First request
Small sign in the window
Some refuse at this point already
Agreed
Days later
Large sign in the front garden
A much bigger request
Result
More agreement than without the first request
Evidence
Freedman and Fraser (1966) ran two field experiments in Palo Alto. In the first they telephoned 156 housewives: those who had first answered eight questions about household soaps were, three days later, more likely to let a team into their home for two hours (52.8 versus 22.2 percent). Only the second experiment took place at the front door, and it is the better known one: there the issue was road safety, a different person made the second request, and 76.0 percent agreed to the large garden sign when the first request matched in both issue and task — against 16.7 percent with no first request.
The small request is not granted as a matter of course: in the first experiment only about two thirds agreed, and nobody who refused it later agreed to the large request.
The meta-analysis by Beaman et al. (1983), covering 120 experimental groups, finds the effect reliable but small. Across all groups the mean effect size is 0.09; across the 85 tests the authors class as clean tests it is 0.21 — with about 30 percent of the individual values at zero or negative. For scale: 0.2 counts as a small effect, 0.5 as medium and 0.8 as large. The 0.67 of Freedman and Fraser’s own six conditions has, according to Gamian-Wilk and Dolinski, never been reached again.
Burger (1999) shows that implementation decides. The technique works best when the first request is actually carried out, costs some effort, and the second looks like its continuation (r = 0.25, compliance almost double the control group). It tips easily when the same person asks for something different immediately afterwards: among 24 such studies, exactly half found more compliance in the control group.
A direct replication succeeded in only one attempt out of four. Gamian-Wilk and Dolinski (2020) repeated the first experiment: in 2003 it worked in Ukraine (53.3 versus 3.3 percent), not in Poland, and in 2013 in neither country. The authors attribute this to market-research calls now being commonplace everywhere.
What decides it is who is asking. Dillard, Hunter and Burgoon (1984) analysed 37 estimates from 28 reports with 4,927 participants. Across all studies the mean correlation is r = 0.11, with individual values ranging from −0.23 to 0.39. The r measures how closely two quantities go together; 0 means not at all, and values around 0.1 count as small.
When the authors split by type of sender, only one half of the effect survived.
| Who is asking | r |
|---|---|
| non-profit cause, no payment involved | 0.17 (across 2,972 participants) |
| self-oriented sender, explicitly including marketing firms | 0.02, with no residual variance at all |
Their own conclusion: the technique is best left to those running prosocial campaigns.
Limits
The studies come from personal requests, not from web forms or checkouts. Freedman and Fraser limit their own scope: all requests came from non-profit organizations and concerned deliberately uncontroversial issues. Whether the technique also works when selling a product remains to be shown, they write. For multi-step funnels the transfer is a hypothesis — and the configuration closest to a checkout (same party, next request immediately) is the least reliable one in Burger’s review.