Package 'factorEx'

Title: Design and Analysis for Factorial Experiments
Description: Provides design-based and model-based estimators for the population average marginal component effects in general factorial experiments, including conjoint analysis. The package also implements a series of recommendations offered in de la Cuesta, Egami, and Imai (2022) <doi:10.1017/pan.2020.40>, and Egami and Imai (2019) <doi:10.1080/01621459.2018.1476246>.
Authors: Naoki Egami [aut, cre], Brandon de la Cuesta [aut], Kosuke Imai [aut]
Maintainer: Naoki Egami <[email protected]>
License: GPL-2
Version: 1.1.0
Built: 2026-05-28 08:16:06 UTC
Source: https://github.com/naoki-egami/factorex

Help Index

Decompose the difference between the pAMCEs

Description

See examples in 'model_pAMCE'.

Usage

decompose_pAMCE(out, effect_name, target_diff)

Arguments

out

An object of class "pAMCE", a result of a call to 'model_pAMCE'.
effect_name

Effect for which the function decomposes the difference in the pAMCEs. The first element should be a factor name and the second element should be a level name.
target_diff

Two target profile distributions for which the function compares the pAMCEs. If missing, the function comapres the first target profile distribution and the in-sample profile distribution.

Value

decompose_pAMCE returns 'data.frame' showing the decomposition of the difference between the pAMCEs.

Estimating the population AMCE using a design-based approach

Description

design_pAMCE implements the design-based approach to estimate the pAMCE. See de la Cuesta, Egami, and Imai (2022) for details. More examples are available at the GitHub page of factorEx.

Usage

design_pAMCE(
  formula,
  factor_name,
  data,
  pair = FALSE,
  pair_id = NULL,
  cross_int = FALSE,
  cluster_id = NULL,
  target_dist,
  target_type,
  partial_joint_name
)

Arguments

formula

Formula
factor_name

Factors for which the function estimates the pAMCEs. If not specified, the function estimates for all factors.
data

Data
pair

Whether we use a paired-choice conjoint design.
pair_id

Unique identifiers for pairs in the paired-choice conjoint design (optional).
cross_int

Include interactions across profiles. Default is FALSE.
cluster_id

Unique identifiers for computing cluster standard errors (optional).
target_dist

Target profile distributions to be used. See Examples in the GitHub page for details.
target_type

Types of target profile distributions. ‘marginal', ’partial_joint', or 'target_data'.
partial_joint_name

(when 'target_type = "partial_joint"') Names of factors representing partial joint distributions. See Examples in the GitHub page for details.

Value

design_pAMCE returns an object of pAMCE class.

  • AMCE: Estimates of the pAMCE for all factors.

  • design_weight: Weight for each observation used in the weighted difference-in-means.

  • approach: "design_based"

  • input: Input into the function.

  • ...: Values for internal use.

References

de la Cuesta, Egami, and Imai. (2022). Improving the External Validity of Conjoint Analysis: The Essential Role of Profile Distribution. Political Analysis.

Egami and Imai. (2019). Causal Interaction in Factorial Experiments: Application to Conjoint Analysis. Journal of the American Statistical Association, Vol.114, No.526 (June), pp. 529–540.

Examples

# Small example
data("OnoBurden")
OnoBurden_data_pr_s <- OnoBurden$OnoBurden_data_pr[1:500, ]
# randomization based on marginal population design
target_dist_marginal_s <- OnoBurden$target_dist_marginal[c("gender", "age")]

# design-based estimation
out_design_mar_s <-
  design_pAMCE(formula = Y ~ gender + age,
               factor_name = "gender",
               data = OnoBurden_data_pr_s,
               pair_id = OnoBurden_data_pr_s$pair_id,
               cluster_id = OnoBurden_data_pr_s$id,
               target_dist  = target_dist_marginal_s, target_type = "marginal")
summary(out_design_mar_s)

  #  Example
  data("OnoBurden")
  OnoBurden_data_pr <- OnoBurden$OnoBurden_data_pr
  # randomization based on marginal population design
  target_dist_marginal <- OnoBurden$target_dist_marginal

  # design-based estimation
  out_design_mar <-
    design_pAMCE(formula = Y ~ gender + age + family + race + experience + party + pos_security,
                 factor_name = c("gender", "age", "experience"),
                 data = OnoBurden_data_pr,
                 pair_id = OnoBurden_data_pr$pair_id,
                 cluster_id = OnoBurden_data_pr$id,
                 target_dist  = target_dist_marginal, target_type = "marginal")
 summary(out_design_mar)

Diagnose modeling assumptions for the model-based approach

Description

Diagnose modeling assumptions for the model-based approach

Usage

diagnose_pAMCE(x, factor_name)

Arguments

x

An object of class "pAMCE", a result of a call to 'model_pAMCE'.
factor_name

A factor for which the function diagnoses modeling assumptions.

