What is a good survey response rate?

Why the published numbers do not line up

The figures in the table measure different things, and that is the first lesson. Pew's telephone rates come from randomly sampled phone numbers, where every call that goes unanswered counts against the rate. SurveyMonkey's platform figures count people who started a survey against everyone who received it, and they mix every kind of sender, from brands emailing loyal customers to researchers emailing strangers. The Census Bureau's American Community Survey is a government household survey with years of follow up built in. None of them is a target for a small business emailing its own customers. What they do show is the range: the same activity of answering questions can attract six people in a hundred or eighty. The honest benchmark for your survey is your own previous rounds, sent the same way to the same kind of audience, because every other number carries somebody else's definition, channel and relationship.

A low rate is not automatically a biased result

The single most useful finding in survey methods for anybody worrying about their rate is that the link between response rate and accuracy is weak. Pew Research Center reported in 2019 that its studies in 1997, 2003, 2012 and 2016 found little relationship between response rates and accuracy, while cautioning that a low rate does signal a higher risk of error. The risk becomes real when taking part is connected to the answer. In Pew's 2017 study, telephone respondents said they had volunteered at a rate of 46 percent, against 8 percent in the Current Population Survey, a gap consistent with the people who agree to surveys also being the people who give their time elsewhere. A satisfaction survey is exposed in the same way: the very pleased and the very annoyed answer, and the quietly indifferent do not.

What shapes the rate you will get

Relationship matters most. People who chose to hear from you, who have just finished an interaction, or who work for you, respond far more than strangers. Channel matters next, and the platform figures show how much: a survey arriving in an inbox from a known sender behaves very differently from a popup interrupting a page. Timing matters because interest fades quickly, so a question about a delivery works best while the parcel is still on the doorstep. Length matters, but mostly through the promise in the invitation, since people decide whether to start from what they are told. Topic matters in both directions, lifting the rate among people with strong views and lowering it among everybody else, which is exactly the pattern that turns a decent rate into a misleading result.

Raising the rate without buying bad data

The dependable methods are unglamorous. Dillman's Tailored Design Method treats the survey as a planned sequence of contacts rather than a single email: an invitation that explains why this person was asked and what happens with the answers, a reminder, and a final contact. State an honest length and keep it. Send from a recognisable person or team. Make the first question easy and relevant. Methods that inflate the number can damage the data: a large prize attracts people who want the prize, and repeated pressure produces rushed answers from people who want the messages to stop. A modest rate from a well defined group, with a reminder and a record of who was invited, is usually a better result than a high rate from a link passed around until the numbers looked respectable.

Reading your own rate from a form

The responses table holds submissions, one row each, so the numerator is easy. The denominator has to come from wherever the invitation was sent, which is why it is worth keeping the delivery count from the mailing tool rather than rounding it from memory. The CSV export places Submitted At in the first column, so responses can be plotted by day to see how much each reminder added and when a survey has effectively stopped collecting. Test runs should use the demo link, which keeps the form interactive but records nothing, so internal testing never pads the count. Report the number invited, the number who responded and the dates alongside the rate.

Questions people ask