regressionComparison: Estimate regression models for comparison

Description Usage Arguments Details Value Author(s) See Also Examples

View source: R/regressionComparison.R

Description

Compare the overall results varying on which respondents are included in the model based on attention.

Usage

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
regressionComparison(
  dataframe = NULL,
  formula = NULL,
  plot_treatment = NULL,
  plot_interact_x = NULL,
  similarity_measures = NULL,
  k = 3,
  model_type = NULL,
  up_down_weight = "down",
  user_seed = 5,
  n_sims = 10000,
  correct_vec = NULL,
  display_plot = F,
  stable_x = T,
  plot_path = NULL,
  print_regs = F,
  return_data = F,
  plotDifferences = F
)

Arguments

dataframe

Dataframe from which you will select the outcome, predictors, and weights for the regression model.

formula

Symbolic representation of the model to be estimated. This is written in "typical" R language (i.e. y ~ x1 + x2), such that y is the outcome variable and x1 and x2 are the predictors.

plot_treatment

Character indicating vector in dataframe that's the treatment. Default is NULL, so users must specify for the function to work.

plot_interact_x

Character indicating vector in dataframe that treatment should be plotted with. If there is an interaction in the formula, plot_interact_x is likely that variable. Default is NULL, so users must specify for the function to work.

similarity_measures

Vector(s) from dataframe that contains the similarity measures to be used as weights.

k

The penalty that you want to set for down-weighting inattentive respondents. Lower levels of k down-weight low attention participants more severely.

model_type

Statistical model to estimate. Currently support OLS and logistic ("ls", "logit").

up_down_weight

Do you want to up-weight or down-weight?

user_seed

Since we make random pulls from the multivariate normal distibution, set seed to get same results again. Default=5.

n_sims

Since we make random pulls from the multivariate normal distibution, set seed to get same results again. Default=10000.

correct_vec

Character indicating vector in dataframe that indicates whether a respondent answered "correctly" as determined by a human coder. Default is NULL, so if a user doesn't include anything function will automatically set a threshold of "correctness" based on a respondent's average similarity (i.e. those respondents that score below say 0.1 will not be included in the "list-wise deletion" sample.

stable_x

Indicates whether plot_interact_x is continuous (stable_x=FALSE) or a factor (stable_x=TRUE). Currently the package only supports TRUE.

plot_path

If user wants to save figure, please provide a character vector for the file path in which the plot should be download. User must decide extension (pdf, jpg, png) in file path.

print_regs

Return table of estimated regression coefficients and fit statistics formatted for LaTeX using texreg(). Default=FALSE.

return_data

Do you want the data that's used to construct the plot? Default = FALSE.

plotDifferences

Do you want to see the marginal effects by model, or the differences between the models with regard to their marginal effects? Default=FALSE.

Details

This function generates three regressions with: (1) the full sample irrespective of attention (2) a subsetted sample based on list-wise deletion of respondents that "failed" manipulation check (3) weighted sample by formula outlined in Ziegler (2020, 5) for average similarity.

Value

Three regression objects are estimated and loaded to your global environment as separate Zelig objects. They are named baseModel_"outcome variable", listwiseModel_"outcome variable", and weightedModel_"outcome variable".

Author(s)

Jeffrey Ziegler (<jeffrey.ziegler[at]emory.edu>)

See Also

complierATE generateMarginalEffect

Examples

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
regressionComparison(dataframe=kaneData, 
  formula=SelectTrump ~ NewsStoryConditions*ideologyFactor3, 
  plot_treatment="NewsStoryConditions",
  plot_interact_x="ideologyFactor3",
  similarity_measures=c("jaccardDist", "cosineDist"),
  k=3, 
  model_type="logit", 
  user_seed=5,  
  n_sims=10000, 
  correct_vec="correct", 
  display_plot=T,
  plot_path=NULL, 
  print_regs=T
  )
   

zieglerjef/openEnded documentation built on Nov. 30, 2020, 2:03 p.m.