Skip to contents

Imputing plausible data for missing values

Source: R/imputeMissing.R
imputeMissing.Rd

Given an estimated model from any of mirt's model fitting functions and an estimate of the latent trait, impute plausible missing data values. Returns the original data in a data.frame without any NA values. If a list of Theta values is supplied then a list of complete datasets is returned instead.

Usage

imputeMissing(x, Theta, warn = TRUE, ...)

Arguments

x

an estimated model x from the mirt package

Theta

a matrix containing the estimates of the latent trait scores (e.g., via fscores)

warn

logical; print warning messages?

...

additional arguments to pass

References

Chalmers, R., P. (2012). mirt: A Multidimensional Item Response Theory Package for the R Environment. Journal of Statistical Software, 48(6), 1-29. doi:10.18637/jss.v048.i06

Author

Phil Chalmers rphilip.chalmers@gmail.com

Examples

# \donttest{
dat <- expand.table(LSAT7)
(original <- mirt(dat, 1))
#> 
Iteration: 1, Log-Lik: -2668.786, Max-Change: 0.18243
Iteration: 2, Log-Lik: -2663.691, Max-Change: 0.13637
Iteration: 3, Log-Lik: -2661.454, Max-Change: 0.10231
Iteration: 4, Log-Lik: -2659.430, Max-Change: 0.04181
Iteration: 5, Log-Lik: -2659.241, Max-Change: 0.03417
Iteration: 6, Log-Lik: -2659.113, Max-Change: 0.02911
Iteration: 7, Log-Lik: -2658.812, Max-Change: 0.00456
Iteration: 8, Log-Lik: -2658.809, Max-Change: 0.00363
Iteration: 9, Log-Lik: -2658.808, Max-Change: 0.00273
Iteration: 10, Log-Lik: -2658.806, Max-Change: 0.00144
Iteration: 11, Log-Lik: -2658.806, Max-Change: 0.00118
Iteration: 12, Log-Lik: -2658.806, Max-Change: 0.00101
Iteration: 13, Log-Lik: -2658.805, Max-Change: 0.00042
Iteration: 14, Log-Lik: -2658.805, Max-Change: 0.00025
Iteration: 15, Log-Lik: -2658.805, Max-Change: 0.00026
Iteration: 16, Log-Lik: -2658.805, Max-Change: 0.00023
Iteration: 17, Log-Lik: -2658.805, Max-Change: 0.00023
Iteration: 18, Log-Lik: -2658.805, Max-Change: 0.00021
Iteration: 19, Log-Lik: -2658.805, Max-Change: 0.00019
Iteration: 20, Log-Lik: -2658.805, Max-Change: 0.00017
Iteration: 21, Log-Lik: -2658.805, Max-Change: 0.00017
Iteration: 22, Log-Lik: -2658.805, Max-Change: 0.00015
Iteration: 23, Log-Lik: -2658.805, Max-Change: 0.00015
Iteration: 24, Log-Lik: -2658.805, Max-Change: 0.00013
Iteration: 25, Log-Lik: -2658.805, Max-Change: 0.00013
Iteration: 26, Log-Lik: -2658.805, Max-Change: 0.00011
Iteration: 27, Log-Lik: -2658.805, Max-Change: 0.00011
Iteration: 28, Log-Lik: -2658.805, Max-Change: 0.00010
#> 
#> Call:
#> mirt(data = dat, model = 1)
#> 
#> Full-information item factor analysis with 1 factor(s).
#> Converged within 1e-04 tolerance after 28 EM iterations.
#> mirt version: 1.44.4 
#> M-step optimizer: BFGS 
#> EM acceleration: Ramsay 
#> Number of rectangular quadrature: 61
#> Latent density type: Gaussian 
#> 
#> Log-likelihood = -2658.805
#> Estimated parameters: 10 
#> AIC = 5337.61
#> BIC = 5386.688; SABIC = 5354.927
#> G2 (21) = 31.7, p = 0.0628
#> RMSEA = 0.023, CFI = NaN, TLI = NaN
NAperson <- sample(1:nrow(dat), 20, replace = TRUE)
NAitem <- sample(1:ncol(dat), 20, replace = TRUE)
for(i in 1:20)
    dat[NAperson[i], NAitem[i]] <- NA
(mod <- mirt(dat, 1))
#> 
Iteration: 1, Log-Lik: -2657.602, Max-Change: 0.18162
Iteration: 2, Log-Lik: -2652.461, Max-Change: 0.13527
Iteration: 3, Log-Lik: -2650.266, Max-Change: 0.10096
Iteration: 4, Log-Lik: -2648.362, Max-Change: 0.04054
Iteration: 5, Log-Lik: -2648.193, Max-Change: 0.03376
Iteration: 6, Log-Lik: -2648.075, Max-Change: 0.02800
Iteration: 7, Log-Lik: -2647.814, Max-Change: 0.00709
Iteration: 8, Log-Lik: -2647.809, Max-Change: 0.00524
Iteration: 9, Log-Lik: -2647.806, Max-Change: 0.00470
Iteration: 10, Log-Lik: -2647.802, Max-Change: 0.00316
Iteration: 11, Log-Lik: -2647.800, Max-Change: 0.00233
Iteration: 12, Log-Lik: -2647.800, Max-Change: 0.00201
Iteration: 13, Log-Lik: -2647.799, Max-Change: 0.00169
Iteration: 14, Log-Lik: -2647.799, Max-Change: 0.00131
Iteration: 15, Log-Lik: -2647.799, Max-Change: 0.00120
Iteration: 16, Log-Lik: -2647.798, Max-Change: 0.00041
Iteration: 17, Log-Lik: -2647.798, Max-Change: 0.00082
Iteration: 18, Log-Lik: -2647.798, Max-Change: 0.00054
Iteration: 19, Log-Lik: -2647.798, Max-Change: 0.00016
Iteration: 20, Log-Lik: -2647.798, Max-Change: 0.00012
Iteration: 21, Log-Lik: -2647.798, Max-Change: 0.00011
Iteration: 22, Log-Lik: -2647.798, Max-Change: 0.00011
Iteration: 23, Log-Lik: -2647.798, Max-Change: 0.00010
Iteration: 24, Log-Lik: -2647.798, Max-Change: 0.00010
#> 
#> Call:
#> mirt(data = dat, model = 1)
#> 
#> Full-information item factor analysis with 1 factor(s).
#> Converged within 1e-04 tolerance after 24 EM iterations.
#> mirt version: 1.44.4 
#> M-step optimizer: BFGS 
#> EM acceleration: Ramsay 
#> Number of rectangular quadrature: 61
#> Latent density type: Gaussian 
#> 
#> Log-likelihood = -2647.798
#> Estimated parameters: 10 
#> AIC = 5315.596
#> BIC = 5364.674; SABIC = 5332.913
#> 
scores <- fscores(mod, method = 'MAP')

