{"site":{"name":"Koji","description":"AI-native customer research platform that helps teams conduct, analyze, and synthesize customer interviews at scale.","url":"https://www.koji.so","contentTypes":["blog","documentation"],"lastUpdated":"2026-10-03T19:24:54.426Z"},"content":[{"type":"documentation","id":"48066a46-1c66-4910-baeb-e49693be96f7","slug":"satisficing-survey-responses","title":"Satisficing: Why Respondents Give You Good-Enough Answers Instead of True Ones","url":"https://www.koji.so/docs/satisficing-survey-responses","summary":"Satisficing is Krosnick's (Applied Cognitive Psychology, 1991, 5(3):213-236) account of respondents giving adequate rather than optimal answers. Optimizing requires comprehension, retrieval, judgment and response mapping; weak satisficing skimps on these and adds noise, while strong satisficing skips retrieval and judgment entirely and takes a cue from the question, adding systematic error sourced from your wording rather than your respondent. Likelihood rises with task difficulty and falls with respondent ability and motivation, so task difficulty is the main controllable lever. One mechanism produces acquiescence, status-quo endorsement, non-differentiation and straightlining, don't-know selection, first-plausible-option choice and random answering. Response-order effects reverse by mode: primacy under visual presentation, recency under oral, so mixing or switching mode can masquerade as opinion change.","content":"Satisficing is what a respondent does when they give you an answer that is good enough to pass rather than the answer that is true. It is not laziness and it is not fraud. It is a predictable response to a task that costs more effort than the respondent is willing to spend, and it is the single mechanism underneath a whole family of problems you may currently treat separately: acquiescence, straightlining, don't-know answers, middle-option clustering and response-order effects.\n\nUnderstanding it as one mechanism is what lets you design it out, instead of building a separate detector for each symptom.\n\n## The short answer\n\nThe concept comes from Jon Krosnick, \"Response strategies for coping with the cognitive demands of attitude measures in surveys,\" *Applied Cognitive Psychology*, 1991, volume 5, issue 3, pages 213 to 236 -- a paper with well over two thousand citations, and still the clearest account of the idea.\n\nThe argument in three moves:\n\n1. **An honest answer is expensive.** Answering a question properly takes four distinct cognitive steps, each of which can be skipped.\n2. **Respondents economise predictably.** When the task is hard, their ability is limited, or their motivation is low, they shortcut -- and the shortcuts they take are specific and recognisable.\n3. **The shortcuts look like separate biases but are one behaviour.** Agreeing with whatever you assert, giving every item in a grid the same rating, picking \"don't know\", choosing the first plausible option: these are all the same economy under different question formats.\n\nSo the lever is not detection. It is reducing the cost of answering honestly, and removing the cheap escape routes your question format offers.\n\n## The four steps an honest answer requires\n\nTo answer an attitude question properly, a respondent has to:\n\n1. **Comprehend** the question -- work out what is actually being asked.\n2. **Retrieve** relevant information from memory.\n3. **Judge** -- integrate what they retrieved into a summary position.\n4. **Respond** -- map that position onto the options you offered.\n\nEvery one of those steps costs effort, and step 4 is where your design either helps or sabotages them. A respondent who has done the first three steps well and then cannot find their actual view among your options will either pick something inaccurate or give up and shortcut.\n\n## Weak and strong satisficing\n\nKrosnick's framework distinguishes two degrees, and the difference matters practically.\n\n**Weak satisficing:** the respondent goes through all four steps, but less thoroughly. They retrieve less, think less carefully, and accept the first answer that seems defensible rather than searching for the best one. The answer is related to their real view, just noisier and biased toward whatever was easiest to reach.\n\n**Strong satisficing:** the respondent skips retrieval and judgment altogether. They look for a cue in the question itself that lets them produce a plausible-looking answer without consulting their own experience at all. The answer is not a noisy version of their view. It is not about their view.\n\nThis is why treating satisficing as \"noise\" is a mistake. Weak satisficing adds noise. Strong satisficing adds *content that came from your question wording rather than from your respondent*, which is a systematic error pointing in whatever direction your question leaned.