JudicialDecisions

This package currently has two purposes (which might be refactored into separate packages at a later point):

Handling of data on judicial decisions, as issued by the German federal patent court (BPatG).
Bayesian inference for statistical models of judicial decisions, e.g. mixed membership multilevel models.

Data handling & import

JudicialDecisions.loaddata(BPatG(), dir) parses files in dir, which are expected to be .jsonl representations of decisisons made by the German federal patent court, into an array with elements of type Decision:

julia> decisions = JudicialDecisions.loaddata(BPatG(), "../data/json")
1565-element Vector{Decision}:
 1 Ni 4/99 (EU)
 1 Ni 6/99(EU)
 1 Ni 8/99
 ⋮
 7 Ni 55/19 (EP)
 7 Ni 60/19 (EP)
 5 Ni 27/19 (EP)

Printing out a single decision showcases its metadata:

julia> decisions[1]
Ruling 1 Ni 4/99 (EU) on EP0389008
Date of decision: 26 September, 2000
Decided by: 1. Senate (Hacker, K Vogel, Henkel, W Maier, van Raden)
Outcome: partially annulled

There are accessor functions for most information in a Decision, e.g.:

julia> outcome(decisions[1])
Outcome(1, "partially annulled")

Bayesian models for judicial decision making

The package contains a series of statistical model implementations for modeling aspects of the decision process. We start with a simple hierarchical Binomial model for the probability that a given senate will nullify a patent (i.e., a model of cross-senate variation):

$$ \begin{align} y_s &\sim \textrm{Binomial}(n_s, p_s), \text{ for } s=1, \ldots, S \\ p_s &= \textrm{logit}^{-1}(\alpha_s) \\ \alpha_s &\sim \textrm{Normal}(\mu, \sigma) \\ \mu &\sim \textrm{Normal}(0, 1) \\ \sigma & \sim \textrm{Exponential}(1) \\ \end{align} $$

where $S$ refers to the total number of senates.

Here is how we would use the implementation of this model in the package:

julia> problem = BinomialGroupsModel(decisions; groupfun=id ∘ senate);
julia> post = sample(problem, 1000)
DynamicHMCPosterior with 1000 samples and parameters (:αs, :μ, :σ)

julia> mean(post.αs)
8-element Vector{Float64}:
 1.1533265420437266
 1.2302096660141322
 1.2042245345786518
 1.2937742490215314
 1.1191711839066207
 1.215801205986333
 1.3576725865081927
 1.166707942601264

jfb-h / judicialdecisions.jl Goto Github PK