The Effect of Learning and Fatigue on Preferences and WTP in a ...

More documents

Recommendations

Info

the hypothetical market case nor in the actual market case, suggesting no learning or fatigue effect being present. Comparing a sequence <strong>of</strong> four CS with a sequence <strong>of</strong> eight CS, Hanley et al. (2002) find only very weak evidence that the number <strong>of</strong> CS affects preferences in any way. Further, they find no significant impact on WTP measures, again suggesting that no learning or fatigue effect is present. Similarly, Johnson & Bingham (2001) compare WTP estimates obtained from the beginning <strong>and</strong> last end <strong>of</strong> a sequence <strong>of</strong> SP questions <strong>and</strong> find no significant differences. In a repeated choice experiment Brouwer et al. (2009) examine learning over five CS with the first <strong>and</strong> the last CS being identical. <strong>The</strong>y also examine a learning effect through the size <strong>of</strong> the error variance, <strong>and</strong> find no differences in the variance throughout the five CS, <strong>and</strong> hence no learning or fatigue effect. In addition, they test the stability <strong>of</strong> preferences by comparing the first <strong>and</strong> the fifth CS which is identical, <strong>and</strong> find no differences in the error variance here as well, although 27 % <strong>of</strong> the respondents did chose another policy alternative in the fifth CS compared to the first CS. Phillips et al. (2002) use a similar approach, also adding an identical first CS at the end <strong>of</strong> the CE. <strong>The</strong>y only examine the rate <strong>of</strong> the same alternative being chosen in the last CS, <strong>and</strong> they find that 75 % <strong>of</strong> the respondents choose the same alternative in the last CS as they did in the first CS. Bateman et al. (2008b) also analyses the effect <strong>of</strong> adding in an identical first CS as the last choice, resulting in 17 CS in total for each respondent. <strong>The</strong>y find that respondents are less likely to choose an identical alternative when it is placed in the end <strong>of</strong> a choice sequence, <strong>and</strong> furthermore that the effect on the marginal utility <strong>of</strong> a given alternative is lower when placed in the end <strong>of</strong> the choice set sequence. Finally they examine a heteroskedastic multinomial logit model in which 16 orderspecific scale parameters were estimated. <strong>The</strong> results <strong>of</strong> this showed that only the scale factors for choice tasks 3, 9, <strong>and</strong> 16 were statistically significantly different from one at the usual significance level (0.05). On the basis <strong>of</strong> this they conclude that the most reliable estimates are found in the 6
eginning <strong>and</strong> in the end <strong>of</strong> the choice task sequence, so the approach <strong>of</strong> using order-specific scale parameters to support a learning hypothesis failed. 3. Method <strong>The</strong> underlying theory <strong>of</strong> CE is based on Lancaster’s Consumer <strong>The</strong>ory (LCT) (Lancaster 1966) <strong>and</strong> r<strong>and</strong>om utility theory (Gravelle & Rees 1992, Luce 1959, McFadden 1974). According to Lancaster, the (indirect) utility V ij that individual i achieves from good j is the sum <strong>of</strong> the utilities obtained from each <strong>of</strong> the K characteristics s kij . Assuming linearity in the valuation <strong>of</strong> attributes, the utility V ij can be written as follows: V = β1s1 + β2 s2 + .......... + β s ij i ij i ij Ki Kij (1) <strong>The</strong> parameter β ki represents the weight by which attribute k is valued by individual i 1 . In r<strong>and</strong>om utility theory, it is assumed that individuals make choices according to a deterministic part along with some degree <strong>of</strong> r<strong>and</strong>omness. Allowing U ij to represent the r<strong>and</strong>om utility that individual i places on alternative j, V ij now represents the deterministic component <strong>of</strong> the utility function <strong>and</strong> ε ij is a r<strong>and</strong>om variable that captures the unsystematic <strong>and</strong> unobserved r<strong>and</strong>om element <strong>of</strong> individual i’s choice (Hanley et al. 2002, Holmes & Adamowicz 2003). We will assume throughout the paper that the error terms are independent Gumbel distributions. Hence, the r<strong>and</strong>om utility U ij can be represented as follows: U ij = β .......... + is ij + β is ij + + β KisKij ε (2) 1 1 2 2 ij 1 Sometimes, the attribute weights differ between alternatives β kj but in the present study it was not meaningful to operate with different attribute weights for the hypothetical alternatives. 7
Page 1 and 2: The Effect
Page 3 and 4: Waldman 2008). However, results are
Page 5: not by changing preferences as have
Page 9 and 10: unobserved variability, and
Page 11 and 12: expectation of the
Page 13 and 14: Note: ASC is the alternative specif
Page 15 and 16: Scarpa et al. 2008, Swait & Louvier
Page 17 and 18: testing for impacts on WTP can be n
Page 19 and 20: preference structure. Turning to a
Page 21 and 22: Dellaert, B. G. C.; Brazell, J. D.

The Effect of Learning and Fatigue on Preferences and WTP in a ...

You also want an ePaper? Increase the reach of your titles

Delete template?

Save as template?