Missing Data: Types, Risks, and Solutions
Understand MCAR, MAR, and MNAR — the three types of missing data — and which methods for handling them are valid versus dangerously misleading.
You'll learn
The three types of missingness (MCAR/MAR/MNAR) and how each affects your analysis differently.
Use this when
Your dataset has missing values and you need to decide how to handle them.
Why Missing Data Is a Serious Problem
Missing data reduces your effective sample size and, depending on why data are missing, can introduce systematic bias. Many researchers delete missing rows or fill with the mean without considering whether that action is valid — and in many cases it is not.
💡 Missing is not random by default
The worst assumption you can make is that missing values are "just noise." In clinical research, missing lab values often occur because a test was not ordered — which is correlated with the patient's clinical state. That is systematic, not random.
The Three Mechanisms of Missingness
Rubin (1976) defined three distinct mechanisms, and the validity of every missing-data method depends on which applies to your data.
| Mechanism | What it means | Example | Safe to delete? |
|---|---|---|---|
| MCAR — Missing Completely At Random | Probability of missing is unrelated to any variable | Lab tube dropped by accident | Yes — unbiased, only loses power |
| MAR — Missing At Random | Probability of missing depends on observed variables but not on the missing value itself | Women less likely to report weight; adjust for sex | No — use multiple imputation |
| MNAR — Missing Not At Random | Probability of missing depends on the unobserved (missing) value itself | Sicker patients skip follow-up visits | No — no standard fix; sensitivity analysis needed |
⚠️ You cannot verify MCAR from data alone
You can test whether missingness is related to observed variables (Little's MCAR test). But MNAR cannot be ruled out from data — it requires domain knowledge about why data are missing.
Common Methods and When They Are Valid
- ●Complete case analysis (listwise deletion): valid only under MCAR — biased under MAR/MNAR
- ●Mean/median imputation: reduces variance artificially, distorts correlations — generally not recommended
- ●Last observation carried forward (LOCF): common in clinical trials but assumes condition is stable — usually optimistic
- ●Multiple imputation (MI): valid under MAR; creates several complete datasets, analyses each, pools results — the current standard
- ●Maximum likelihood (FIML): valid under MAR; estimates parameters using all available data without creating imputed datasets
💡 Reporting missing data
Always report: (1) the number and percentage of missing values per variable, (2) whether missingness appears to be related to outcome or exposure, (3) the method used to handle missing data and why that method is appropriate.
Practical Steps in VibeResearch
Before running any analysis, use the Prepare tab to inspect each variable's completeness. The data quality panel shows missing count, percentage, and flags variables with >10% missingness for review.
- 1.Check the missing count per variable in Prepare → Data quality
- 2.Identify whether the missingness pattern is systematic (e.g., correlated with age or disease severity)
- 3.Decide on your strategy: exclude, impute, or sensitivity analysis
- 4.Document your decision in the Methods section
Trusted sources behind this lesson
Further reading
Read next
Effect Size: Why Statistical Significance Is Not Enough
Understand Cohen's d, odds ratios, and other effect size measures — and why a statistically significant result can still be clinically meaningless.