Service-request withdrawal attrition bias
Withdrawn requests can change service metrics when they disappear from the denominator or are relabeled as successful closure. Start with every eligible request received, preserve its later disposition, and compare results with and without withdrawals so the reporting effect is visible.
Withdrawal is an outcome category, not a verdict about service quality. A requester may withdraw because the need changed, the request duplicated another record, an answer arrived elsewhere, or waiting made the request irrelevant.
The federal statistical standard gives a useful reporting trigger: when a survey's expected unit response rate is below 80 percent, agencies must plan a nonresponse-bias analysis. The OMB survey standards do not make 80 percent a universal quality score, but they show why missing outcomes require analysis rather than silent removal.
Build an entry cohort
Freeze the population at a declared intake event before examining final status. Retain the original identifier, requester grouping, channel, received time, first response, withdrawal time, recorded reason, and any link to a replacement request.
Do not define eligibility from the final status. A rule that selects only completed cases will omit withdrawals by construction and cannot measure attrition.
Reconcile status labels across systems before grouping outcomes. Cancelled, duplicate, no longer needed, merged, and administratively closed may have different meanings even when a dashboard collapses them.
Separate timing and reason
Measure whether withdrawal occurred before acknowledgment, after a first response, during dependency waiting, or after proposed resolution. Timing helps distinguish intake churn from later service abandonment without assigning motives that the record does not support.
Use a small reason taxonomy with an explicit unknown class. Preserve the submitted note separately and do not infer a reason from silence.
Report reason completeness and review a sample against source records. A neat distribution built from mostly default values is not reliable evidence.
Recalculate the service measures
Show request volume, response coverage, completion rate, and elapsed-time summaries with every entry retained. Then publish a sensitivity table showing how each measure changes if withdrawals are excluded, counted as unresolved, or presented as their own outcome.
Keep duplicate replacements linked so one need is not mistaken for two independent failures. Also retain both records, because deleting the earlier entry hides actual intake work.
Break results out by channel and request type only when the definitions are stable. A high withdrawal share in one channel may reflect its intake design rather than the team handling the work.
Use bounded conclusions
The analysis can show whether attrition changes a reported rate and where withdrawals occur. It cannot prove why a requester left or whether a withdrawn request would have been completed on time.
An operations coordinator can maintain the cohort, reason dictionary, and linked-record review. The service owner sets the disposition policy and decides which withdrawals need direct follow-up.
The nonresponse coverage analysis covers missing replies, the denominator integrity analysis tests population drift, and the Research index provides related methods.
Sources and limits
- OMB survey standards, for the 80 percent nonresponse-analysis trigger and required documentation.
- Census quality standards, for relevance, objectivity, transparency, and error assessment.
- NHIS documentation, for survey instruments, sample design, and response documentation.
- NCES statistical standards, for response-rate calculation, bias evaluation, and disclosure.
- NHIS survey description, for an official example of defined samples, participation, and survey outcomes.
- FDA missing-data guidance, for sensitivity analysis when outcomes are not observed.
- Attrition-bias review, for assessing missing outcome data and its effect on conclusions.
- NIH missing-data resource, for documenting missingness assumptions and alternative analyses.
- GAO data reliability guide, for tracing completeness and accuracy limitations to source systems.
- NIST missing-data guidance, for documenting missing observations before analysis.
The 80 percent figure belongs to the OMB survey standard and is not a service-desk target. A local withdrawal analysis must preserve the actual intake denominator, explain status semantics, and avoid inferring a requester's reason when the record is blank.