[Durgical](https://durgical.com/en/) › [DurgiPedia](https://durgical.com/en/durgipedia/) › [Epidemiology](https://durgical.com/en/epidemiology/) › [Biostatistics](https://durgical.com/en/epidemiology/areas/stats/) › Non-response, losses and missing data

# Non-response, losses and missing data

- Concept

When some participants' data are missing, the result can shift. How many are missing matters, and above all why they are missing.

Updated Nov 13, 2026

## Key points

- MCAR: missingness depends on nothing (a tube was lost); complete cases are unbiased, only less powerful.
- MAR: missingness depends on observed variables (younger people miss more); handled with multiple imputation or models that include those variables.
- MNAR: missingness depends on the unobserved value (those who get worse drop out); no method fully solves it: sensitivity analyses are done.
- Single imputation (mean, last observation carried forward) understates uncertainty and can bias.
- Prevention works best: active follow-up, flexible visits and recording the reason for every loss.

On this page

- The basics Key points
- What it is
- Key facts
- Worked example
- Missing-data mechanisms
- Markers
- In practice
- Common pitfalls
- Memory aid

- Study Test yourself

- Connections Connections

## What it is

Missing data come from people who do not respond, drop out, withdraw or could not be measured. Their effect depends on the mechanism: missing completely at random (MCAR), missing at random given what was observed (MAR) or missing not at random (MNAR), when missingness depends on the unobserved value itself. Analysing complete cases only is simple but can bias; multiple imputation and sensitivity analyses show how much the conclusion depends on the assumptions.

## Key facts

- Worst-case and best-case analysis: assume all missing people in one group failed and those in the other succeeded, and the other way round.
- If the conclusion changes between scenarios, the result is fragile to the losses.

## Worked example

- Made-up trial: 200 per group; outcome, treatment success. Data are missing for 20 in the treated group and 10 in the control group.
- Complete cases: treated 120 of 180 (66.7 %); control 95 of 190 (50.0 %). Difference: 16.7 points.
- Worst case for treatment (missing treated = failure; missing control = success): 120/200 = 60.0 % vs 105/200 = 52.5 %.
- Difference in the worst case: 60.0 − 52.5 = 7.5 points.
- Best case (the reverse): 140/200 = 70.0 % vs 95/200 = 47.5 %; a difference of 22.5 points.
- Reading: the direction of the effect holds in all three analyses, but its size ranges from 7.5 to 22.5 points; the losses weigh on the estimate.

## Missing-data mechanisms

| Mechanism | What it means | Example | Which analysis helps |
| --- | --- | --- | --- |
| MCAR | Missingness is unrelated to any variable | A sample was damaged in transport | Complete cases (unbiased, less power) |
| MAR | Missingness depends on observed variables | Younger participants miss more visits | Multiple imputation with those variables |
| MNAR | Missingness depends on the unobserved value | Those with more side effects drop out | Sensitivity analyses under several assumptions |

## Markers

- MCAR
- MAR
- MNAR
- MI

## In practice

- When reading a trial, look at losses by group and their reasons: unequal losses between groups are a warning sign.
- In surveys, responders and non-responders are compared on whatever data exist for both.

## Common pitfalls

- Quietly excluding those who did not complete the study.
- Filling in the last observation as if it were a real value.
- Assuming MCAR without checking whether those with missing data differ from the rest.

## Memory aid

‘Missing data speak too’: if a value is missing because the patient got worse, that silence is information.

## Test yourself

Think of the answer, then tap the question to see it.

## What does MNAR mean?

That the chance of a value being missing depends on the unobserved value itself, for example when those who get worse drop out.

## Why is single imputation a problem?

Because it treats an invented value as real: it understates uncertainty and can bias the effect.

## Of 100 allocated, 90 have data and 60 improve. In the worst case, what proportion improves?

60 / 100 = 60 %, against 60 / 90 ≈ 66.7 % in complete cases.

## Which method is used when data are missing at random given what was observed (MAR)?

Multiple imputation, or models that include the variables related to missingness.

[Add to my cards (4)](https://durgical.app/estudiar/?add=/en/epidemiology/non-response-losses-and-missing-data/)

## Connections

## Related 5

- [Attrition bias](https://durgical.com/en/epidemiology/attrition-bias/)
- [Intention-to-treat analysis](https://durgical.com/en/epidemiology/intention-to-treat-analysis/)
- [Statistical analysis plan](https://durgical.com/en/epidemiology/statistical-analysis-plan/)
- [Sample size](https://durgical.com/en/epidemiology/sample-size/)
- [Selection bias](https://durgical.com/en/epidemiology/selection-bias/)

[← All of epidemiology](https://durgical.com/en/epidemiology/)

Educational content on epidemiological and statistical methods; the examples use made-up numbers to teach the calculation. Not a clinical or legal guide: current Colombian rules are in the official sources.

Sources: [Chapter 10: Analysing data and undertaking meta-analyses](https://www.cochrane.org/authors/handbooks-and-manuals/handbook/current/chapter-10) (Cochrane, 2026); [E9(R1) Statistical Principles for Clinical Trials: Addendum: Estimands and Sensitivity Analysis in Clinical Trials](https://www.fda.gov/regulatory-information/search-fda-guidance-documents/e9r1-statistical-principles-clinical-trials-addendum-estimands-and-sensitivity-analysis-clinical) (U.S. Food and Drug Administration, 2021); [Attrition bias | Catalog of Bias](https://catalogofbias.org/biases/attrition-bias/) (Catalogue of Bias Collaboration, 2017).

## **Spotted a mistake?** We are always improving for the community. Tell us what is wrong and our team will review it. Report a mistake

Representative model, for educational purposes only.

---
Source: https://durgical.com/en/epidemiology/non-response-losses-and-missing-data/
License: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Cite as: DurgiPedia (durgical.com), https://durgical.com/en/epidemiology/non-response-losses-and-missing-data/