DeclareDesign

Declare and diagnose your research design

DeclareDesign is a set of software tools to plan, implement, analyze, and communicate about empirical research

Get Started

Software

Read Book

MIDA framework for describing research designs

The MIDA framework describes the four elements of any empirical research design:

Model: the worlds you consider
Inquiry: the question you ask
Data strategy: sampling, treatment assignment, and measurement procedures
Answer strategy: estimation, testing, interpretation, and visualization procedures

Read Chapter 5 of Research Design in the Social Sciences to learn how these four research design elements connect to reality and simulations that can be used to plan and improve research designs.

Declare-Diagnose-Redesign algorithm for designing research

Declare designs in code following the MIDA framework.
Diagnose declared designs through Monte Carlo simulation to learn their properties, such as bias and power.
Redesign data and answer strategy features to optimize designs under logistical, financial, and ethical constraints.

Here is an illustration of using DeclareDesign for a two-arm randomized trial:

library(DeclareDesign)

sample_size <- 100

# Declare a two-arm trial in code
two_arm_trial <-
  declare_model(N = sample_size,
                U = rnorm(N),
                potential_outcomes(Y ~ 0.2 * Z + U)) +
  declare_inquiry(ATE = mean(Y_Z_1 - Y_Z_0)) +
  declare_assignment(Z = complete_ra(N, prob = 0.5)) +
  declare_measurement(Y = reveal_outcomes(Y ~ Z)) +
  declare_estimator(Y ~ Z, inquiry = "ATE")

# Draw a simulated dataset 
draw_data(two_arm_trial)

ID	U	Y_Z_0	Y_Z_1	Y
001	0.08	0.08	0.28	0.08
002	1.06	1.06	1.26	1.06
003	1.47	1.47	1.67	1.47

# Obtain a simulated estimate and estimand
run_design(two_arm_trial)

estimate	std.error	p.value	inquiry	estimand
0.3	0.19	0.11	ATE	0.2

# Simulate the research design 500 times and 
#   summarize the simulations
diagnosis <- diagnose_design(two_arm_trial, sims = 500)
tidy(diagnosis)

diagnosand	estimate	std.error
mean_estimand	0.200	0.000
mean_estimate	0.209	0.008
bias	0.009	0.008
sd_estimate	0.190	0.005
rmse	0.190	0.005
power	0.162	0.014
coverage	0.964	0.008

library(ggplot2)

# Visualize simulated sampling distribution
ggplot(data = get_simulations(diagnosis), 
       aes(x = estimate)) +
  geom_histogram()

# Redesign over sample size and calculate power
diagnosis <-
  two_arm_trial |>
  redesign(sample_size = c(250, 500, 750, 1000, 1250)) |>
  diagnose_designs() |>
  tidy() |>
  filter(diagnosand == "power")

# Visualize power curve over sample sizes
ggplot(diagnosis, aes(sample_size, estimate)) + 
  geom_point() +
  geom_line()

Library of common research designs

The MIDA framework can accomodate observational and experimental, descriptive and causal, qualitative and quantitative research designs. Part III of Research Design in the Social Sciences illustrates the framework for these common designs.

MIDA framework for describing research designs

Declare-Diagnose-Redesign algorithm for designing research

Library of common research designs

Observational designs for descriptive inference

Observational designs for causal inference

Experimental designs for descriptive inference

Experimental designs for causal inference

Complex designs