Anchors ha¨e a suggestive effect on the answers to binary questions. With scales
bounded on one side such as the dollar scale. this effect causes an upward bias in
binary answers, relative to corresponding open-ended responses. In the context of CV
surveys, this bias explains the discrepancy previously observed between estimates of
WTP from referendum questions and from open-ended questions. The design em-
ployed by Jacowitz and Kahneman 1995. allows a comparison between two
proportions: i. the proportion of respondents in the original group unanchored.
who spontaneously offered an estimate higher than an anchor; ii. the proportion
of respondents in the anchored group who stated that the same anchor was lower
than the true value of the quantity. In the absence of bias, the two proportions
should be the same. However, the results showed a pronounced bias: on average,
respondents in the anchored group judged the high anchor to be lower than the
true value on 27% of occasions, very significantly more than the 15% expected
from the responses of the unanchored group. Furthermore, there was a pronounced
asymmetry in the bias: the low anchors were judged to be too high on only
14% of occasions. The asymmetry was due to the prevalence of estimation
problems in which the range of possible answers is bounded by zero, e.g., the
height of the tallest redwood. The result of this bias, of course, is that the estimates
inferred from the binary question were generally much higher than the estimates
obtained directly from open-ended questions. The discrepancy between the two
response modes is similar to the discrepancy observed in CV research between
estimates of WTP derived from open-ended and from referendum questions
Desvousges et al., 1992; McFadden, 1994..