Analysis of plausible values with the survey package

survey-pv-analysis.R

# https://nces.ed.gov/surveys/piaac/datafiles.asp
# Load the survey data and prepare it for analysis ----
  piaac17_puf <- file.path(
    "C:\\Users\\benja\\Downloads\\spss\\prgusap1_puf.sav"
  ) |> haven::read_sav()
  
  piaac17_rep_svy <- svrepdesign(
    data = piaac17_puf,
    repweights = "SPFWT[1-9]{1}0{0,1}",
    weights = ~ SPFWT0,
    type = "JK2"
  )

# Repeat the analysis separately for each plausible value ----
  separate_pv_results <- withPV(
    data = piaac17_rep_svy,
    # Use 'PVLITi' as shorthand for one of the i=1,...,10
    # literacy plausible value variables
    mapping = PVLITi ~ PVLIT1 + PVLIT2 + PVLIT3 + PVLIT4 + PVLIT5 +
                       PVLIT6 + PVLIT7 + PVLIT8 + PVLIT9 + PVLIT10,
    # Within `quote()`, specify how to analyze the data with PVLITi
    action = quote(
      svymean(x = ~ PVLITi, design = piaac17_rep_svy, na.rm = TRUE)
    )
  )

# Combine the analysis results using Rubin's Rules ----
  
  combined_pv_results <- MIcombine(mi_results)
  
  coef(combined_pv_results) # Point estimate(s)
  SE(combined_pv_results) # Std. error(s)
  vcov(combined_pv_results) # Variance-covariance matrix
  confint(combined_pv_results) # Confidence interval(s)