\n\n## When it happens\n\nKrosnick frames the likelihood of satisficing as rising with task difficulty and falling with respondent ability and motivation. That gives you three distinct levers, and the useful observation is that only some of them are yours to pull:\n\n| Factor | Direction | Can you change it? |\n| --- | --- | --- |\n| Task difficulty | Harder task, more satisficing | Yes, and this is your main lever |\n| Respondent ability | Lower ability, more satisficing | Not directly, but you can lower the demand |\n| Respondent motivation | Lower motivation, more satisficing | Partly -- framing, relevance, length, incentive |\n\nTask difficulty is the one you control almost completely, and it is mostly made of things you chose: how abstract the question is, how many options there are, how much memory search it demands, how long the instrument already ran before this question.\n\nThere is also a stable individual component. Sturgis and Brunton-Smith, writing in *Public Opinion Quarterly* in 2023 (volume 87, issue 3, pages 689 to 718), report that \"People who score high on Conscientiousness and Agreeableness were less likely to be in the top decile of straightlining and midpoint distributions,\" and that they find \"large effects of these personality dimensions on the propensity to satisfice in both face-to-face and self-administration modes.\" Two implications: some of this travels with the person rather than the instrument, and it is not fixed by switching mode.\n\n## The fingerprints\n\nKrosnick catalogued the specific tactics respondents use. Each is a familiar \"bias\" in its own right, and each is this same economy wearing a different question format:\n\n- **Choosing the first reasonable option** rather than reading to the end. Produces response-order effects.\n- **Agreeing with whatever is asserted.** This is [acquiescence bias](/docs/acquiescence-bias), and it is why agree-disagree batteries are the most satisficing-prone format in common use.\n- **Endorsing the status quo** over any proposed change, because evaluating a change is work.\n- **Failing to differentiate** among the items in a battery -- non-differentiation, visible as straightlining down a grid.\n- **Answering \"don't know\"** when a don't-know option is offered as an easy exit.\n- **Selecting more or less randomly** when no cue is available -- mental coin-flipping.\n\nThe practical consequence: if you see straightlining and separately see acquiescence and separately see heavy don't-know use, you do not have three problems. You have one instrument that is too expensive to answer honestly.\n\n## Response-order effects: why voice and text fail differently\n\nOne prediction of the theory has a direct bearing on how you collect data, and it is the detail most often missed.\n\nSatisficing predicts response-order effects, because a respondent who stops searching early is disproportionately likely to choose whatever they encountered first. But which end of the list benefits depends on how the list reaches the respondent, and the long-standing pattern in the survey methodology literature is that visually presented options produce primacy effects -- the early options win -- while orally presented options produce recency effects, where the last options heard win.\n\nThe mechanism is straightforward once stated. Reading a list, a respondent processes the top items most deeply and can stop whenever something plausible appears. Hearing a list, they have only the most recent items still in working memory when the question ends.\n\nThis matters because it means **the same question with the same options can shift its results depending on whether you ran it in text or in voice**, and in opposite directions. If you mix modes, or switch modes between waves, order effects can masquerade as a real change in opinion. Two defences: randomise option order within each mode, and keep mode constant when you are comparing waves. See [mode effects](/docs/mode-effects-mixed-mode-research) for the broader version of this problem.\n\n## What actually reduces satisficing\n\nIn rough order of effect:\n\n**Cut the task, not the question count alone.** Replacing one abstract attitude question with a concrete behavioural one removes more satisficing than deleting three easy questions. \"How satisfied are you with our reporting?\" is expensive. \"When did you last export a report, and what did you do with it?\" is cheap and more informative.\n\n**Abandon agree-disagree batteries.** They combine every risk factor: an assertion to acquiesce to, a grid to straightline down, and no requirement to retrieve anything. Use construct-specific scales where each item's options name the actual thing being measured.\n\n**Stop offering a free exit.