Value

No return value, called for side effects

Estimating the population AMCE using a model-based approach

Description

model_pAMCE implements the model-based approach to estimate the pAMCE. See de la Cuesta, Egami, and Imai (2022) for details. More examples are available at the GitHub page of factorEx.

Usage

model_pAMCE(
  formula,
  formula_three = NULL,
  data,
  reg = TRUE,
  ord_fac,
  pair = FALSE,
  pair_id = NULL,
  cross_int = FALSE,
  cluster_id = NULL,
  target_dist,
  target_type,
  difference = FALSE,
  cv_type = "cv.1Std",
  nfolds = 5,
  boot = 100,
  seed = 1234,
  numCores = NULL
)

Arguments

formula

Formula
formula_three

Formula for three-way interactions (optional)
data

Data
reg

TRUE (regularization) or FALSE (no regularization). Default is TRUE
ord_fac

Whether we assume each factor is ordered. When not specified, we assume all of them are ordered
pair

Whether we use a paired-choice conjoint design
pair_id

Unique identifiers for pairs in the paired-choice conjoint design (optional)
cross_int

Include interactions across profiles. Default is FALSE.
cluster_id

Unique identifiers for computing cluster standard errors (optional).
target_dist

Target profile distributions to be used. This argument should be 'list'
target_type

Types of target profile distributions. 'marginal' or 'target_data'. See Examples for details.
difference

Whether we compute the differences between the multiple pAMCEs. Default is FALSE.
cv_type

(optimal only when 'reg = TRUE“) 'cv.1Std“ (stronger regularization; default) or 'cv.min' (weaker regularization).
nfolds

Number of cross validation folds. Default is 5.
boot

The number of bootstrap samples.
seed

Seed for bootstrap.
numCores

Number of cores to be used for parallel computing. If not specified, detect the number of available cores internally.

Value

model_pAMCE returns an object of pAMCE class.

  • AMCE: Estimates of the pAMCE for all factors.

  • boot_AMCE: Estimates of the pAMCE for all factors in each bootstrap sample.

  • boot_coef: Estimates of coefficients for the linear probability model in each bootstrap sample.

  • approach: "model_based"

  • input: Input into the function.

  • ...: Values for internal use.

References

de la Cuesta, Egami, and Imai. (2022). Improving the External Validity of Conjoint Analysis: The Essential Role of Profile Distribution. Political Analysis.

Egami and Imai. (2019). Causal Interaction in Factorial Experiments: Application to Conjoint Analysis. Journal of the American Statistical Association, Vol.114, No.526 (June), pp. 529–540.

Examples

# Small example
  target_dist_marginal <- OnoBurden$target_dist_marginal
  OnoBurden_data <- OnoBurden$OnoBurden_data
  OnoBurden_data_small <- OnoBurden_data[1:300, ]
  target_dist_marginal_small <- target_dist_marginal[c("gender", "race")]

  # model-based estimation without regularization
  out_model_s <-
      model_pAMCE(formula = Y ~ gender + race,
           data = OnoBurden_data_small, reg = FALSE,
           pair_id = OnoBurden_data_small$pair_id,
           cluster_id = OnoBurden_data_small$id,
           target_dist  = target_dist_marginal_small,
           target_type = "marginal")


  # model-based estimation with regularization
  out_model <-
    model_pAMCE(formula = Y ~ gender + age + family + race + experience + party + pos_security,
                 data = OnoBurden_data_small, reg = TRUE,
                 pair_id = OnoBurden_data_small$pair_id,
                 cluster_id = OnoBurden_data_small$id,
                 target_dist  = target_dist_marginal,
                 target_type = "marginal",
                 nfolds = 2,   # suggest "nfolds = 5" in practice
                 numCores = 1, # when numCores = NULL, it automatically uses all cores.
                 boot = 5)  # suggest "boot = 500" in practice
 summary(out_model, factor_name = c("gender"))

