README.md

In aterui/avpc: Average Predictive Comparison

R package avpc

Overview

An R package avpc provides a function apcomp to estimate average predictive comparisons. The average predictive is a predicted change in an orginal-scale response variable with a unit change of input variable (= explanatory variable) of interest (see Gelman and Pardoe, 2007). The average predictive comparison is identical to regression coefficients when the model is linear (i.e., link function is identity) and contains no interaction terms. However, when the model contains interactions or non-linear variable transformation (e.g., logit, log), the average predictive comparison is different from regression coefficients and gives deeper insights into how the input variable of interest affects the response variable at the original scale while properly controlling for influences of other variables.

Installation

Use the following script to install the package:

remotes::install_github("aterui/avpc")

Usage

Function apcomp uses model object m and input variable of interest u as minimal arguments. Users can specify other input variables v manually but this is NOT recommended.

# simulated data
set.seed(123)

n_sample <- 100
x1 <- rnorm(n_sample)
x2 <- rnorm(n_sample)
x3 <- sample(letters[1:3], size = n_sample, replace = TRUE)
mat <- model.matrix(model.frame(~ x1 + x2 + x3))
v_b <- runif(n = ncol(mat), min = 0, max = 1)

y <- mat %*% v_b + rnorm(n_sample)
m <- lm(y ~ x1 + x2 + x3)

# average predictive comparison
library(avpc)
apcomp(m, u = "x1")

## link function: identity - the inverse function was used to estimate an average predictive comparison

## $estimate
## [1] 0.6858253
## 
## $sim_estimate
## [1] 0.6827217
## 
## $sim_se
## [1] 0.1074857
## 
## $df_uv
## # A tibble: 10,000 x 13
##    sq_distance row_id col_id weight      u1     u2  sign   x2_v1 x3b_v1 x3c_v1
##          <dbl>  <dbl>  <dbl>  <dbl>   <dbl>  <dbl> <dbl>   <dbl>  <dbl>  <dbl>
##  1       0          1      1  1     -0.560  -0.560     1 -0.710       0      1
##  2       1.03       2      1  0.493 -0.230  -0.560    -1  0.257       0      1
##  3       0.236      3      1  0.809  1.56   -0.560    -1 -0.247       0      1
##  4       6.09       4      1  0.141  0.0705 -0.560    -1 -0.348       0      0
##  5       6.52       5      1  0.133  0.129  -0.560    -1 -0.952       0      0
##  6       0.486      6      1  0.673  1.72   -0.560    -1 -0.0450      0      1
##  7       6.01       7      1  0.143  0.461  -0.560    -1 -0.785       1      0
##  8       1.01       8      1  0.498 -1.27   -0.560     1 -1.67        0      1
##  9       5.94       9      1  0.144 -0.687  -0.560     1 -0.380       1      0
## 10       7.81      10      1  0.114 -0.446  -0.560    -1  0.919       0      0
## # ... with 9,990 more rows, and 3 more variables: x2_v2 <dbl>, x3b_v2 <dbl>,
## #   x3c_v2 <dbl>
## 
## $df_u1v1
## # A tibble: 10,000 x 5
##    `(Intercept)`      x1      x2   x3b   x3c
##            <dbl>   <dbl>   <dbl> <dbl> <dbl>
##  1             1 -0.560  -0.710      0     1
##  2             1 -0.230   0.257      0     1
##  3             1  1.56   -0.247      0     1
##  4             1  0.0705 -0.348      0     0
##  5             1  0.129  -0.952      0     0
##  6             1  1.72   -0.0450     0     1
##  7             1  0.461  -0.785      1     0
##  8             1 -1.27   -1.67       0     1
##  9             1 -0.687  -0.380      1     0
## 10             1 -0.446   0.919      0     0
## # ... with 9,990 more rows
## 
## $df_u2v1
## # A tibble: 10,000 x 5
##    `(Intercept)`     x1      x2   x3b   x3c
##            <dbl>  <dbl>   <dbl> <dbl> <dbl>
##  1             1 -0.560 -0.710      0     1
##  2             1 -0.560  0.257      0     1
##  3             1 -0.560 -0.247      0     1
##  4             1 -0.560 -0.348      0     0
##  5             1 -0.560 -0.952      0     0
##  6             1 -0.560 -0.0450     0     1
##  7             1 -0.560 -0.785      1     0
##  8             1 -0.560 -1.67       0     1
##  9             1 -0.560 -0.380      1     0
## 10             1 -0.560  0.919      0     0
## # ... with 9,990 more rows
## 
## $interaction_term
## NULL

aterui/avpc documentation built on March 16, 2021, 12:05 a.m.