Every form tool sells you the same number. Completion rate. Response rate. The percentage of people who hit submit. Typeform shows it on the dashboard. Google Forms surfaces it in the summary. The entire industry treats this number as the scoreboard — higher is better, full stop.
In 2006, Robert Groves published a paper that should have ended this. “Nonresponse rates and nonresponse bias in household surveys” examined the relationship between how many people respond and whether those responses are any good. The correlation was weak. Sometimes it was nonexistent. A high response rate did not predict low bias. A low response rate did not predict high bias. The two metrics were, in practice, decoupled.
Response rate vs. response quality · weak correlation
Groves & Peytcheva (2008): the relationship between nonresponse rate and nonresponse bias is empirically weak.
The number everyone optimizes is the wrong one
Groves and Peytcheva followed up in 2008 with an even more pointed analysis. They examined 59 methodological studies and found that the relationship between nonresponse rate and nonresponse bias was “empirically weak.” Their conclusion was blunt: nonresponse rate is a poor predictor of nonresponse bias. You cannot look at a 95% completion rate and conclude the data is good. You cannot look at a 60% rate and conclude the data is bad.
Holbrook, Krosnick, and Pfent reinforced this in 2007. They studied whether high cooperation rates — the survey researcher's gold standard — produced better data. They didn't. High-cooperation samples showed no consistent advantage in representativeness or quality. The teams that worked hardest to get everyone to respond didn't end up with better answers. They ended up with more answers, many of them from people who didn't want to be answering.
This is counterintuitive. More responses should mean more signal. But it only works that way if the additional responses carry information. A respondent who types “fine” into three textareas and hits submit adds to your completion rate and subtracts from your data quality.
Completion is not commitment
There's a distinction the form industry doesn't make. Completion means someone clicked submit. Commitment means someone thought about the questions and wrote something specific enough to act on.
A five-word response that names a date, a person, or a system carries more information than a thirty-word response that says nothing. “Deploy pipeline broke Wednesday, took 4 hours to fix” is eleven words with three facts. “Things are generally going pretty well, a few minor bumps along the way but nothing too concerning” is seventeen words with zero facts.
Form tools count both as completed responses. Both contribute equally to the 94% bar on the dashboard. The distinction between them — the difference between signal and noise — is invisible in every analytics view the industry ships.
Dashboard · what you measure shapes what you get
Why nobody measures quality
Response rate is popular because it's easy to measure. You count submissions, divide by recipients, get a percentage. Done. Clean number. Goes in the quarterly review.
Response quality is hard to measure. What makes an answer “good”? Word count is a proxy, but a verbose non-answer is still a non-answer. Sentiment analysis tells you whether the words sound positive, not whether they contain information. Topic modeling catches themes but misses specificity.
The useful signal hides in structural features. Does the response contain named entities — people, teams, systems, dates? Does it reference specific events? Does it include numbers? These are the markers of a respondent who actually retrieved information and formed a judgment, rather than one who typed the first plausible sentence and moved on.
Groves didn't prescribe a quality metric in his 2006 paper. He argued for a shift in how the field thinks about survey error — away from response rates and toward bias measurement. The form industry mostly ignored him. Nineteen years later, the completion percentage is still the headline number.
The metric shift
Here is what a quality-focused dashboard shows you that a rate-focused one does not.
Specificity score. What percentage of responses contain at least one named entity, date, or number? A sprint retro where 80% of answers name specific tickets or dates is producing signal. One where 15% do is producing noise, regardless of the completion rate.
Word count distribution. Not average word count — the distribution. A bimodal distribution (some people wrote 80 words, most wrote 5) tells you more than the mean. It tells you who committed and who satisficed.
Source breakdown. Did the respondent write from scratch, or did they start with an artifact-grounded draft and edit it? Assisted responses tend to be more specific because the draft contains facts by default. The respondent adds judgment. The form sender gets both.
None of these metrics are hard to compute. They're just not what anyone tracks. The industry settled on completion rate as the success metric, and the incentive structure reinforced it: tools compete on who gets more people to hit submit, not on who gets people to say something worth reading.
What changes when you measure signal
When you optimize for response quality instead of response rate, the design decisions change. You stop adding incentives to get more people to submit. You start asking whether the submissions you have are specific enough to act on. You stop shortening the form to reduce dropoff. You start giving respondents a draft to edit so their answers contain facts instead of platitudes.
Groves was right in 2006. The form industry just hasn't caught up. The metric that matters is not “did they respond?” It's “can you act on what they said?”
See how Pluck measures response quality, not just completion →