** A prominent don't-know option is an invitation to strong satisficing. Where genuine non-opinion matters, ask about it separately rather than inlining it as an escape.\n\n**Randomise order.** It does not reduce satisficing but it stops it biasing a specific option, converting a systematic error into noise you can at least reason about.\n\n**Shorten the instrument.** Motivation decays through a study, so satisficing concentrates late. This is the same mechanism behind [survey fatigue](/docs/survey-fatigue) and breakoff, and [ideal survey length](/docs/ideal-survey-length-guide) covers the trade-offs.\n\n**Intervene in real time.** Conrad, Couper, Tourangeau and Zhang, in *Survey Research Methods* (2017, volume 11, issue 1, pages 45 to 61), tested prompting respondents who answered implausibly fast and found that \"This prompting technique reduced speeding on subsequent questions compared to a no prompt control.\" A live, non-punitive nudge changes behaviour in a way post-hoc cleaning cannot.\n\n## How Koji handles this\n\nSatisficing is an economics problem, and Koji attacks the cost side rather than detecting the symptom.\n\n- **Conversation lowers comprehension cost.** A respondent who misunderstands a question in a form has no recourse and will shortcut. Koji's AI interviewer can rephrase, give an example, and confirm understanding, which removes the most common source of task difficulty.\n- **Adaptive follow-ups close the strong-satisficing escape.** Strong satisficing works by harvesting a cue from the question and never consulting experience. That fails when the next question is generated from what the respondent just said and asks for a specific. The built-in `mom_test` framework is designed around exactly this, with patterns like \"Walk me through how you currently handle [task]\" and \"What happened after that?\" -- behaviour probes that cannot be answered from a cue.\n- **Six typed question formats let you avoid the worst ones.** Koji supports `open_ended`, `scale`, `single_choice`, `multiple_choice`, `ranking` and `yes_no`. That range means you never need an agree-disagree grid: use `ranking` where you genuinely want discrimination between items, `scale` for a single calibrated judgment, and `single_choice` or `multiple_choice` with named options instead of a battery of assertions.\n- **Ranking defeats non-differentiation structurally.** A respondent cannot straightline a `ranking` question. The format requires discrimination, which is precisely the work non-differentiation avoids.\n- **Quality scoring surfaces it per interview.** Koji scores each interview 1-5 across relevance, depth and coverage. A satisficed conversation reliably scores low on depth even when every question was technically answered, which gives you a per-interview signal rather than a post-hoc guess.\n- **Mode is explicit, so order effects stay controllable.** Because Koji records whether an interview ran in voice or text, you can hold mode constant across waves and avoid confounding a primacy-to-recency flip with real change.\n\n## Common mistakes\n\n- **Treating the symptoms as separate problems.** Building a straightlining detector, an acquiescence correction and a don't-know rule separately, while leaving the instrument that causes all three untouched.\n- **Adding an attention check and calling it fixed.** [Attention checks](/docs/attention-check-questions) find some low-effort respondents. They do not reduce the task difficulty that produced the behaviour.\n- **Reading straightlining as a bad respondent.** It is usually a bad grid. The same person answers a well-formed question carefully.\n- **Assuming a long answer means an optimised one.** Weak satisficing produces plausible, fluent, adequately-long answers. Length is not evidence of retrieval.\n- **Switching mode mid-study.** Changing between text and voice can invert your response-order bias, and the shift will look like a change in customer opinion.\n- **Offering don't-know prominently to seem unbiased.** It raises data loss and invites strong satisficing. Measure genuine non-opinion deliberately instead.\n\n## Frequently asked questions\n\n### What is satisficing in survey research?\n\nSatisficing is giving an answer that is good enough to pass rather than the answer that is accurate. The term was applied to survey response by Jon Krosnick in 1991, who argued that answering properly requires comprehending the question, retrieving information, forming a judgment and mapping it onto the options offered, and that respondents skip or skimp on those steps when the task is hard or their motivation is low.\n\n### What is the difference between weak and strong satisficing?