research methods
Recall bias
What is recall bias and how does it affect research?
Recall bias is the gap between what people remember doing and what they actually did. Human memory isn't a recording; it's a reconstruction - lossy, smoothed, and skewed toward the vivid, the recent and the emotional. So any research that asks people to remember their past behaviour collects an edited highlights reel, not an accurate log.
Also known as: recall bias, memory bias, retrospective bias
Prefer to watch?
Watch the recap 0:51
The demo
“How often do you use the app?” Compare what a user remembers with what their usage log actually recorded - and mind the gap.
What this demo shows (text version)
The same question - how often do you use the app? - is answered two ways. The "what they remember" view gives a vague, deliberate-sounding self-report: "a couple of times a week, when I actually need something." The "what the logs show" view records the reality: dozens of short opens, mostly habitual and barely conscious. The two don't match, and nobody lied - memory simply kept the intentional moments and dropped the routine ones. (The figures illustrate the effect rather than report real data.)
That gap is recall bias: memory is a reconstruction skewed toward the recent, unusual and emotional, so retrospective self-report of "how often / how long / when" is unreliable. The fixes are to shrink the gap between event and question (diary studies, in-the-moment prompts) or, better, to measure behaviour directly with analytics instead of asking people to remember it.
When you ask people to recall their behaviour, you get a reconstruction, not a record: routine actions are forgotten, frequencies are guessed, and the dramatic moments crowd out the typical ones. So self-reported "how often / how long / when" is unreliable. Capture experiences close to when they happen (diary studies), or measure behaviour directly (analytics), rather than asking people to remember it weeks later.
Ask them and it's "a couple of times a week, when I need it." Check the logs and it's thirty-odd quick opens, mostly out of habit. Nobody's lying - their memory just kept the deliberate moments and quietly dropped the routine ones. That gap between remembered and actual behaviour is recall bias, and it's why "how often do you…?" answers can't be trusted.
Memory is reconstructive, so it fails in patterned ways: habitual, low-attention actions barely register and get under-counted; frequencies and durations are estimated, not recalled; and what survives is biased toward the recent, the unusual and the emotionally charged (entangled with the peak-end rule). The longer the gap between the event and the question, the worse it gets.
This makes retrospective self-report a weak foundation for "how often", "how long" or "when" questions. Surveys and interviews that lean on memory - "how many times last month did you…?" - gather confident guesses dressed as data. It compounds with social-desirability bias when the truth is also a little embarrassing.
The fixes are about timing and measurement. Shrink the gap: capture experiences in the moment with diary studies, experience sampling or quick post-task prompts. Better still, measure the behaviour directly with analytics rather than asking people to remember it. And when you must ask retrospectively, anchor to specifics ("the last time you did X") rather than vague averages.