mvp

Version 1.0.0 | 2026-04-08

mvp analyzes missing-data structure across variables, reports distinct missingness patterns, and can graph missingness as bar charts, pattern-frequency charts, observation-by-variable heatmaps, and missingness-correlation heatmaps. It is a Stata 16+ fork of mvpatterns with added graphing, stratification, generated indicators, and monotone-pattern diagnostics.

Requirements

• Stata 16 or later
• No external dependencies. When tetrachoric is unavailable, mvp, correlate falls back to Pearson correlations among missingness indicators.

Installation

capture ado uninstall mvp
net install mvp, from("https://raw.githubusercontent.com/tpcopeland/Stata-Tools/main/mvp") replace

Commands

How It Works

• If you omit varlist, mvp analyzes all variables in memory and drops variables with no missing values from the pattern display unless you specify nodrop.
• The main text output has two layers: a variable-level missingness summary and a pattern table where + means observed and . means missing.
• generate(stub) creates one missingness indicator per analyzed variable plus stub_pattern and stub_nmiss for downstream analysis.
• save(name) stores the pattern dataset either in a Stata frame or in a .dta file, depending on the name you supply.
• graph(bar), graph(patterns), graph(matrix), and graph(correlation) provide four complementary views of missingness. gby() and over() let you compare groups inside the graph workflows.

Worked Examples

1. Quick audit of missingness in a built-in dataset

sysuse auto is a good first check because rep78 has missing values and the example runs immediately after installation.

sysuse auto, clear
mvp price mpg rep78 headroom trunk, percent sort

Use this pattern when you want a fast table of missingness by variable and by pattern before modeling or imputation planning.

2. Generate missingness indicators and save the pattern data

This workflow is useful when you want to reuse the missingness structure later in the same Stata session.

sysuse auto, clear
mvp price mpg rep78, generate(m) save(patterns)
tab m_pattern
frame patterns: list

3. Compare missingness across groups with a graph

gby() creates faceted graphs so you can see whether missingness changes across categories.

sysuse auto, clear
mvp price mpg rep78, graph(bar) gby(foreign)

4. Check monotonicity and co-occurring missingness in a richer synthetic example

For monotone checks and correlation heatmaps, it helps to have several variables with distinct missingness rates.

clear
set seed 12345
set obs 300

gen age = rnormal(50, 12)
gen female = runiform() < 0.5
gen bmi = rnormal(27, 5)
gen sbp = rnormal(130, 18)
gen ldl = rnormal(3.5, 1.1)
gen hba1c = rnormal(5.8, 0.9)

replace bmi = . if runiform() < 0.08
replace sbp = . if runiform() < 0.05
replace ldl = . if runiform() < 0.15
replace hba1c = . if runiform() < 0.20
replace hba1c = . if missing(ldl) & runiform() < 0.4

mvp age female bmi sbp ldl hba1c, percent sort monotone correlate
mvp bmi sbp ldl hba1c, graph(correlation) textlabels

Graph Workflows

For large datasets, graph(matrix) samples 500 observations by default. Use graph(matrix, sample(#)) to change that, and add sort inside graph(matrix, sort) when you want observations grouped by pattern.

Demo

. mvp age female bmi sbp ldl hba1c, percent sort monotone correlate

Variables with no missing: age female

Variable     | Type     Obs    Miss   %Miss   Variable label
-------------+-----------------------------------------
hba1c        |  double    369    131   26.2   HbA1c
ldl          |  double    417     83   16.6   LDL cholesterol
bmi          |  double    452     48    9.6   Body mass index
sbp          |  double    470     30    6.0   Systolic BP
-------------------------------------------------------

Missing value patterns

  +----------------------------------+
  | _pattern   _miss   _freq    _pct |
  |----------------------------------|
  |     ++++       0     276   55.20 |
  |     .+++       1      74   14.80 |
  |     ..++       2      40    8.00 |
  |     +.++       1      35    7.00 |
  |     ++.+       1      32    6.40 |
  |     +++.       1      18    3.60 |
  |     .+.+       2       9    1.80 |
  |     .++.       2       5    1.00 |
  |     ++..       2       3    0.60 |
  |     +..+       2       3    0.60 |
  |     +.+.       2       2    0.40 |
  |     ..+.       3       2    0.40 |
  |     ...+       3       1    0.20 |
  +----------------------------------+

--------------------------------------------------
Total observations:             500
Complete cases:                 276  ( 55.2%)
Incomplete cases:               224  ( 44.8%)
Unique patterns:                 13
Variables analyzed:               4
Max missing/obs:                  3
Mean missing/obs:              0.58
--------------------------------------------------

Monotone missingness test:
  Observations with monotone pattern:      297 ( 59.4%)
  Pattern is non-monotone

Tetrachoric correlations of missingness:
(correlations among missingness indicators)

        hba1c     ldl     bmi     sbp
hba1c   1.000
  ldl   0.453   1.000
  bmi  -0.097  -0.217   1.000
  sbp  -0.045  -0.070   0.012   1.000

Graph gallery

Multiple treatment groups

The gby() and over() options support any number of groups. Here the dataset has 3 treatment arms (Placebo, Low dose, High dose) with the high-dose arm receiving extra lab dropout missingness.

. mvp bmi sbp ldl hba1c if arm == 0, percent sort   // Placebo

Variable     | Type     Obs    Miss   %Miss   Variable label
-------------+-----------------------------------------
hba1c        |  double    130     37   22.2   HbA1c
ldl          |  double    144     23   13.8   LDL cholesterol
bmi          |  double    151     16    9.6   Body mass index
sbp          |  double    153     14    8.4   Systolic BP
-------------------------------------------------------

Complete cases:           95  ( 56.9%)    Unique patterns: 10

. mvp bmi sbp ldl hba1c if arm == 1, percent sort   // Low dose

Variable     | Type     Obs    Miss   %Miss   Variable label
-------------+-----------------------------------------
hba1c        |  double    121     46   27.5   HbA1c
ldl          |  double    144     23   13.8   LDL cholesterol
bmi          |  double    149     18   10.8   Body mass index
sbp          |  double    163      4    2.4   Systolic BP
-------------------------------------------------------

Complete cases:           94  ( 56.3%)    Unique patterns: 8

. mvp bmi sbp ldl hba1c if arm == 2, percent sort   // High dose

Variable     | Type     Obs    Miss   %Miss   Variable label
-------------+-----------------------------------------
hba1c        |  double     97     69   41.6   HbA1c
ldl          |  double    116     50   30.1   LDL cholesterol
bmi          |  double    152     14    8.4   Body mass index
sbp          |  double    154     12    7.2   Systolic BP
-------------------------------------------------------

Complete cases:           67  ( 40.4%)    Unique patterns: 13

Regenerate with:

do mvp/demo/demo_mvp.do

Selected Returned Results

mvp stores several useful summaries in r(), including:

• r(N): observations analyzed
• r(N_complete) and r(N_incomplete): complete and incomplete cases
• r(N_patterns): number of unique patterns displayed
• r(N_vars): number of variables analyzed
• r(varlist): analyzed variables with missing values
• r(monotone_status): monotone or non-monotone when monotone is requested
• r(corr_miss): missingness correlation matrix when correlate or graph(correlation) is used

Version History

• 1.0.0 (2026-04-08): Initial Stata-Tools release of the enhanced mvpatterns fork with graphics, generated indicators, saved pattern data, and monotone diagnostics.

Author

Timothy P Copeland, Karolinska Institutet

Fork of mvpatterns version 2.0.0 by Jeroen Weesie.