Correlation/Regression Modeling Steps

Below is an overview of the steps in regression modeling which seeks to find a correlation between predictor variables and a response variable.

1. Defining the Question

• Define the requirements of the model including the response variables.
• Research and define what the process should respond to, the potential predictor variables.

2. Collect and/or Find Data

• Sample Data
• Predictor Variables

3. Qualify Data

• All Data
  • Map it
  • Research collection methods, metadata
  • Visit the site if possible
• Sample Data
  • Response variable
  • Coordinate accuracy
• Predictor Variable Data

4. Data Exploration

Looking for: relationships between the response variable and covariates

• Response variable
  • Occurrences
    • Histogram against each covariate, compare with histograms of entire sample area
      • If there is a difference, you may be able to model it, otherwise you can't
    • Convert to density (continuous)? Kernel density functions
    • Convert to "count"
      • Convert to a raster using a "count" operation, then convert back to points
  • Binary (presence/absence)

4. Prepare the Data sets

• How will the model be validated?

• • Test and training data sets
• Three options:
  • Use predictors that were collected with samples
  • Use continuous predictors
  • Use a combination of predictors collected with samples and continuous predictors
• Gridding data
• Eliminating duplicates

5. Define Modeling Approaches

• Type of response variable:
  • Occurrences (i.e. just "1's")
  • Binary (e.g. presence/absence, alive/dead)
  • Categorical
  • Continuous
• Type of covariates/predictors
  • Combination of continuous and categorical
• Residuals:
  • Constant variance?
  • Correcting variance (transforming data)
• Shape of the expected response:
  • Linear
  • Logistic: Presence/Absence, True/False
  • Exponential
  • Logarithmic
  • Complex

6. Run Models

• Bootstrapping
• Jackknifing
• Optimizing parameters with Monte Carlo methods

7. Validating and Selecting Models

• AIC, AICc, BIC
• AUC
• Deviance explained

8. Estimating Uncertainty

• Computing precision & accuracy
• Jiggling
• RMSE
• etc.

9. Documenting Results

• Types of maps:
  • Gridded
  • Interpolated
  • Continuous
• Uncertainty maps
• Caveats
• Significant Digits