Table 1 reports swap rates by treatment. The Endowments column shows the number of subjects initially endowed with each of the two bundles. The Swaps column reports the total number of swaps and (in parentheses) the number of swaps in each possible direction: swapping crisps for lemonade (c → l) or lemonade for crisps (c ← l). The Swap rate is the proportion of subjects who swapped. The final column reports p-values for Boschloo tests of the null hypothesis (based on standard preference theory) that the final allocation is independent of endowment (i.e. there is 50% swapping rate), against the alternative hypothesis that there is an endowment effect (i.e. the swap rate is less than 50%).Footnote 5
We comment first on the results for the BASELINE and TASTING treatments. While the swap rates for these treatments have the expected pattern with BASELINE < TASTING < 0.5, neither has a statistically significant endowment effect. The absence of an endowment effect in the BASELINE treatment is noteworthy, and we examine this further in Sect. 5. Its absence, however, means that we cannot conduct a meaningful test of the taste experience hypothesis (which would require us to look for a reduction of the endowment effect in TASTING relative to BASELINE).
We now test the choice experience hypothesis by comparing behaviour between the PASSIVE and CHOOSING treatments. There is a significant endowment effect in the PASSIVE treatment (the trading rate is only 0.23) and the experience of choosing weakens it. In line with the choice experience hypothesis, the trading rate rises to 0.35 for the CHOOSING treatment where subjects are approximately 50% more likely to trade. This treatment difference just fails to reach significance at the 10% level (p = 0.1023, Boschloo test with one-sided alternative hypothesis) though we do also find a (weakly) significant effect of choice experience in the individual-level analysis below (see analysis of Table 2).
An unanticipated feature of our results is the difference between the treatments in which the acquisition of endowments occurred in two steps (PASSIVE, CHOOSING), rather than one step (BASELINE, TASTING). While 44% of subjects swapped in the one-step treatments, only 29% of subjects did so in two-step treatments (p = 0.0190, Boschloo test with two-sided alternative hypothesis). Comparing the BASELINE and PASSIVE treatments, which control for the experiences of choosing between and tasting the goods, respectively 42% and 23% of subjects swapped their endowment (p = 0.0461, Boschloo test, two-sided alternative hypothesis). These tests provide evidence that acquiring an endowment in stages strengthens the endowment effect. We think this is an intriguing discovery and briefly discuss its interpretation and potential significance in Sect. 6.
We supplement the analysis of treatment effects by using logit regression (following List 2003) to model the probability that a subject swaps, taking account of individual characteristics. Observations from all treatments are pooled. This provides a clear overall view of treatment effects within the models we estimate (specifically models 3 and 5) and increases the statistical power of the tests. Across different specifications, as independent variables, we included a dummy for the treatment, the individual experiences, plus a set of individual-level characteristics elicited in the post-decision questionnaire, including age and gender. We also included a measure of individual-level loss aversion constructed by ranking subjects’ from least to most loss averse based on their responses to a series of hypothetical tasks (see supplementary materials). The results are reported in Table 2.
Model 1, which includes only a constant, provides a simple econometric test for the presence of an endowment effect. The highly significant negative coefficient confirms the presence of an endowment effect in our data.
In all three models that include individual-level characteristics (models 2, 4 and 5), the coefficient for measured loss aversion is negative (other characteristics are never significant). Tests of the null hypothesis that the swap rate is independent of loss aversion are rejected at the 5% level (model 2, p = 0.0436; model 4, p = 0.0424; model 5, p = 0.0369). Hence, in these data, more loss averse individuals were less likely to trade. While this result supports theories, including LOS, which invoke loss aversion to explain the endowment effect, we note that we do not replicate this association in the follow-up study reported in Sect. 5.
Models 3 and 5 provide evidence that the experience of choosing part of the endowment increases the trading rate (and reduces the endowment effect). Tests of the null hypothesis that the trading rates in the PASSIVE and CHOOSING treatments are equal are rejected at the 10% level in favour of the alternative hypothesis that the trading rate is higher in the CHOOSING treatment (model 3, p = 0.0898; model 5, p = 0.0894).
Finally, this analysis confirms that acquiring endowments in two steps decreases the trading rate (increases the endowment effect). This is evidenced by the significant negative coefficients on PASSIVE in models 3 and 5 and by the significant coefficient for ‘Two-step’ in model 4.