DeclareDesign is a system for describing research designs in code and simulating them in order to understand their properties. Because DeclareDesign employs a consistent grammar of designs, you can focus on the intellectually challenging part – designing good research studies – without having to code up simulations from scratch. For more, see declaredesign.org.
To install the latest stable release of DeclareDesign, please ensure that you are running version 3.5 or later of R and run the following code:
install.packages("DeclareDesign")Designs are declared by adding together design elements. Here’s a minimal example that describes a 100 unit randomized controlled trial with a binary outcome. Half the units are assigned to treatment and the remainder to control. The true value of the average treatment effect is 0.05 and it will be estimated with the difference-in-means estimator. The diagnosis shows that the study is unbiased but underpowered.
library(DeclareDesign)
design <-
declare_model(
N = 100,
potential_outcomes(Y ~ rbinom(N, size = 1, prob = 0.5 + 0.05 * Z))
) +
declare_inquiry(ATE = 0.05) +
declare_assignment(Z = complete_ra(N, m = 50)) +
declare_measurement(Y = reveal_outcomes(Y ~ Z)) +
declare_estimator(Y ~ Z, .method = lm_robust, inquiry = "ATE")
diagnosands <-
declare_diagnosands(bias = mean(estimate - estimand),
power = mean(p.value <= 0.05))
diagnosis <- diagnose_design(design, diagnosands = diagnosands)diagnosis| Inquiry | Estimator | Outcome | Bias | SE(Bias) | Power | SE(Power) | n sims |
|---|---|---|---|---|---|---|---|
| ATE | estimator | Y | -0.004 | 0.004 | 0.076 | 0.01 | 500 |
The core DeclareDesign package relies on four companion packages, each of which is useful in its own right.
- randomizr: Easy to use tools for common forms of random assignment and sampling.
- fabricatr: Imagine your data before you collect it.
- estimatr: Fast estimators for social scientists.
- DesignLibrary: Templates to quickly adopt and adapt common research designs.
-
To get started, have a look at this vignette on the idea behind DeclareDesign, which covers the main functionality of the software.
-
For an explanation of the philosophy behind DeclareDesign, examples in code and words of declaring and diagnosing common research designs in the social sciences, as well as examples of how to incorporate DeclareDesign into your own research, see the book Research Design in the Social Sciences (Blair, Coppock, Humphreys, 2023).
Each of these declare_*() functions returns a function.
declare_model()(describes dimensions and distributions over the variables, including potential outcomes)declare_inquiry()(takes variables in the model and calculates estimand value)declare_sampling()(takes a population and selects a sample)declare_assignment()(takes a population or sample and adds treatment assignments)declare_measurement()(takes data and adds measured values)declare_estimator()(takes data produced by sampling, assignment, and measurement and returns estimates linked to inquiries)declare_test()(takes data produced by sampling, assignment, and measurement and returns the result of a test)
To declare a design, connect the components of your design with the + operator.
Once you have declared your design, there are four core post-design-declaration commands used to modify or diagnose your design:
diagnose_design()(takes a design and returns simulations and diagnosis)draw_data()(takes a design and returns a single draw of the data)draw_estimates()(takes a design and returns a single simulation of estimates)draw_estimands()(takes a design and returns a single simulation of estimands)
A few other features:
- A designer is a function that takes parameters (e.g.,
N) and returns a design.expand_design()is a function of a designer and parameters that return a design. - You can change the diagnosands with
declare_diagnosands().
This project was generously supported by a grant from the Laura and John Arnold Foundation and seed funding from EGAP.
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent
