opossum 0.0.6

Simulated Data Generating Process

Simulated Data Generating Process

Task: Simulate realistic data set with known treatment effects

Properties of the data generator:

• High dimensional and non-linear data
• flexible
• user friendly

Motivation

• Provide high quality data for research

• Useful for you to test your models

Model Equations

Y - Outcome Variable

- True treatment effect

D - Treatment Dummy

- Covariates

Treatment Assignment

Random

Choose probability

Conditioned covariates

Weight vector

Create assignment vector

Treatment effect

Options 1/2: Constant, positive and negative

Options 3/4: Continuous heterogeneous effect, positive and negative

Option 5: No treatment effect

Composition of dependent variable

Non-linearity:

Option 1: Continuous

Option 2: Binary

Application of module/package

from opossum import UserInterface

u = UserInterface(N = 10000,k = 10, seed = 12)

u.generate_treatment(random_assignment = True,
                     assignment_prob = 0.5,
                     constant_pos = True,
                     constant_neg = False,
                     heterogeneous_pos = False,
                     heterogeneous_neg = False,
                     no_treatment = False,
                     treatment_option_weights = [0, 0, 0.7, 0.1, 0.2], # treatment_option_weights default: None 
                     # [constant_pos,constant_neg, heterogeneous_pos, heterogeneous_neg, no effect]
                     intensity = 5)

y, X, assignment_vector, treatment_effect = u.output_data(binary=False)

Correlation Matrix of X

%matplotlib notebook
u.plot_covariates_correlation()

<IPython.core.display.Javascript object>

Distribution of propensity scores according to treatment assignment

Treatment effect options

Customized treatment distribution

Outputs depending on treatment assignment