Imputation Illustration

library(ColeridgeInitiative)
library(tidyverse)
library(dbplyr)
library(zoo) ## need to install

options(java.parameters = c("-XX:+UseConcMarkSweepGC", "-Xmx16000m"))
gc()

con <- adrf_redshift(usertype = "training")

select
        ncm.*,
        fpuw.year_quarter_key,
        fpuw.ui_quarterly_wages
from tr_state_impact_ada_training.nb_analysis_matched ncm
left join tr_state_impact_ada_training.fact_person_ui_wage fpuw
    on ncm.person_key = fpuw.person_key
    and (fpuw.year_quarter_key >=27 and fpuw.year_quarter_key <= 39 or fpuw.ui_quarterly_wages is null)
order by ncm.person_key, fpuw.year_quarter_key 

Releveling a factor reference category

table(data_did$eth_recode)

data_did <- data_did |>
  mutate(eth_recode = fct_relevel(eth_recode, "Wht"))

table(data_did$eth_recode)

Imputation examples

group wise modal imputation for categorical variables

must install the statip package

#install.packages('statip')

# Group-wise modal imputation example
data_did_modal <- data_did |> 
  group_by(eth_recode) |> 
  mutate(
    gender = ifelse(is.na(gender),
                    as.character(statip::mfv(gender, na_rm = TRUE)),
                    as.character(gender))
  ) %>%
  ungroup()

group wise median imputation for continuous variables

#install.packages('statip')

# Group-wise modal imputation example
data_did_median <- data_did |> 
  group_by(eth_recode) |> 
  mutate(
    quartery_wages_imp = if_else(is.na(ui_quarterly_wages)==T,
                    median(data_did$ui_quarterly_wages, na.rm=T),
                    ui_quarterly_wages)
  ) %>%
  ungroup()

Nearest neighbor imputation

specify the variables you want imputed in the “variable” list, and the variables you want to use to measure the closeness of a given observation in the “dist_var” list

# kNN imputation example

#install.packages('VIM')
df_knn <- VIM::kNN(data_did, k = 3,
                   variable = c("gender", "age_cat", "edu_cat", "wages_q37", "wages_q41", "ui_quarterly_wages"),
                   dist_var = c("gender", "age_cat", "edu_cat"),
                   imp_var = FALSE)

df_knn

summary(data_did$ui_quarterly_wages)

summary(df_knn$ui_quarterly_wages)