Causal Model Details

  1. Home
  2. Approach
  3. Causal Model Details
Zoom In
Zoom In
Zoom In
Zoom In

Causal Model information

Datasets Used

Number of Years in Dataset

Data Science Techniques Applied

Subject Matter Experts Involved

The Use of Judge Name as a Segment in our Predictive and Causal Models

Due to direct feedback from our subject matter experts, the predictive and causal modelling approaches have been run with and without judge information for comparison reasons.

We chose counties with enough judge information for our modelling purposes and noted that many of the judges names we misspelled, shortened or abbreviated. We have applied "fuzzy" search logic, which you can find in this notebook, to group these variations in judges names together.

  • Number of judges used: 3 (Andrews, Quesada, Peters)
  • Link to judge reference: Florida Sentencing Details by Judge
  • Challenges: We believe that OCR (or similar) was used to scan court documents as the team encountered several variations of judges names. Fuzzy logic (levenstein distance) was used to collate the judge names and sanity checks were made to ensure accuracy.

    • The Use of Segments for the Predictive and Causal Models

    To narrow the scope for our MVP we segmented the three circuits and three crime types listed below

    • Circuits: CIRCUIT 06 - CLEARWATER, CIRCUIT 11 - MIAMI, CIRCUIT 17 - FT. LAUDERDALE
    • Crime Type: Drugs, Robbery and Traffic Control


    Causal Modelling:

    For causal modelling we focused on robustness and applicability through a Quasi-experimental design using matching methods.

  • Approach: Clustering based matching (unsupervised)
    • Generate data clusters based on feature proximity determined through k-means and cosine similarity on gender, age and total sentencing points
    • Clusters are generated using distance based clustering (Good explainability)
    • Clustering based matching is less situational than many other matching methods
  • Model Inputs:
    • Age: 20-86 years of age
    • Gender: M/F
    • Total Points: Number of points for the defendent
  • Model Outputs:Sentence Difference in Days, Confidence interval and p-value for the Difference
  • Assumptions:
    • Stable unit treatment value assumption (SUTVA): Treatment of one unit does not affect the potential outcome of other units
    • Ignorability: Treatment assignment is independent of the potential outcome
  • Evaluation:
    • Clustering:
      • Elbow method
      • Intra cluster variance
    • Sensitivity Analysis
    • Confidence Interval and Significance test evaluation
  • Trade-offs:
    • Bias and Variance
    • Matching Explainability
    • Computation Efficiency
  • Challenges: The method depends on enough data points for each cluster and may be hard to extend to scenarios with less data points to prove meaningful results