The rate and its denominator
A response rate is completed interviews divided by eligible contacts, and every dispute about the number is really a dispute about the denominator. The conservative standard (the AAPOR convention used by serious survey houses) keeps refusals, break-offs and unreachable-but-eligible contacts in the denominator, and removes only contacts who were never eligible: wrong numbers, people outside the target population, businesses in a household survey. Computed that way the rate is a hard measure of the whole recruitment effort, not a compliment to it.
1,200 invitations, 400 people start the survey, 342 finish it. Response rate: 342 ÷ 1,200 = 28.5%. Start rate: 33.3%. Completion rate among starters: 342 ÷ 400 = 85.5%. Three honest numbers describing three different stages: recruitment pulled a third in, the survey itself kept six of every seven who began. Reporting only the 85.5% as "the response rate" is the commonest flattering error in survey write-ups.
What a low rate does and does not mean
A low response rate raises the risk of nonresponse bias; it does not automatically create it. Bias needs the nonrespondents to differ from respondents on the things being measured. A 25% rate on a survey whose respondents match the population on known benchmarks can be safer than a 70% rate from a skewed frame. The professional habit: report the rate by the standard definition, compare respondents against known population margins, and say plainly what could not be checked. Chasing the rate itself (relentless callbacks, incentives that change who answers) can distort the sample while improving the headline number.
Counting partial interviews inconsistently. Decide before fieldwork what counts as complete (the standard is the core questions answered), log break-offs separately, and do not quietly promote partials into the numerator when the rate looks thin.
Frequently asked questions
What is a good response rate?
Context decides: probability-based phone surveys now often land under 10%, well-run organisational surveys 30 to 50%, employee surveys higher. The trend matters less than whether respondents resemble the population.
Do partial completions count in the numerator?
Only if your pre-set definition of complete is met. AAPOR defines several rates (RR1 to RR6) that differ exactly on this point; state which one you are reporting.
How do I improve a response rate without biasing the sample?
Shorter instruments, clear sender identity, reminders spaced over days, and modes the audience actually uses. Avoid incentives or pressure that selectively pull in one kind of respondent.