Survey methodology covers sampling, instrument design, data collection methods, and evidence traceability. Learn the types and how to build one that holds up.
Five sources of error live inside every survey. Methodology is the discipline of finding them, naming them, and reducing the ones that matter most for the decision the data has to support.
This guide explains the framework in plain practitioner terms: the five sources of total survey error, the six methodology principles that hold across every kind of survey, and the diagnostics that surface error before it ends up in a report. The worked example follows a multi-language workforce program running across English, Spanish, and Vietnamese cohorts.
Survey design
What the data should be able to claim. Decisions about types, identifiers, wave structure, and analysis output.
Survey methodology
How confidently the data can claim it. Diagnostics that find the five error sources and reduce the ones that matter.
The two pages cover the same survey at different moments. Design controls the ceiling. Methodology measures whether the actual data hit it.
Total Survey Error, often called the TSE framework, decomposes data quality into five separable sources. Each is reducible through different practices. The discipline is not eliminating error (impossible) but knowing which sources threaten the specific claim the survey has to support, and reducing those first.
The frame does not include people the survey is trying to describe. A workforce program that surveys only English-speaking participants when forty percent of the cohort speaks Spanish has a coverage gap before fielding starts.
Reduces by: mapping the population before drawing the frame.
Even with a perfect frame, drawing 200 people out of 1,000 produces estimates that differ from the true population value. Sampling error is calculable when probability methods are used and reduces with larger samples.
Reduces by: probability sampling and adequate sample size.
Invited participants who do not respond differ systematically from those who do. A 30 percent response rate where the responding 30 percent skews toward more engaged participants overstates outcomes for the whole cohort.
Reduces by: planned follow-up, mode flexibility, and honest reporting.
Answers that do not reflect what the question intended to measure. Causes include confusing wording, social desirability bias, recall failure, mode effects, and translation drift across languages.
Reduces by: piloting questions, cognitive interviews, back-translation.
Error introduced between the moment a participant answers and the moment the answer becomes a number in a report. Manual data entry, inconsistent coding of open-ended responses, and version drift between tools all compound here.
Reduces by: automation, audit trails, single-source-of-truth data architecture.
The order matters. Coverage error cannot be fixed during analysis. Measurement error cannot be fixed during reporting. Each error has one window where it is reducible at low cost; methodology is the discipline of recognizing that window before it closes.
The TSE framework is the canonical practitioner model for survey data quality, originating in academic survey methodology research and now standard in nonprofit evaluation, government statistical agencies, and applied research. Every error source above can be diagnosed without specialized statistical training; the principles section below shows the discipline that catches each one.
The terms below are the ones practitioners hear most often. The differences matter because they decide which decisions need to be made before the survey runs and which need to be checked after responses arrive.
Survey methodology is the discipline of measuring where survey data is accurate and where it is not. It covers the five sources of error that exist in every survey, the diagnostics that surface them, and the workflow that reduces error before results are reported.
Methodology is what makes the difference between a survey that produces a number and a survey that produces a number you can defend. Every survey contains coverage, sampling, nonresponse, measurement, and processing error in some amount. The methodology question is not whether error exists but which kinds threaten the specific decision the data has to support, and how much of each is acceptable for that decision.
Survey methodology means the practice of measuring data quality, naming the failure modes, and reducing the most consequential ones before reporting. The meaning is diagnostic: methodology answers the question how confident can we be in this result.
Most practitioners encounter the term as something academic researchers worry about. The framework is equally useful for nonprofit evaluators, program managers, customer feedback teams, and government statistical analysts. The vocabulary differs across fields. The five-error decomposition does not.
Survey methodology covers five error types that together describe total survey error. Coverage error is who the survey missed entirely because they were not in the frame. Sampling error is the variation introduced by surveying some rather than all. Nonresponse error is the bias from people who were invited but did not answer. Measurement error is what people answered that did not reflect reality. Processing error is what got changed or lost between collection and report.
Each error type calls for a different diagnostic and a different fix. A survey can have low sampling error and high nonresponse error at the same time. The methodology workflow finds which errors are concentrated where, and reduces the ones that threaten the report most.
