bid differences are significantly different from zero with p values of less than 0.01 according to a
Wilcoxon signed rank test4. The mean bids for quality B are significantly higher than the mean bids
for quality V, even though they have the same median tenderness. Before tasting, the bid difference
between B and V may partly be explained by the fact that the average valuation of A and C is 3.5%
less than the valuation of B. However, after tasting, the average valuation of A and C is 1% higher
than the valuation of B.
Tasting
We randomly drew one fillet from each quality for tasting. The fillet representing quality V would
have been categorized as very tender (quality A). Consequently, the fillet representing quality B was
less tender than the fillet representing quality V and tasting appears to have affected the bidding in R3
and R4. The mean bid for qualities A, C, and V increased by NOK 0.61 (p = 0.29), 2.67 (p = 0.12) and
0.88 (p = 0.17), respectively, from R2 to R3, whereas the mean bid for quality B decreased by NOK
2.84 (p = 0.23). None of these mean bids changed significantly between R2 and R3, although the mean
bid for B was reduced compared with the bids for A (p = 0.06), C (p = 0.01), and V (p = 0.02). These
changes indicate that the participants used the information obtained from tasting to update their beliefs
about the four qualities, even though they were instructed that the loins were drawn randomly.
Rising bids
Figure 2 illustrates the rising mean bids throughout the trials. With the exception of a decline in the
mean bid for quality B after tasting, the mean bids for all qualities are stable or rising. Rising bids in
multi-trial auctions are one of the most persistent results in experimental valuation literature and are
usually explained as learning effects. Framing and learning effects in multi-trial auctions have been
studied in a number of papers (e.g., Knetsch, Tang, and Thaler, 2001; List and Shogren, 1999;
Shogren et al. 1994a; Shogren, List, and Hayes, 2000). List and Shogren (1999) also found some
evidence of bids being correlated with posted prices, but concluded that most of the increase in bids
may be explained by participants learning their optimal strategy. The only study not reporting
increasing bids is Knetsch, Tang, and Thaler (2001), who found that, in a ninth-price sealed-bid
auction, bids were decreasing. Their results may be explained, at least partly, by the lack of
competition in a market where almost all participants are buyers.
Bidding on all qualities simultaneously ensures that the rules and learning processes are
identical for all qualities. Even though our bids are rising over the trials, the differences in bids
between the different qualities seem to be robust estimators for the differences in WTP between these
qualities.
Risk aversion
We compare the expected value, calculated as described by equation (3), to the bids for quality V in
the six trials. In the first hypothetical trial, 12 participants bid more for V than the expected value of V.
In the following five trials, 5, 6, 9, 12 and 18 participants bid the same or more for V than the expected
value of V. Three participants (6%) bid the same or more for V than the expected value of V in all real
trials, 25 participants (49%) bid the same or more in some trials and less in other trials, and 23
participants (45%) bid less for V than the expected value of V in all four real trials. In the four real
trials, 13 participants (25%) bid on average more for V than the expected value of V. For the
remaining 38 participants (75%), the average bid for quality V was lower than the expected value of V
in the four real trials. These 38 participants showed evidence of risk aversion and were willing to pay
less for the uncategorized beef than its expected value. Furthermore, the results from the four real
trials imply that the categorization of the beef increased the total value of the beef by 8%.
As shown in table 3, the mean risk premiums were positive and the mean risk ratios were
below one in all trials. The mean risk premium in R1 and R2 was NOK 7.00 and NOK 6.44 while it
was reduced to NOK 4.96 and NOK 3.59 in R3 and R4. As discussed above, the tasting may explain
this reduction. The results of the non-parametric Wilcoxon signed rank test show that the risk premium
was statistically significant in all trials.
The cumulative distribution of the individual risk ratios in the four real trials is presented in
figure 3. Two participants (4%) had a risk ratio below 0.7, five participants (10%) had a risk ratio