1 ### EXAMPLE 2_b: demonstrates how measurement error can lead to a type sign error in a covariate
2 ### This is the same as example 2, only instead of x->k we have k->x.
3 ### Even when you have a good predictor, if it's biased against a covariate you can get the wrong sign.
4 ### Even when you include the proxy variable in the regression.
5 ### But with some ground truth and multiple imputation, you can fix it.
15 library(predictionError)
16 options(amelia.parallel="no",
19 source("simulation_base.R")
22 ### we want to estimate x -> y; x is MAR
23 ### we have x -> k; k -> w; x -> w is used to predict x via the model w.
24 ### A realistic scenario is that we have an NLP model predicting something like "racial harassment" in social media comments
25 ### The labels x are binary, but the model provides a continuous predictor
28 #### how much power do we get from the model in the first place? (sweeping N and m)
31 simulate_data <- function(N, m, B0=0, Bxy=0.2, Bgy=-0.2, Bgx=0.2, y_explained_variance=0.025, gx_explained_variance=0.15, prediction_accuracy=0.73, seed=1){
33 g <- rbinom(N, 1, 0.5)
35 x.var.epsilon <- var(Bgx *g) * ((1-gx_explained_variance)/gx_explained_variance)
36 x.epsilon <- rnorm(N,sd=sqrt(x.var.epsilon))
37 xprime <- Bgx * g + x.epsilon
38 x <- as.integer(logistic(scale(xprime)) > 0.5)
40 y.var.epsilon <- (var(Bgy * g) + var(Bxy *x) + 2*cov(Bxy*x,Bgy*g)) * ((1-y_explained_variance)/y_explained_variance)
41 y.epsilon <- rnorm(N, sd = sqrt(y.var.epsilon))
42 y <- Bgy * g + Bxy * x + y.epsilon
44 df <- data.table(x=x,xprime=xprime,y=y,g=g)
47 df <- df[sample(nrow(df), m), x.obs := x]
49 df <- df[, x.obs := x]
54 df <- df[sample(1:N,(1-prediction_accuracy)*N),w_pred:=(w_pred-1)**2]
55 df <- df[,':='(w=w, w_pred = w_pred)]
59 parser <- arg_parser("Simulate data and fit corrected models")
60 parser <- add_argument(parser, "--N", default=500, help="number of observations of w")
61 parser <- add_argument(parser, "--m", default=100, help="m the number of ground truth observations")
62 parser <- add_argument(parser, "--seed", default=4321, help='seed for the rng')
63 parser <- add_argument(parser, "--outfile", help='output file', default='example_2_B.feather')
64 args <- parse_args(parser)
71 df <- simulate_data(args$N, args$m, B0, Bxy, Bgy, Bgx, seed=args$seed, y_explained_variance = 0.025, gx_explained_variance = 0.15)
72 result <- list('N'=args$N,'m'=args$m,'B0'=B0,'Bxy'=Bxy,'Bgy'=Bgy, 'Bgx'=Bgx, 'seed'=args$seed)
73 outline <- run_simulation(df, result)
75 outfile_lock <- lock(paste0(args$outfile, '_lock'),exclusive=TRUE)
76 if(file.exists(args$outfile)){
77 logdata <- read_feather(args$outfile)
78 logdata <- rbind(logdata,as.data.table(outline))
80 logdata <- as.data.table(outline)
84 write_feather(logdata, args$outfile)