Survey methodology best practices reduce the five errors in priority order based on which ones threaten the decision most. Map the population before drawing the sample to reduce coverage error. Use probability sampling when feasible to make sampling error calculable. Plan nonresponse follow-up before fielding rather than reacting to low response. Pilot the questions for measurement validity before launch, including cognitive interviews when stakes are high. Track error sources at every handoff between collection, coding, and analysis. Document the methodology so the next reader can audit it.
Generic survey methodology best practices that focus on writing better questions address only the measurement error layer. The other four error sources need their own attention. A perfect questionnaire fielded to the wrong frame produces clean answers from the wrong people.
Survey design is the architectural decisions made before any question is written: type of design, identifier strategy, wave structure, analysis workflow. Survey methodology is the discipline of measuring whether those decisions worked once data starts arriving.
Design controls what the data should be able to claim. Methodology controls how confidently it can claim it. Most programs need both. They run sequentially: design before launch, methodology continuously during and after. A program that does design well and skips methodology produces clean architecture filled with unverified data. A program that does methodology well without design has nothing coherent to verify.
Survey design is the architectural decisions made before any question is written. Methodology measures whether those decisions held in the field. Read the survey design guide.
Questionnaire design is the question-writing layer. It addresses one of the five methodology error sources (measurement) and leaves the other four unaddressed.
Survey research methodology is the academic counterpart, used in research methods textbooks. The five-error framework is the same. The acceptable error thresholds differ by field and stakes.
Data analysis methodology covers what happens after collection ends. Survey methodology covers everything from the frame to the moment data enters analysis, including the handoff itself.
Each principle below reduces a different one of the five error sources. The order matches the order error enters a survey: frame, sample, response, measurement, handoff. Methodology that follows this sequence catches errors early, when they are still cheap to fix.
Who you can reach and who you cannot.
Before drawing a sample, document who the survey is trying to describe and which subgroups exist within that population. Write down which subgroups will be reachable through the planned channels and which will not. A program serving Spanish-speaking participants that mails English-only surveys has documented its coverage gap rather than discovering it after analysis.
Reduces: coverage error. Caught early by population mapping before frame selection.
Random when feasible. Census when small.
For populations larger than the budget allows surveying entirely, probability sampling makes sampling error calculable. For populations small enough to survey completely, census sampling eliminates sampling error. Convenience samples (whoever shows up) make sampling error unknowable, which means the report cannot honestly state how confident the estimates are.
Reduces: sampling error. Caught early by choosing the sampling method before fielding.
Reminders are a workflow, not a reflex.
The moment to plan nonresponse follow-up is before launch, not after low response rates appear. Plan the reminder cadence, the alternative modes (SMS, in-person, phone), and the cutoff at which the survey closes. Plan how response rate by subgroup will be reported. Programs that react to nonresponse instead of planning for it produce uneven data across subgroups and cannot disentangle the cause.
Reduces: nonresponse error. Caught early by a written nonresponse plan before fielding.
A question that confuses ten people will confuse a thousand.
Pilot the survey with five to ten participants from the target population before full launch. Watch them answer. Ask afterward what each question meant to them. Measurement error caught in pilot testing is fixed in hours; the same error caught after fielding 500 responses requires choosing between bad data and a redesign that loses comparability. Translation drift is a measurement-error sub-category that benefits especially from cognitive interviews in each language.
Reduces: measurement error. See qualitative survey guidance for open-ended question piloting.
Each tool transition is a place for data to drift.
Survey responses typically pass through three or four systems between collection and report: collection tool, data export, coding system, analysis tool. Each handoff is a place where a column gets renamed, a value gets miscoded, or a row drops. Track row counts, value distributions, and flag any handoff where they change unexpectedly. Manual handoffs introduce more processing error than any other source in modern survey workflows.
Reduces: processing error. Single-source-of-truth architectures eliminate most handoffs entirely.
A reader should be able to audit your error decisions.
Every report includes a methodology section. Frame description, sampling method, response rate by subgroup, mode mix, instrument version, coding rules, and any known error sources. The reader should be able to look at this section and know how confident to be in the headline number. Methodology documentation is not bureaucratic overhead; it is what makes a finding defensible to a skeptical reader.
Reduces: reader misinterpretation. Always required regardless of error magnitude.
Each row below names a methodology decision and which error source it controls. The broken-way column describes the workflow that fails quietly. The working-way column describes the practice that catches error early. The decides column names what the choice actually controls in the final report.
Frame definition
Who the survey can reach. Controls coverage error.
The frame is whoever is on the email list, with no documented mapping of who is missing. Spanish-speaking participants are not in the frame because the program never collected their language preference. The gap is invisible until report time.
Population mapped before drawing the frame. Subgroups documented. Coverage gaps named explicitly, with a decision about whether to extend the frame or accept the gap and report the limitation.
Whether the report can claim to describe the program participants or only the reachable subset. Coverage error caps the population claim.
Sampling method
How participants are selected. Controls sampling error.
Whoever responds first becomes the sample. Sample size set by gut estimate. No record of how the responding participants compare to the frame on basic demographics. Confidence intervals are not calculable.
Probability sampling when budget allows, census sampling when the population is small enough. Sample size justified against the precision the analytical question requires. Sample composition compared to frame composition before reporting.
Whether the report can offer honest confidence intervals or only point estimates of unknown precision.
Nonresponse plan
What happens when invited people do not answer. Controls nonresponse error.
One reminder email sent the day before the survey closes. No follow-up by phone or in person. No comparison of respondents to non-respondents on observable variables. Response rate by subgroup not reported.
A planned reminder cadence with multiple modes (email, SMS, in-person) targeted to non-respondents only. Response rate tracked by subgroup and reported transparently. Known nonresponse bias acknowledged in the methodology section.
Whether response rate caps the strength of any claim about the whole cohort, or whether the data can speak only for respondents.
Question piloting
Whether questions get tested before launch. Controls measurement error.
Survey is launched on the same day questions are finalized. Confusing items are discovered when 200 responses arrive with the same write-in clarification asking what the question meant. By then, fixing the question loses comparability with prior responses.
Pilot with 5 to 10 participants from the target population. Cognitive interviews after the pilot ask each respondent what each question meant to them. Translation drift checked through back-translation when running multi-language instruments.
Whether the responses to a question actually reflect what the question intended to measure, or capture confusion masquerading as data.
Translation strategy
How multi-language surveys keep instruments equivalent. A measurement-error sub-decision.
A staff member translates the English instrument into Spanish and Vietnamese in an afternoon. No back-translation, no cognitive interviews, no review by native speakers from the participant population. The Spanish version of confident carries different connotations than the English version, and nobody catches it.
Professional translation followed by independent back-translation. Cognitive interviews in each language with participants from the target population. Anchor labels and rating scales reviewed for connotative equivalence, not only literal accuracy.
Whether responses across languages can be compared at all, or only described separately within each language cohort.
Handoff tracking
How data quality is checked between systems. Controls processing error.
Data exports as CSV, opens in a spreadsheet, gets cleaned manually, copies to the analysis tool. Each step is unverified. Three weeks later, an analyst notices a subgroup is missing 40 rows, traces it to a sort that was never undone, and rebuilds.
Single-source-of-truth architecture where collection and analysis share the same data. Row counts and value distributions audited at every transition. Manual handoffs eliminated where possible. Audit log preserved for every change.
Whether the numbers in the final report match the answers participants actually gave, or have drifted through unverified transitions.
Errors compound additively across the five sources. A survey with 10 percent coverage error, 5 percent sampling error, 30 percent nonresponse error, and 5 percent measurement error has roughly 50 percent total survey error against the population claim, even though each individual source looks acceptable. Methodology forces practitioners to look at the sum, not only each component.
A workforce training program serves participants across English, Spanish, and Vietnamese language cohorts. Total enrollment is 480 across the three. The program runs the same outcome survey at intake and at twelve weeks. Methodology questions are mapped to the five error sources and tracked across the program lifecycle. The deeper coding workflow lives in the multilingual survey analysis guide.
We knew funders would push back if we reported one outcome number for the whole program when forty percent of participants speak Spanish or Vietnamese. So we mapped the population first. Drew the frame to include all three language groups. Piloted the translated surveys in each language with native speakers from the cohort. Tracked response rate by language across the twelve weeks. By the time we wrote the methodology section, every error source was either reduced or honestly named.
Workforce program evaluation lead, end of cycle
Coverage
Diagnosed: Initial frame from the email list missed 28 percent of Vietnamese-speaking participants who registered through community partners and never received program emails.
Reduced by: extending the frame to include community-partner intake records. Coverage gap closed from 28 percent to 4 percent.
Sampling
Diagnosed: The cohort of 480 was small enough for census sampling. Sampling error was eliminated by surveying everyone in the frame rather than drawing a sample.
Reduced by: census sampling. Sampling error effectively zero against the frame.
Nonresponse
Diagnosed: Initial response rates were 71 percent English, 58 percent Spanish, and 38 percent Vietnamese. The 33-point gap threatened the cross-language comparison the report needed to make.
Reduced by: in-person follow-up by a Vietnamese-speaking staff member during the final two program sessions. Final response rates: 89 percent / 81 percent / 76 percent.
Measurement
Diagnosed: Cognitive interviews in pilot revealed that the Vietnamese translation of confident carried a stronger meaning than the English version, biasing ratings downward in that cohort.
Reduced by: rewording the Vietnamese anchor to a less absolute term, then re-piloting. Cross-language ratings became comparable.
Processing
Diagnosed: Open-ended responses in three languages would normally pass through a manual translation step before coding, introducing translation choices that could not be audited later.
Reduced by: coding in native language first, then aggregating themes across languages. Original responses preserved at every step. Audit trail intact.
A defensible cross-language comparison
Confidence gains comparable across all three language cohorts because measurement equivalence was confirmed in pilot. Cohort-by-cohort breakdowns hold up to funder review.
An honest methodology section
The report names every error source and what was done to reduce each. Reviewers see the residual coverage gap (4 percent), the response rate by language, and the translation methodology.
Native-language qualitative themes
Open-ended responses coded in source language preserve nuance that bulk pre-translation would lose. Themes connected to specific language cohorts and outcome groups.
Reduced rework on the next cohort
The methodology decisions documented for cohort one carry forward. Cohort two does not relitigate the translation methodology; it inherits it.
A single-language report with caveats
The Vietnamese cohort gets reported separately in a footnote. The headline number describes the English cohort by default. The cross-language outcome question goes unanswered.
A methodology section written from memory
Six weeks after fielding, the evaluator reconstructs what happened from emails. Response rate by language is approximate. Translation choices are not auditable.
English-first qualitative coding
Open-ended responses translated in bulk, then coded once. Translation drift compounds with coding decisions. The report cannot answer whether language cohorts described different experiences.
Same problems in cohort two
Methodology decisions were not documented because they were not deliberate. The next cohort starts from the same gaps, learns the same lessons, and produces the same caveats.
Sopact Sense was designed so methodology checks happen continuously rather than at report time. Native-language coding, response-rate-by-subgroup tracking, and audit trails across handoffs are part of the data architecture, not separate workflows. Methodology stops being something you remember to do and becomes something the platform makes hard to skip. The 480-participant program above produced a methodology-complete report at week thirteen. The same workflow on disconnected tools produces the report at week twenty-one with reconstructed methodology.
Methodology rigor varies with stakes. The contexts below differ in audience scrutiny, error tolerance, and the consequences of methodology gaps. The five-error framework holds across all three. The thresholds for acceptable error differ.
Multi-grantee, multi-program, comparable across portfolios.
The setup. A foundation funds 12 to 40 grantees, each running their own program with their own participants. The foundation needs comparable outcome data across grantees without dictating every question. Methodology gets harder because each grantee introduces its own coverage, sampling, and processing decisions.
What goes wrong. Each grantee designs its own frame. Some include all program participants, some only the ones who completed orientation, some only the ones who consented to follow-up. Coverage error compounds across the portfolio, and the foundation cannot tell whether differences in reported outcomes reflect program differences or methodology differences.
What works. A core stakeholder survey instrument shared across grantees, with shared frame definitions and a shared methodology section template. Each grantee retains program-specific questions but adheres to the methodology requirements. The foundation reports out as a true portfolio comparison rather than a portfolio of incomparable anecdotes.
Workforce-development foundation, 22 grantees, annual cycle. Shared core instrument with mandatory frame mapping. Foundation-level methodology section assembles grantee-level methodology disclosures. Cross-grantee error comparison enabled.
Three or more languages, equivalent measurement required.
The setup. Programs serving immigrant communities, refugee resettlement, multilingual workforce training, or international development typically run surveys across three to seven languages. Participants give responses in their preferred language; the report needs cross-language comparability for any aggregate claim.
What goes wrong. Translation drift drives most measurement error in multi-language settings. The English version of confident does not carry the same connotation as the Spanish or Vietnamese counterparts. Anchor labels on rating scales drift across translations. Reverse-scored items confuse translators. Processing error compounds when a manual translation step happens between collection and coding.
What works. Professional translation followed by independent back-translation, cognitive interviews in each language, and analysis tools that handle native-language responses without forcing English-first coding. The deeper workflow lives in the multilingual survey analysis guide, including theme extraction across languages.
Refugee resettlement program, 5 languages, 600 participants. Cognitive interviews per language. Native-language coding preserved. Cross-language comparison validated through anchor-vignette checks.
Statutory deadlines, audit-ready documentation.
The setup. Government-funded workforce programs (WIOA, apprenticeship, sector partnerships) report outcomes against statutory definitions on quarterly or annual schedules. The audience includes federal program officers, state oversight bodies, and increasingly the Government Accountability Office. Methodology gaps trigger audits.
What goes wrong. The single most common audit finding is response rate by subgroup. Programs report aggregate response rates but cannot break them out by race, language, age band, or program track. The aggregate looks acceptable; the subgroup distribution reveals that some populations are barely represented in the response data.
What works. Continuous tracking of response rate by every reporting subgroup. Documented nonresponse follow-up procedures. Multi-rater feedback structures for programs evaluating staff and partner performance, with audit-ready handoff trails. Methodology sections written for federal review rather than for the next program meeting.
State workforce board, 14 sector partnerships, federal reporting. Subgroup-level response rate tracking automatic. Audit trail preserved across all data handoffs. Methodology section assembled from continuous logs rather than reconstructed for each report.
General survey tools handle collection competently. The five-error decomposition exposes where they hand methodology back to the practitioner. Coverage and frame mapping happen outside the tool. Response-rate-by-subgroup tracking depends on connecting survey responses to a participant record the tool does not own. Native-language coding requires exporting open-ended responses into a separate analysis system. Every handoff is a place for processing error to enter.
Sopact Sense was designed so methodology checks happen continuously rather than at report time. Persistent participant records track response rate by every subgroup automatically. Native-language qualitative coding runs at submission rather than weeks later. Audit trails preserve every transition. The methodology section of the final report is assembled from continuous logs rather than reconstructed from memory.
The most common questions about survey methodology, total survey error, and how methodology relates to design. Each answer follows the five-error framework used throughout this guide.
Survey methodology is how you measure where your survey data is right and where it is not. It covers the five sources of error, the diagnostics that surface them, and the workflow that reduces error before results are reported. Every survey contains coverage, sampling, nonresponse, measurement, and processing error in some amount. Methodology is the discipline of finding which ones matter most for the decision the data must support and reducing those first.
Survey methodology means the practice of measuring data quality. It is what makes the difference between a survey that produces a number and a survey that produces a number you can defend. The methodology layer covers who you reached and missed, who responded and did not, what they answered and whether the answers were accurate, and what happened to those answers between collection and report.
Survey methodology covers five error categories that together describe total survey error. Coverage error is who the survey missed entirely. Sampling error is the variation introduced by surveying some rather than all. Nonresponse error is the bias from people who did not answer. Measurement error is what people answered that did not match reality. Processing error is what got changed or lost between answer and analysis. Each type calls for a different diagnostic and a different fix.
Survey methodology best practices reduce the five errors in priority order based on which one threatens the decision most. Map the population before sampling to reduce coverage error. Use probability sampling when feasible to make sampling error calculable. Plan nonresponse follow-up before fielding rather than reacting to low response. Pilot questions for measurement error before launch. Track error sources at each handoff between collection, coding, and analysis. Document the methodology so the next reader can audit it.
Survey design is the architectural decisions made before any question is written: type of design, identifier strategy, wave structure, analysis workflow. Survey methodology is the discipline of measuring whether those decisions worked once data starts arriving. Design controls what the data should be able to claim. Methodology controls how confidently it can claim it. Most programs need both. They run sequentially: design first, methodology continuously.
Total survey error, sometimes called the TSE framework, is a practitioner framework for diagnosing survey quality by decomposing total error into five sources: coverage error, sampling error, nonresponse error, measurement error, and processing error. The framework lets you see where error is concentrated rather than guess. A survey can have low sampling error and high nonresponse error at the same time, and the report needs to be honest about both.
Sampling error is the variation introduced because a sample is drawn from a frame rather than the whole population. Even with a perfect frame, picking 200 people out of 1000 produces estimates that differ from the true population value. Sampling error is calculable when probability sampling is used, reduces with larger samples, and never reaches zero unless the entire population is surveyed.
Nonresponse error is bias introduced when invited participants do not respond and the non-respondents differ systematically from respondents. A program survey with a 30 percent response rate where the responding 30 percent skews toward more engaged participants overstates outcomes for the whole cohort. The fix is not always more reminders. Sometimes it is a different mode, a different timing, or accepting that the response rate caps what the data can claim.
Measurement error is what participants answered that did not accurately reflect what the question intended to measure. Causes include confusing question wording, social desirability bias, recall failure for events long past, mode effects (people answer differently on phone versus paper), and translation drift across languages. Measurement error is reduced through pilot testing, cognitive interviews, and checking that translated versions ask the same thing.
Coverage error is error introduced when the people the survey can reach do not match the people the survey is trying to describe. The clearest example is a workforce program that surveys only English-speaking participants when forty percent of the cohort speaks Spanish. Those participants are not in the frame and cannot be measured no matter how good the questions are. Coverage error is fixed at the frame, not at the question level.
Processing error is error introduced between the moment a participant gives an answer and the moment the answer becomes a number in a report. Causes include data entry mistakes, miscoding of open-ended responses, incorrect aggregation, and version drift between collection and analysis tools. Processing error is the easiest to ignore because it happens between systems rather than during fielding, but it can swamp other error sources when handoffs are manual.
Six principles guide practitioner survey methodology. Map the population before drawing the sample. Use probability sampling when feasible. Plan nonresponse follow-up before fielding. Pilot for measurement validity before launch. Track error sources at every handoff. Document methodology so the next reader can audit it. Each principle reduces a different error source. Together they produce a survey that can be defended in front of a skeptical reader.
Multilingual survey methodology is the practice of running the same survey across multiple languages while keeping coverage, measurement, and processing error under control. The core challenge is that translation drift introduces measurement error: the Spanish version of confident may carry different connotations than the English version. The fix combines back-translation, cognitive interviews in each language, and analysis tools that handle native-language responses without forcing English-first coding. The deeper workflow lives in the multilingual survey analysis guide.
Survey methodology in research uses the same five-error framework as in applied program evaluation, with more rigorous sampling, larger samples, and explicit error reporting. Academic research methodology emphasizes probability sampling, weighting strategies, and statistical confidence intervals. Applied program methodology more often uses census sampling of available participants and reports findings as directional rather than statistically significant. The framework is the same. The thresholds for acceptable error differ.
The canonical sources for survey methodology are the academic literature on Total Survey Error, especially the work of Robert Groves and Paul Biemer, and standards published by the American Association for Public Opinion Research (AAPOR). For practitioner application, the same framework appears in nonprofit evaluation guides and federal statistical agency methodology handbooks. The shared takeaway across sources is the five-error decomposition and the practice of naming which errors threaten the specific claim being made.
The sister pillar on survey design covers the architectural decisions methodology measures. The five guides below cover specific methodology decisions in depth.
The architectural decisions made before any question is written. The input methodology measures. Five types, six principles, and the decision matrix.
Read the guide →Translation drift, native-language coding, and the analysis architecture that handles three or more languages without forcing English-first coding.
Read the guide →Methodology rules that hold across multiple waves: persistent identifiers, instrument consistency, and the response-rate diagnostics that compound over time.
Read the guide →Question precision for codeable narratives, with cognitive interview protocols that reduce measurement error in open-ended responses.
Read the guide →Anchor-label discipline, cross-language equivalence, and the measurement-error diagnostics specific to rating scale instruments.
Read the guide →Pairing quantitative and qualitative instruments so each compensates for the others' error sources. Methodology across coupled measurements.
Read the guide →Bring the methodology section of your most recent survey report, or the one that came back from a funder or board with questions you struggled to answer. We walk it against the five error sources, name the gaps in coverage, sampling, nonresponse, measurement, and processing, and show what a continuous-methodology workflow looks like in Sopact Sense. No procurement decision required.
A 60-minute working session, screen-share. Founder, not a sales rep.
Your most recent methodology section, or the response rate report that raised questions, or the question that data could not answer.
A five-error audit of your current methodology, plus a workflow sketch for the next survey cycle that holds across platforms.
