Assess Survey Nonresponse Bias

Evidence and applicability: use only supplied evidence for source values, historical facts, completed checks and approvals. Cite source labels and row, page or section references. Distinguish observations, source claims, calculations, assumptions and proposals. Preserve conflicting evidence and mark missing evidence unknown or not evidenced. Do not invent values, authority, permissions, sign-off or agreement requirements. Apply only supplied applicable obligations. Keep recommendations conditional on unresolved checks and the responsible owner's decisions. Descriptive change alone does not establish program impact.

Use only materials supplied in this conversation. Ask me for the survey purpose/estimand, target population, sampling design and selection counts, final disposition table and eligibility/completion rules, permitted frame covariate summaries, respondent summaries and any documented weights. Use safe aggregate extracts rather than participant records. If essential records are missing, give targeted requests and an explicitly conditional diagnostic plan, not invented rates or bias findings.

Assess postcollection survey unit nonresponse and what the achieved sample can support. This is a diagnosis and adjustment feasibility assessment, not a new sample-size plan.

1. Establish the intended population and estimand, sampling unit and design, selected sample and frame coverage, fieldwork dates/modes, contact protocol, eligibility rules, final disposition definitions, survey completion rule and existing weights. Ask for actual safe disposition counts and rules if missing; a target and achieved total alone do not establish a response rate or its denominator. Separate unit nonresponse from item missingness, valid skips, ineligibility and frame coverage error. For opt-in or other nonprobability designs, explain which selected/eligible denominator is known and limit population inference accordingly.
2. Reconcile final mutually exclusive unit dispositions to the documented selected sample. Separate complete and partial interviews, refusals, noncontacts, other eligible noninterviews, known ineligible units and unknown eligibility where the supplied records permit. Distinguish contact attempts from sampled units and documented original units from replacements/reserves. Do not treat unknown eligibility as ineligible, duplicates as separate units, or replacements as if the original unit responded. Flag unresolved totals and classification conflicts; do not silently repair them.
3. Calculate only rates supported by the counts and an explicitly named, fully specified rule. For each reported response, cooperation or contact rate, show its numerator, denominator, treatment of partials and unknown eligibility, substitutions and source references. Do not label different rates interchangeable or invent an official standard. If eligibility is fully resolved, a simple complete-interview proportion among all eligible sampled units may be shown with that exact definition; do not present it as a standardized rate without its applicable specification. With unknown eligibility, leave the corresponding rate unresolved or show explicitly conditional alternatives under justified stated assumptions; do not guess an eligible fraction. Separate weighted from unweighted rates. Disclose unresolved or excluded dispositions. Do not apply a universal acceptable response-rate cutoff.
4. Compare respondents with eligible sampled nonrespondents or the relevant frame using safe variables available for both, and external population benchmarks only when definitions and coverage match. Specify comparator, denominator, missing covariate coverage, design/weights and subgroup disclosure limits for each comparison. Report supported differences or response propensities by relevant observed characteristics, with calculations only when inputs permit. Distinguish nonresponse imbalance from frame undercoverage. Balance on observed characteristics cannot establish absence of bias on the outcome; a low rate does not quantify bias and a high rate does not rule it out. If information on nonrespondents is absent, state that this diagnostic cannot be completed rather than assuming they resemble respondents.
5. Assess feasible adjustments, without automatically constructing weights: preserve supplied design weights and explain candidate response-cell, propensity or calibration adjustments only when selection information, auxiliary data, overlap/positivity and benchmark quality support them. Identify assumptions linking response and outcomes, sparse cells, extreme-weight risks and design-aware uncertainty needs. Adjustment cannot recover a wholly unobserved group by assumption or guarantee representativeness. Label untestable assumptions. Calculate adjustments only with sufficient authorized inputs and reproducible specifications; otherwise describe a plan and identify specialist checks needed. Do not infer weights, replace missing outcomes with zeros or declare bias corrected.
6. Propose sensitivity comparisons tied to the estimand: unadjusted versus defensibly adjusted results if feasible, or clearly labeled outcome/eligibility scenarios or bounds when their mathematical and substantive assumptions can be stated. Keep scenarios separate from observed results and do not invent nonrespondent outcomes. Explain assumptions, plausible justification, data requirements and how conclusions could change; where these cannot be supplied, list unresolved sensitivity needs. Separate item-missingness treatment for a dedicated analysis rather than silently imputing it here.
7. Deliver a disposition reconciliation and rate worksheet, comparator/imbalance table, bias-evidence and unknowns note, adjustment and sensitivity feasibility plan, and concise inference/disclosure statement. Distinguish the achieved respondent sample from the intended population, describe precision/design limitations, and identify claims the evidence does and does not support. Do not claim population representativeness, causal impact, completed validation or approval from this assessment. Prefer aggregate evidence and suppress unsafe small cells; do not request names or contact records.