# re-estimate imputed dataset (good to do this multiple times and average over)
fulldata <- imputeMissing(mod, scores)
(fullmod <- mirt(fulldata, 1))
#> 
Iteration: 1, Log-Lik: -2664.434, Max-Change: 0.18487
Iteration: 2, Log-Lik: -2659.219, Max-Change: 0.13796
Iteration: 3, Log-Lik: -2656.941, Max-Change: 0.10327
Iteration: 4, Log-Lik: -2654.915, Max-Change: 0.04137
Iteration: 5, Log-Lik: -2654.730, Max-Change: 0.03441
Iteration: 6, Log-Lik: -2654.603, Max-Change: 0.02949
Iteration: 7, Log-Lik: -2654.310, Max-Change: 0.00695
Iteration: 8, Log-Lik: -2654.304, Max-Change: 0.00464
Iteration: 9, Log-Lik: -2654.302, Max-Change: 0.00425
Iteration: 10, Log-Lik: -2654.299, Max-Change: 0.00330
Iteration: 11, Log-Lik: -2654.298, Max-Change: 0.00236
Iteration: 12, Log-Lik: -2654.298, Max-Change: 0.00204
Iteration: 13, Log-Lik: -2654.297, Max-Change: 0.00171
Iteration: 14, Log-Lik: -2654.297, Max-Change: 0.00128
Iteration: 15, Log-Lik: -2654.297, Max-Change: 0.00115
Iteration: 16, Log-Lik: -2654.297, Max-Change: 0.00049
Iteration: 17, Log-Lik: -2654.297, Max-Change: 0.00031
Iteration: 18, Log-Lik: -2654.297, Max-Change: 0.00024
Iteration: 19, Log-Lik: -2654.296, Max-Change: 0.00020
Iteration: 20, Log-Lik: -2654.296, Max-Change: 0.00094
Iteration: 21, Log-Lik: -2654.296, Max-Change: 0.00014
Iteration: 22, Log-Lik: -2654.296, Max-Change: 0.00014
Iteration: 23, Log-Lik: -2654.296, Max-Change: 0.00062
Iteration: 24, Log-Lik: -2654.296, Max-Change: 0.00010
Iteration: 25, Log-Lik: -2654.296, Max-Change: 0.00010
#> 
#> Call:
#> mirt(data = fulldata, model = 1)
#> 
#> Full-information item factor analysis with 1 factor(s).
#> Converged within 1e-04 tolerance after 25 EM iterations.
#> mirt version: 1.44.4 
#> M-step optimizer: BFGS 
#> EM acceleration: Ramsay 
#> Number of rectangular quadrature: 61
#> Latent density type: Gaussian 
#> 
#> Log-likelihood = -2654.296
#> Estimated parameters: 10 
#> AIC = 5328.593
#> BIC = 5377.67; SABIC = 5345.91
#> G2 (21) = 31.58, p = 0.0645
#> RMSEA = 0.022, CFI = NaN, TLI = NaN

# with multipleGroup
set.seed(1)
group <- sample(c('group1', 'group2'), 1000, TRUE)
mod2 <- multipleGroup(dat, 1, group, TOL=1e-2)
#> 
Iteration: 1, Log-Lik: -2657.602, Max-Change: 0.19467
Iteration: 2, Log-Lik: -2648.992, Max-Change: 0.14635
Iteration: 3, Log-Lik: -2646.059, Max-Change: 0.11272
Iteration: 4, Log-Lik: -2643.584, Max-Change: 0.04382
Iteration: 5, Log-Lik: -2643.446, Max-Change: 0.03509
Iteration: 6, Log-Lik: -2643.372, Max-Change: 0.02823
Iteration: 7, Log-Lik: -2643.267, Max-Change: 0.00963
fs <- fscores(mod2)
fulldata2 <- imputeMissing(mod2, fs)

# }