1 ### EXAMPLE 1: demonstrates how measurement error can lead to a type sign error in a covariate
2 ### What kind of data invalidates fong + tyler?
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",
20 source("simulation_base.R")
23 ### we want to estimate x -> y; x is MAR
24 ### we have x -> k; k -> w; x -> w is used to predict x via the model w.
25 ### A realistic scenario is that we have an NLP model predicting something like "racial harassment" in social media comments
26 ### The labels x are binary, but the model provides a continuous predictor
29 #### how much power do we get from the model in the first place? (sweeping N and m)
32 ## one way to do it is by adding correlation to x.obs and y that isn't in w.
33 ## in other words, the model is missing an important feature of x.obs that's related to y.
34 simulate_data <- function(N, m, B0, Bxy, Bgy, Bkx, Bgx, seed, xy.explained.variance=0.01, u.explained.variance=0.1){
37 ## the true value of x
39 g <- rbinom(N, 1, 0.5)
41 # make w and y dependent
44 xprime <- Bgx * g + rnorm(N,0,1)
46 k <- Bkx*xprime + rnorm(N,0,1.5) + 1.1*Bkx*u
48 x <- as.integer(logistic(scale(xprime)) > 0.5)
50 y <- Bxy * x + Bgy * g + B0 + u + rnorm(N, 0, 1)
52 df <- data.table(x=x,k=k,y=y,g=g)
54 w.model <- glm(x ~ k,df, family=binomial(link='logit'))
57 df <- df[sample(nrow(df), m), x.obs := x]
59 df <- df[, x.obs := x]
64 w <- predict(w.model, df) + rnorm(N, 0, 1)
67 df[,':='(w=w, w_pred = as.integer(w>0.5),u=u)]
71 schennach <- function(df){
73 fwx <- glm(x.obs~w, df, family=binomial(link='logit'))
74 df[,xstar_pred := predict(fwx, df)]
75 gxt <- lm(y ~ xstar_pred+g, df)
80 parser <- arg_parser("Simulate data and fit corrected models")
81 parser <- add_argument(parser, "--N", default=5000, help="number of observations of w")
82 parser <- add_argument(parser, "--m", default=200, help="m the number of ground truth observations")
83 parser <- add_argument(parser, "--seed", default=432, help='seed for the rng')
84 parser <- add_argument(parser, "--outfile", help='output file', default='example_2.feather')
85 args <- parse_args(parser)
94 outline <- run_simulation(simulate_data(args$N, args$m, B0, Bxy, Bgy, Bkx, Bgx, args$seed)
95 ,list('N'=args$N,'m'=args$m,'B0'=B0,'Bxy'=Bxy,'Bgy'=Bgy, 'Bkx'=Bkx, 'Bgx'=Bgx, 'seed'=args$seed))
97 outfile_lock <- lock(paste0(args$outfile, '_lock'),exclusive=TRUE)
98 if(file.exists(args$outfile)){
99 logdata <- read_feather(args$outfile)
100 logdata <- rbind(logdata,as.data.table(outline))
102 logdata <- as.data.table(outline)
106 write_feather(logdata, args$outfile)