| Title: | Classify Missing Data as MCAR, MAR, or MNAR |
| Version: | 1.0.1 |
| Maintainer: | Noah William Trelawny Hellen <noahhellen@gmail.com> |
| Description: | Classify missing data as missing completely at random (MCAR), missing at random (MAR), or missing not at random (MNAR). This step is required before handling missing data (e.g. mean imputation) so that bias is not introduced. See Little (1988) <doi:10.1080/01621459.1988.10478722> for the statistical rationale for the methods used. |
| License: | MIT + file LICENSE |
| URL: | https://github.com/NoahHellen/missr, https://noahhellen.github.io/missr/ |
| BugReports: | https://github.com/NoahHellen/missr/issues |
| Depends: | R (≥ 3.5) |
| Imports: | norm, tibble, lifecycle |
| Suggests: | knitr, rmarkdown, testthat (≥ 3.0.0) |
| VignetteBuilder: | knitr |
| Config/testthat/edition: | 3 |
| Encoding: | UTF-8 |
| Language: | en-GB |
| LazyData: | true |
| RoxygenNote: | 7.3.2 |
| NeedsCompilation: | no |
| Packaged: | 2025-06-04 10:43:02 UTC; noahhellen |
| Author: | Noah William Trelawny Hellen [aut, cre, cph] |
| Repository: | CRAN |
| Date/Publication: | 2025-06-04 11:20:01 UTC |
missr: Classify Missing Data as MCAR, MAR, or MNAR
Description
Classify missing data as missing completely at random (MCAR), missing at random (MAR), or missing not at random (MNAR). This step is required before handling missing data (e.g. mean imputation) so that bias is not introduced. See Little (1988) doi:10.1080/01621459.1988.10478722 for the statistical rationale for the methods used.
Author(s)
Maintainer: Noah William Trelawny Hellen noahhellen@gmail.com [copyright holder]
See Also
Useful links:
Report bugs at https://github.com/NoahHellen/missr/issues
Simulated animal health data (MCAR)
Description
A toy dataset with heart rate data for various animals.
Usage
animalhealth
Format
A 200 x 2 data frame:
- animal
The animal of interest
- hear_rate
The corresponding heart rate of the animal (bpm)
Simulated company data (MNAR)
Description
A toy dataset with typical company metrics across various firms.
Usage
companydata
Format
A 500 x 5 data frame:
- sales
Sales in the last fiscal year (USD, million)
- marketing_spend
Marketing spend in last fiscal year (USD, million)
- product_rating
Average rating across all products
- employees
Total employee count in last fiscal year
- gross_profit
Gross profit in last fiscal year (USD, million)
Simulated health check data (MAR)
Description
A toy dataset with typical health check-up metrics for various individuals.
Usage
healthcheck
Format
A 200 x 5 data frame:
- bone_mass
Bone mass of individual (kg)
- body_fat
Body fat percentage of individual
- height
Height of individual (cm)
- age
Age of individual
- rbc
Red blood cell count of individual (million/mm^3)
Missing at random (MAR) test
Description
![[Stable]](./figures/lifecycle-stable.svg)
mar() performs multiple logistic regressions to test for MAR.
The null hypothesis for each is that the data are not MAR.
Usage
mar(data, debug = FALSE)
Arguments
data |
A data frame. |
debug |
A logical value used only for unit testing. |
Details
In the following, each column of M with missing data is regressed on
D_obs. Each regression produces a vector of p-values (one for each
variable in D_obs). The smallest p-value is the most important. This
is because missing data need only be dependent on one observed variable
for the data to be MAR. If each reported smallest p-value is significant,
the data is MAR. See vignette("background") for definitions of M and
D_obs.
Value
missing |
Column of M with missing data |
p_value |
Smallest p-value of the logistic regressions |
explanatory |
Variable corresponding to |
p_values |
The p-values of the logistic regressions |
variables |
Variables corresponding to |
combined |
Paired |
Examples
mar(healthcheck)
Little's missing completely at random (MCAR) test
Description
mcar() performs Little's MCAR test to test for MCAR.
The null hypothesis is that the data is MCAR.
Usage
mcar(data, debug = FALSE)
Arguments
data |
A data frame. |
debug |
A logical value used only for unit testing. |
Details
This function reproduces the d^2 statistic in equation (5) from [1].
This statistic is used to test for MCAR. Comments reference variables
from vignette("background") (in brackets) to improve readability and
traceability.
Value
statistic |
The d^2 statistic |
degrees_freedom |
Degrees of freedom of chi-squared distribution |
p_val |
P-value of the test |
missing_patterns |
Number of missing patterns |
Note
Code is adapted from mcar_test() from the naniar package
using base R instead of the tidyverse.
References
[1] Little RJA. A Test of Missing Completely at Random for Multivariate Data with Missing Values. Journal of the American Statistical Association. 1988;83(404):1198-202.
Examples
mcar(pollutionlevels)
Missing not at random (MNAR) classification
Description
mnar() presents the statistics from mar() and mcar(). If at least one
p-value in mar() is not significant, and the p-value in mcar() is
significant then the data is MNAR.
Usage
mnar(data)
Arguments
data |
A data frame |
Details
There exists no formal test for MNAR data. This function therefore
presents the statistics for the tests in mar() and mcar(). If the
results suggest the data is neither MAR nor MCAR, one can use process of
elimination to deduce that the data is MNAR.
Value
A list:
mcar |
Results of Little's MCAR test |
mar |
Results of MAR test |
Examples
mnar(companydata)
Simulated pollution level data (MCAR)
Description
A toy dataset with typical pollution level metrics for various settlements.
Usage
pollutionlevels
Format
A 200 x 4 data frame:
- light
Light pollution of settlement (mag/arcsec^2)
- visual
Visual pollution of settlement (VPI)
- noise
Noise pollution of settlement (dB)
- air
Air pollution of settlement (AQI)
Simulated test scores data
Description
A toy dataset with test scores of various students.
Usage
testscores
Format
A 200 x 2 data frame:
- id
The ID of the student
- score
The student's score in the test