A one-off survey is a photograph. A standing panel is a time-lapse. Here is how to recruit it, pay it, keep it, and why wave three is when it starts paying you back. Sources are named in the text and linked at the end.

1. Why ask the same people again when I could just run a fresh survey?

• You measure change inside people, not just across crowds. A fresh sample shows the net number moved. A panel shows who moved, which way, and whether the switchers look alike.
• The statistics are shockingly generous. Two independent samples of 100 give you a margin of error of about ±13.9 points on a change in a 50% share. Ask the same 100 people, and if their answers correlate at 0.7 from wave to wave, that shrinks to ±7.6 points (Chart 5, our calculation). Same budget, half the noise.
• Joined people answer. Pew Research Center’s American Trends Panel (ATP) got 92% of sampled panelists to respond in its November 2025 wave (10,357 of 11,258), even though its cumulative response rate after recruitment and attrition is just 3% (Pew, 2026). The hard work is getting people in. After that, they show up.

2. Why a hundred, and how many do I actually need to recruit?

• One hundred is affordable and still directional. A single wave of 100 carries about ±9.8 points of error at 95% confidence. Great for 15-point shifts, useless for hair-splitting.
• Attrition compounds like a bad loan. Keep 80% of people each quarter and 100 becomes 64 by wave 3 and 51 by wave 4. Start with 140 and you still have about 90 at wave 3 and 72 at wave 4 (Chart 2). Recruit 130 to 150 and plan a top-up.
• The pros refill, too. Pew recruited ATP members in 2014, 2015 and 2017, then switched to address-based mail recruitment in 2018, explicitly to “replenish panel membership because of inevitable attrition” (Pew, 2015). Of 30,283 people who ever joined, 12,448 were still active in late 2022, about 41% (Pew, 2023).

3. Where do I find people who will actually stick around?

• Define the population in one sentence, then recruit only from it. Customers who bought in the last year, members of a trade group, heads of procurement at mid-size manufacturers.
• Ask warm people at a warm moment. When Pew invited respondents at the end of a long phone survey in 2014, 54.4% (5,338 of 9,809) said yes (Pew, 2015). The end of a webinar, a support call or an onboarding email works the same way.
• Recruitment quality echoes down the funnel. Two CDC panel vendors ended at cumulative response rates of 13.3% vs. 3.8%, mostly because of who they recruited and kept (Table 1).
• Over-recruit the people who leave first. Understanding Society, the huge UK household panel, finds attrition runs higher among younger people, ethnic minorities, lower-income households and people without qualifications (Understanding Society, 2025). Pad those groups by 20% to 30% at intake.

Table 1. Two panel funnels, one survey (CDC NCHS RSS, 2025)

Vendor	Recruitment rate	Panel retention	Survey completion	Cumulative response
Panel Provider 1	29.2%	80.8%	56.5%	13.3%
Panel Provider 2	8.8%	28.6%	70.5%	3.8%

Cumulative response is the product of the stages (AAPOR CUMRR1). A higher completion rate could not rescue a weak recruitment and retention stage.

4. What should I pay them, and when?

• Pay before you ask. Church’s classic 1993 meta-analysis found prepaid cash incentives lifted response by 19.1 points, while promised rewards lifted it only 3.8 points (Chart 3). Singer and Ye’s 2013 review of four decades of experiments agrees.
• Money has diminishing returns. Singer and colleagues (1999) estimated about one third of a point per dollar in interviewer surveys, with the curve flattening as amounts rise (FORS Guide 08).
• Copy the professionals. Pew paid $10 cash to join in 2014 and now pays $5 to $20 per survey, with more going to groups that are harder to reach (Pew, 2015; 2026). The Swiss Election Study found an unconditional first-wave incentive plus conditional payments later improved retention (Pekari, 2017).
• A workable recipe: a $10 welcome, $20 per completed wave, and a small bonus for finishing all four. For 140 people that is about $14,000 a year. Add the perk money can’t buy: panelists see the results before the public does. People love reading about themselves.

5. How bad will attrition get, and how do I slow it down?

• Expect a range from graceful to brutal (Chart 1). A young-adult panel interviewed every four months still had 84% of its original sample after six waves (American Sociological Review, 2017). The English Longitudinal Study of Ageing kept 47% of its first-wave members through every one of five waves (IJE, 2012). A five-wave COVID panel kept just 20% through all five (arXiv, 2022). Online panels lose 1% to 2% per month (AAPOR); the Understanding America Study loses 7% to 8% a year.
• Loyalty snowballs. In that same COVID panel, people who had answered before responded at 51% and 42% in later waves, while fresh invitees responded at just 14% to 19%. Median completion time fell every wave, too.
• Tactics that work: same week each quarter, under 10 minutes, two reminders, and share the findings. Keep people who miss a wave. In the ORANJ BOWL study, re-engaged dropouts were more often lower-income and African American, so chasing them made the sample more representative (2021).

Table 2. Retention benchmarks from real panels

Panel	Cadence	What happened	Source
Pew American Trends Panel	Roughly monthly	92% to 93% per-survey response; 3% cumulative	Pew, 2025 to 2026
Young-adult panel (TLT)	Every 4 months	84% of wave 1 left after 6 waves	Am. Sociol. Rev., 2017
UCL COVID-19 Social Study	Weekly / monthly	92.7% / 95.9% came back for a later wave	UCL, 2025
NZ Attitudes and Values Study	Yearly	68% to 86% wave-to-wave; 45% of T1 by T9	NZAVS, 2020

6. Why do the questions have to stay frozen?

• Because one word can move a number more than a year of real change. In the 2014 General Social Survey, 62% of Americans said the country spends too little on “assistance to the poor.” Only 19% said the same about “welfare.” And 49% said too much on welfare, versus 12% for assistance to the poor (AP-NORC, 2015; Chart 4). Across studies the gap averages 39 points.
• Build two modules. A frozen core (same wording, answer scale, order and position every wave) and a rotating module for timely topics. The core is your trend line.
• If you must change a question, split it. Ask half the panel the old version and half the new one for one wave, and you’ll know how much of the shift is you.