 # decompose the difference in the pAMCEs
 decompose_pAMCE(out_model, effect_name = c("gender", "Female"))

Dataset from Ono and Burden (2018)

Description

Dataset from Ono and Burden (2018)

Usage

OnoBurden

Format

A list containing the conjoint data, target profile distributions based on (1) marginals, (2) a combination of marginal and partial joint distributions, and (3) the full joint distribution.

Plot decomposition of the difference between pAMCEs

Description

See examples in 'model_pAMCE'.

Usage

plot_decompose(x, effect_name, target_diff, mar = 12)

Arguments

x

An object of class "pAMCE", a result of a call to 'model_pAMCE'.
effect_name

Effect for which the function decomposes the difference in the pAMCEs. The first element should be a factor name and the second element should be a level name.
target_diff

Two target profile distributions for which the function compares the pAMCEs. If missing, the function comapres the first target profile distribution and the in-sample profile distribution.
mar

Space on the left side of the plot. Default is 12.

Value

No return value, called for side effects

Plotting diagnostic checks

Description

Plotting diagnostic checks

Usage

plot_diagnose(
  x,
  factor_name,
  legend_pos = "topright",
  target_dist_name,
  xlim,
  mar
)

Arguments

x

An object of class "pAMCE", a result of a call to 'model_pAMCE.'
factor_name

A factor for which the function diagnoses modeling assumptions.
legend_pos

Position of the legend. Default is 'topright'.
target_dist_name

Names of the target profile distributions to be used.
xlim

Range for the x-axis.
mar

Space on the left side of the plot. Default is 12.

Value

No return value, called for side effects

Plotting the estimated population AMCEs

Description

Plotting the estimated population AMCEs

Usage

## S3 method for class 'pAMCE'
plot(
  x,
  factor_name,
  target_dist_name,
  legend_pos = "topright",
  main = "Estimated population AMCEs",
  xlim,
  mar = 12,
  diagnose = FALSE,
  ...
)

Arguments

x

An object of class "pAMCE", a result of a call to 'model_pAMCE' or 'design_pAMCE'
factor_name

Factors for which the function visualizes the pAMCEs
target_dist_name

Names of the target profile distributions to be used
legend_pos

Position of the legend. Default is 'topright'
main

Title of the plot
xlim

Range for the x-axis
mar

Space on the left side of the plot. Default is 12
diagnose

Whether we plot diagnostic checks recommended in de la Cuesta, Egami, and Imai (2019). Default is FALSE
...

Other graphical parameters

Value

No return value, called for side effects

Summarizing the estimated population AMCEs

Description

Summarizing the estimated population AMCEs

Usage

## S3 method for class 'pAMCE'
summary(object, factor_name, target_dist_name, sample = FALSE, ...)

Arguments

object

An object of class "pAMCE", a result of a call to 'model_pAMCE' or 'design_pAMCE'
factor_name

Factors for which the function visualizes the pAMCEs
target_dist_name

Names of the target profile distributions to be used
sample

Whether to print the sample AMCEs, which is estimated based on the profile distribution used for randomization
...

Other parameters

Value

No return value, called for side effects

Computing weights for estimating the population AMCE using a design-based approach

Description

Computing weights for estimating the population AMCE using a design-based approach

Usage

weights_pAMCE(
  formula,
  factor_name,
  data,
  pair,
  pair_id,
  cross_int,
  target_dist,
  target_type,
  partial_joint_name
)

Arguments

formula

Formula
factor_name

A factor for which the function computes weights
data

Data
pair

Whether we use a paired-choice conjoint design
pair_id

Unique identifiers for pairs in the paired-choice conjoint design (optional)
cross_int

Include interactions across profiles. Default is FALSE
target_dist

Target profile distributions to be used. This argument should be 'list'
target_type

Types of target profile distributions. 'marginal', 'partial_joint' or 'target_data'. See Examples for details
partial_joint_name

Names of factors representing partial joint distributions. See Examples for details.

Value

No return value, called for side effects