PS 1-45
END-AVERSION IN TIME TRADE-OFF VALUATION?
Value algorithms for the EQ-5D based on time trade-off (TTO) valuation generally display a disproportionately large utility drop between having no problems (value of 1) and having any problem, however small (value of around 0.9). This may result in overspending on minor health issues. We hypothesize that the observed drop may be caused in part by a general tendency to contrast the presented health states, i.e. a form of end-aversion. Such contrasting could be specific to comparisons involving full health, or apply to all health states. The aim of the study was to test if such contrasting could be at work.
Method:
We performed an experimental TTO study with a fixed set of EQ-5D-5L health states (12111, 11122, 12122, 13224, and 24445). Each health state was traded off against full health and all dominating states. 120 general population respondents were interviewed face to face, with the 14 TTO health state pairs presented in random order. Let A refer to state 11111 (full health), B an intermediate imperfect states, and C a state worse than B. We have trade-offs AB and AC expressed in units of A, and BC in units of B. Without contrasting/end aversion, AC should be equal to AB x BC. Predictions for AC using AB and BC were compared to observed AC using mean square error (MSE). We tested three alternative models:
1: Base case according to utility theory:
AC = AB * BC + ε
2: Fixed end-aversion specific to comparisons to full health
AC + Φ = (AB + Φ) * BC + ε
AC = (AB + Φ) * BC - Φ + ε
3. End-aversion not specific to full health
AC + Φ = (AB + Φ) * (BC + Φ) + ε
AC = (AB + Φ) * (BC + Φ) - Φ + ε
Result:
model |
Φ |
p |
MSE |
|||
1 |
- |
0.216 |
||||
2 |
0.142 |
<0.001 |
0.189 |
|||
3 |
0.178 |
<0.001 |
0.176 |
Conclusion:
Predictions were improved by adding terms representing tendencies reflecting contrasting/end aversion. The improved fit of model 3 over model 2 suggests that there may be a general upper-end-aversion in TTO not specific to comparison to full health. A possible explanation for the observation could be that values close to 1 inevitably have an error distribution biased downwards. Preliminary findings suggest that this phenomenon warrants further investigation.