7. Won’t being in a panel change how people answer?

• Sometimes, but usually less than people fear. Pew tested its own panel and found no evidence of “panel conditioning” bias for news consumption, discussing politics, partisanship or voting, with only a slight uptick in voter registration (Pew, 2021).
• It is real for some items. In the Current Population Survey, repeat respondents increasingly claim to have left the labor force, pushing measured unemployment down (Halpern-Manners and Warren, 2012). Quiz-style knowledge questions suffer too.
• The fix is a refresh cohort. Add 20 to 30 newcomers at wave 4 and compare them to veterans.

8. Why is wave three where the panel finally gets valuable?

• Wave 1 is a baseline, which is a polite way of saying it interests you and your mother. Wave 2 is a comparison, and you can’t tell whether a change is real or wobble. Wave 3 is a trend, with a direction, a shape and a headline: “for the third straight quarter.”
• The math agrees. David Heise showed in 1969 that three waves let you separate measurement noise from real change, something two waves can’t do (American Sociological Review; Table 3). Researchers testing for conditioning in the GSS needed “at least three waves” for the same reason (Halpern-Manners et al., 2017).

Table 3. What wave three unlocks (Heise model, illustrative)

Correlation between waves	Value	What you can compute	Result
Wave 1 and wave 2	0.60	Reliability = (r12 x r23) / r13	0.80 (the question is solid)
Wave 2 and wave 3	0.60	Real stability, wave 1 to 3 = r13² / (r12 x r23)	0.56 (opinions truly moved)
Wave 1 and wave 3	0.45	Real stability each quarter = r13 / r23	0.75

With only two waves you see a 0.60 correlation and can’t tell whether it means a sloppy question or real change. The third wave settles it.

9. How does a panel turn into demand and distribution?

• Original research still wins, and fewer people are doing it. Orbit Media’s 2026 survey of 1,042 content marketers found just 13.9% report “strong results,” a 12-year low. Publishing original research improves those odds by about 50%, and marketers using six or more proven strategies hit 39% (Chart 6).
• Trend beats snapshot. Orbit has asked the same questions since 2014, which is exactly why its 2026 drop made news. A quarterly panel hands you four dated releases a year, each with a fresh “up from” or “down from.”
• Make it easy to cite. Publish a methods box the way Pew does: sample size, field dates, response rate and exact question wording. Journalists and AI answer engines both lift clean Q&A text and tables.

Table 4. A year-one plan

Quarter	What you do	What you publish
Q0	Recruit 140; intake profile survey; $10 welcome	Nothing yet. Resist.
Q1	Wave 1: frozen core + rotator	Baseline snapshot, short
Q2	Wave 2; chase non-responders	“What changed in 90 days”
Q3	Wave 3; run the three-wave checks	First real trend report
Q4	Wave 4; add a refresh cohort of 20 to 30	Annual report + open data tables

Bottom line: recruit more than you need, pay up front, keep the core boring and identical, and don’t judge the panel until wave three. Then you own a number that didn’t exist before.

Sources

1. Pew, Typology 2026 survey methodology (pewresearch.org)

2. Pew, Building the American Trends Panel (2015) (pewresearch.org)

3. Pew, ATP methods: 54.4% acceptance (2015) (pewresearch.org)

4. Pew, AI and jobs methodology (2023) (pewresearch.org)

5. Pew, Risks of Panel Conditioning (2021) (pewresearch.org)

6. CDC NCHS, Rapid Surveys System Round 7 (2025) (cdc.gov)

7. Singer & Ye, Incentives in Surveys, ANNALS (2013) (journals.sagepub.com)

8. Church (1993) prepaid vs. promised, summary (papersurvey.io)

9. FORS Guide 08, Incentives in surveys (iris.unil.ch)

10. Five-wave COVID-19 panel, arXiv 2209.11805 (arxiv.org)

11. TLT waves and attrition, Am. Sociol. Rev. (2017) (pmc.ncbi.nlm.nih.gov)

12. ELSA samples and response rates, IJE (2012) (pmc.ncbi.nlm.nih.gov)

13. NZ Attitudes and Values Study retention (pmc.ncbi.nlm.nih.gov)

14. AAPOR, Data Quality Metrics for Online Samples (aapor.org)

15. Understanding America Study, arXiv 2512.13270 (arxiv.org)

16. UCL COVID-19 Social Study retention (ncbi.nlm.nih.gov)

17. Understanding Society, Retention and attrition (understandingsociety.ac.uk)

18. ORANJ BOWL attrition and retention (ncbi.nlm.nih.gov)

19. AP-NORC, welfare vs. assistance (2015) (apnorc.org)

20. Programmatic Equivalence study (39-pt gap) (academia.edu)

21. Heise (1969) three-wave model, panelr (panelr.jacob-long.com)

22. Halpern-Manners et al., Conditioning in GSS (doi.org)

23. Panel conditioning in CPS labor items (pmc.ncbi.nlm.nih.gov)

24. Orbit Media, Blogging Statistics 2026 (orbitmedia.com)