| Oxbow | 370 |
| Species | Failed | Successful |
|---|---|---|
| EAME | 56 | 60 |
| RWBL | 208 | 197 |
| SAVS | 3 | 2 |
| GRSP | 3 | 10 |
| BOBO | 19 | 47 |
Landcover Data
I’m using the Chesapeake Bay Program’s One-meter Resolution Land Use/Land Cover (LULC) Data to extract landscape covariates. Let’s take a look at that.
Now let’s take a look at some of our covariates. A few covariates are heavily skewed with zeros and will need to be log transformed.
Let’s take take a look at collinearity.
I will be taking a step or hierarchical approach to modeling.
- Step 1 focuses on incorporating random effects to account for variability across different levels of the data.
- Step 2 adds temporal effects to capture changes over time.
- Step 3 examines single-scale effects, including nested scales, treatment scales, and landscape scales, to identify how different spatial levels influence the outcome.
- Step 4 integrates multi-scale effects by combining the most significant single-scale effects, allowing for a comprehensive analysis of interactions across different scales.
## $`base model`
##
## Call: glm(formula = survoutcome ~ 1, family = binomial(link = logexp(cov_data$exposure)),
## data = cov_data)
##
## Coefficients:
## (Intercept)
## 2.978
##
## Degrees of Freedom: 616 Total (i.e. Null); 616 Residual
## Null Deviance: 730.3
## Residual Deviance: 750.6 AIC: 752.6
##
## $robel
##
## Call: glm(formula = survoutcome ~ robe1, family = binomial(link = logexp(cov_data$exposure)),
## data = cov_data)
##
## Coefficients:
## (Intercept) robe1
## 3.145521 -0.003341
##
## Degrees of Freedom: 616 Total (i.e. Null); 615 Residual
## Null Deviance: 730.3
## Residual Deviance: 749.6 AIC: 753.6
##
## $height
##
## Call: glm(formula = survoutcome ~ height1, family = binomial(link = logexp(cov_data$exposure)),
## data = cov_data)
##
## Coefficients:
## (Intercept) height1
## 1.85827 0.01149
##
## Degrees of Freedom: 616 Total (i.e. Null); 615 Residual
## Null Deviance: 730.3
## Residual Deviance: 736.1 AIC: 740.1
##
## $pgrass
##
## Call: glm(formula = survoutcome ~ pgrass, family = binomial(link = logexp(cov_data$exposure)),
## data = cov_data)
##
## Coefficients:
## (Intercept) pgrass
## 2.7796 0.3216
##
## Degrees of Freedom: 616 Total (i.e. Null); 615 Residual
## Null Deviance: 730.3
## Residual Deviance: 749.6 AIC: 753.6
##
## $pforb
##
## Call: glm(formula = survoutcome ~ pforb, family = binomial(link = logexp(cov_data$exposure)),
## data = cov_data)
##
## Coefficients:
## (Intercept) pforb
## 3.0915 -0.2768
##
## Degrees of Freedom: 616 Total (i.e. Null); 615 Residual
## Null Deviance: 730.3
## Residual Deviance: 749.8 AIC: 753.8##
## Model selection based on AICc:
##
## K AICc Delta_AICc AICcWt Cum.Wt LL
## height 2 740.08 0.00 0.99 0.99 -368.03
## base model 1 752.61 12.52 0.00 1.00 -375.30
## robel 2 753.57 13.49 0.00 1.00 -374.78
## pgrass 2 753.63 13.55 0.00 1.00 -374.81
## pforb 2 753.80 13.71 0.00 1.00 -374.89##
## Call:
## glm(formula = survoutcome ~ height1, family = binomial(link = logexp(cov_data$exposure)),
## data = cov_data)
##
## Coefficients:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 1.858267 0.299036 6.214 5.16e-10 ***
## height1 0.011491 0.003048 3.770 0.000163 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## (Dispersion parameter for binomial family taken to be 1)
##
## Null deviance: 730.26 on 616 degrees of freedom
## Residual deviance: 736.06 on 615 degrees of freedom
## AIC: 740.06
##
## Number of Fisher Scoring iterations: 5Summary of our best model shows anestimated change in the log odds (1.0115) of the response variable for a one-unit increase in height, meaning survival increases and vegetation height increases.