\n\nIn weak satisficing the respondent completes all four steps but less thoroughly, accepting the first defensible answer instead of searching for the best one, so the answer is a noisier version of their real view. In strong satisficing they skip retrieval and judgment entirely and take a cue from the question to manufacture a plausible answer. That second case is more damaging, because the answer reflects your question wording rather than the respondent.\n\n### Is straightlining the same thing as satisficing?\n\nStraightlining is one visible form of it, specifically the non-differentiation tactic applied to a grid. Acquiescence, status-quo endorsement, don't-know selection, picking the first plausible option and effectively random answering are the others. Treating them as one underlying behaviour is more useful than detecting each separately, because they share a cause and therefore a remedy.\n\n### Why do voice and text interviews produce opposite order effects?\n\nBecause of what remains available to the respondent at the moment of answering. Reading a list of options, they process the earliest ones most deeply and can stop as soon as something plausible appears, which produces primacy effects. Hearing the options read aloud, only the most recent remain in working memory, which produces recency effects. The practical rule is to randomise option order and to keep mode constant when comparing waves.\n\n### Does offering a don't-know option make my data more honest?\n\nUsually the opposite. A prominent don't-know option is the cheapest available escape from a demanding question, so it attracts respondents who hold a view but do not want to do the work of reporting it, mixing them in with genuine non-opinion. If distinguishing real non-opinion matters to your study, measure it with a separate deliberate question rather than inlining it as an easy exit.\n\n### How do I know if satisficing is affecting my study?\n\nLook for its fingerprints together rather than one at a time: agreement rates that rise on agree-disagree items, identical ratings down grids, don't-know use concentrated on the hardest questions, and all of these intensifying in the second half of the instrument. Koji's per-interview quality scoring also helps, because a satisficed conversation scores low on depth while still appearing complete.\n\n## Related Resources\n\n- [Structured Questions Guide](/docs/structured-questions-guide) - the six question types, and choosing formats that resist satisficing\n- [Acquiescence Bias](/docs/acquiescence-bias) - the agree-with-anything tactic in detail\n- [Attention Check Questions](/docs/attention-check-questions) - what detection can and cannot do here\n- [Mode Effects](/docs/mode-effects-mixed-mode-research) - why voice and text can give different answers\n- [Survey Fatigue](/docs/survey-fatigue) - why satisficing concentrates late in an instrument\n- [Ideal Survey Length](/docs/ideal-survey-length-guide) - trading coverage against respondent effort","category":"Research Methods","lastModified":"2026-10-02T03:47:16.127357+00:00","metaTitle":"Satisficing in Surveys: Why Respondents Give Good-Enough Answers","metaDescription":"Satisficing explains acquiescence, straightlining and don't-know answers as one behaviour. The mechanism, and how to design it out.","keywords":["satisficing survey responses","weak and strong satisficing","krosnick satisficing","straightlining","non-differentiation","response order effects"],"aiSummary":"Satisficing is Krosnick's (Applied Cognitive Psychology, 1991, 5(3):213-236) account of respondents giving adequate rather than optimal answers. Optimizing requires comprehension, retrieval, judgment and response mapping; weak satisficing skimps on these and adds noise, while strong satisficing skips retrieval and judgment entirely and takes a cue from the question, adding systematic error sourced from your wording rather than your respondent. Likelihood rises with task difficulty and falls with respondent ability and motivation, so task difficulty is the main controllable lever. One mechanism produces acquiescence, status-quo endorsement, non-differentiation and straightlining, don't-know selection, first-plausible-option choice and random answering. Response-order effects reverse by mode: primacy under visual presentation, recency under oral, so mixing or switching mode can masquerade as opinion change.","aiPrerequisites":["Familiarity with survey question formats","Basic understanding of response bias"],"aiLearningOutcomes":["Define satisficing and distinguish its weak and strong forms","Name the four cognitive steps an optimised answer requires","Predict when satisficing will occur from difficulty, ability and motivation","Recognise the six behavioural fingerprints as one mechanism","Explain why voice and text produce opposite response-order effects","Choose question formats and interventions that lower task difficulty"],"aiDifficulty":"intermediate","aiEstimatedTime":"13 min read"}],"pagination":{"total":1,"returned":1,"offset":0}}