Completer Balance Table¶
Diagnostic table — instantiates Attrition and Dropout Monitoring
Lines up the people who stayed against the people who left, covariate by covariate, to show whether the two groups were ever the same population.
Attrition only becomes a validity problem when the leavers differ from the stayers. Completer Balance Table is the artifact that checks. It is a two-column comparison — completers in one column, non-completers in the other — laid out across the baseline covariates and burden indicators the study already measured, so a reader can see, variable by variable, whether the people who remained resemble the people who disappeared. Its defining move is to convert attrition from a count into a characterized difference: not "we lost 18%" but "the 18% we lost were younger, poorer, and more heavily burdened than the ones we kept." It is a description of observed data, deliberately not an inference test — it shows whether and how the groups diverge, and stops short of declaring what that does to the conclusion.
Example¶
A long-running household income panel reaches its fourth annual wave and wants to know whether the people who kept answering still represent the sample it started with. The balance table sets wave-4 completers beside those who dropped out after earlier waves and compares them across baseline fields: household income, education, age, employment status, region, and a set of burden indicators — interview length, number of prior contact attempts, and past item-nonresponse. The pattern is unmistakable. Attriters were, at baseline, younger, lower-income, more mobile, and had recorded markedly higher response burden. The standardized differences on income and age are large; on region they are trivial. The table does not pronounce the panel biased. It shows, in black and white, that the survivors are not a random slice of who started — which is exactly the finding the next mechanism needs.
How it works¶
Its distinguishing discipline is standardized description over significance testing:
- Define completer and non-completer at a specific stage and hold that definition fixed across the table.
- Tabulate each covariate in both columns — means for continuous variables, proportions for categorical — side by side.
- Report standardized differences, not p-values. How large the gap is matters for bias; whether it is "significant" does not, and testing it invites a classic error.
- Include burden indicators as their own rows, reading whether the people who left were carrying more of the study's load than those who stayed.
Tuning parameters¶
- Covariate set — which baseline variables to compare; more rows catch more forms of imbalance but dilute attention and invite cherry-picking.
- Completer definition — the stage and threshold at which a unit counts as retained; moving it reshapes both columns.
- Standardized-difference threshold — how large a gap is treated as material; stricter thresholds flag more imbalance.
- Burden indicators included — which load measures (visit count, form length, contact attempts) enter the table.
- Subgroup stratification — whether the contrast is drawn overall or within arm and site, exposing whether imbalance is itself differential.
When it helps, and when it misleads¶
Its strength is that it turns attrition into a validity diagnosis on evidence the study already holds — no new data collection, just an honest side-by-side. It is the step that tells a team whether their missingness is plausibly benign or almost certainly consequential.
Its honest limit is that it can only compare observed baseline covariates. Two groups can look perfectly balanced on everything measured and still differ sharply on the unobserved, outcome-related reason they left — the table cannot see that. The classic misuse is the Table 1 fallacy[n1]: attaching significance tests to baseline differences and reading "not significant" as "balanced," which conflates sample size with the size of the imbalance. The guarding discipline is to report standardized differences, treat balance-on-observed as necessary but not sufficient, and hand the result to a sensitivity analysis rather than concluding the study is safe.
How it implements the components¶
completer_non_completer_contrast— it is the contrast: the covariate-by-covariate comparison of retained versus lost units.participant_burden_audit— it reads the study's burden indicators into the comparison, showing whether leavers bore a heavier participation load than stayers.
It does not record which missingness assumption the observed differences license, nor test whether the conclusion survives them (missingness_assumption_record, its nearest twin Missing-Data Sensitivity Analysis) — the table describes, it does not infer — and it does not act to reduce that burden or recover the missing endpoints (endpoint_recovery_channel, Retention Outreach Protocol).
Related¶
- Instantiates: Attrition and Dropout Monitoring — the balance table is the archetype's diagnostic step, turning loss into a measured difference.
- Sibling mechanisms: Participant Flow Diagram · Attrition Dashboard · Withdrawal Reason Survey or Interview · Missing-Data Sensitivity Analysis · Retention Outreach Protocol · Data Monitoring Review
Editorial Notes¶
Form Classification¶
Form family: Assessment, Review & Assurance
Rationale: Lines up the people who stayed against the people who left, covariate by covariate, to show whether the two groups were ever the same population, making its operative form a bounded evaluation of existing evidence or work that produces a finding or disposition.
Independent corroboration: The frozen evidence defines Completer Balance Table as 'Lines up the people who stayed against the people who left, covariate by covariate, to show whether the two groups were ever the same population', so its operative form is Assessment, Review & Assurance.
Review outcome: Independent reviewer agreement; high confidence.
Origin Attribution¶
Primary origin: Statistics & Experimental Design
Origin pattern: Convergent development
Present-day reach: Specialized
Rationale: Trial and longitudinal-study methodology cohered baseline tables comparing completers with non-completers to diagnose attrition bias.
Related originating lineages:
- Medicine & Healthcare — Clinical research made loss-to-follow-up and differential burden especially consequential applied uses.
Review resolution: Both reviewers agree on statistics_experimental_design as primary. Reading the source mechanism confirms that its defining operation belongs to that lineage; the final record retains medicine_healthcare only where it materially formed the mechanism and keeps present-day application breadth separate from provenance.
Review outcome: Reconciled after independent review; high confidence.
Notes¶
The balance table and the Missing-Data Sensitivity Analysis are the two most easily conflated mechanisms here, because both ask whether attrition biased the result. The separating idea in one sentence: the balance table describes the observed difference between leavers and stayers, while the sensitivity analysis records a missingness assumption and tests the conclusion under a range of guesses about the unobserved — the table stops at what the data show, the analysis reasons past it.
[n1] The Table 1 fallacy is the practice of running significance tests on baseline covariate differences (especially in randomized trials) and interpreting non-significant results as evidence of balance. Statisticians warn against it because a p-value reflects sample size as much as imbalance; standardized differences, which the balance table reports instead, measure the magnitude that actually bears on bias. ↩