Learning how to choose (?) Effects of instructional choice sets in DCE

Transcription

1 Learning how to choose (?) Effects of instructional choice sets in DCE Jürgen Meyerhoff (TU Berlin) Klaus Glenk (SRUC) ICMC Sydney July 13

2 What do we mean by ICS? We define an Instruction Choice Set (ICS) as a choice set that is presented to respondents of a DCE before they start to answer the sequence of tasks that are used to estimate WTP values. Respondents answered the ICS but were informed that the choice does not count.

3 Why use an ICS? It may facilitate institutional learning It may generally reduce error variance Especially problems with the 1 st set in the sequence of choices may decrease Czajkowski et al Learning and fatigue Carlsson et al The first time is the hardest

4 No free lunch anchoring may occur ICS may induce bias related to anchoring or referencing similar to typical starting point bias Ladenburg/Olsen (2008) an ICS with varying price levels was shown prior to the actual choice task Differing prices impact preferences and WTP estimates, results are particularly gender specific Carlsson/Martinsson (2008) an additional first choice set was included with alternatives offering large improvements at relatively low costs There was no significant effect on overall model parameters (LR test)

5 Do ICS do more good than harm? Given the trade-off between advantages (institutional learning) and disadvantages (anchoring), should an ICS be used? First systematic attempt to investigate effects of an ICS with variation in both price and non-price attribute levels

6 SURVEY

7 Water quality improvements in Berlin & Brandenburg Based on EU Water Framework Directive Current quality in B&B poor to moderate Online survey with 751 completed interviews Respondents received 12 choice sets

8 RESEARCH DESIGN

9 Treatments Treatment N Choices T1: No Instruction Choice Set (ICS) T2: ICS: high quality & low prices T3: ICS: high quality & high prices T4: ICS: low quality & low prices T5: ICS: low quality & high prices

10 Example ICS Respondents were randomly assigned to the five treatments, i.e., one group was not presented an ICS

11 Treatment T2 High Quality Low Price T3 High Quality High Price T4 Low Quality Low Price T5 Low Quality High Price Expected effect on WTP and choice of SQ get a lot for little high quality is cheap wait for equally good situation Following trade-offs are relatively worse (lower quality at higher prices) Status quo choices: more than in T1 WTP compared to T1: on average lower high quality has its price get a lot for much less later - a lot of offers Some following trade-offs are relatively better Status quo choices: less than in T1 WTP compared to T1: on average higher close to status quo and cheap does not impress less anchoring Following trade-offs worse, equally good or better Status quo choices: similar to T1 WTP compared to T1: on average similar probably too expensive, maybe unacceptable anchor for high WTP Following trade-offs better Status quo choices: more than in T1 WTP compared to T1: on average higher

12 Hypotheses Differences in SQ choices? H0 H1 SQ ICS_T1 = SQ ICS_T2/T3/T4/T5 SQ ICS_T1 SQ ICS_T2/T3/T4/T5 Differences in WTP estimates? H0 H1 WTP ICS_T1 = WTP ICS_T2/T3/T4/T5 WTP ICS_T1 WTP ICS_T2/T3/T4/T5 Differences in error variance? H0 H1 λ ICS_T1 = λ ICS_T2/T3/T4/T5 λ ICS_T1 λ ICS_T2/T3/T4/T5

13 Econometric approach For each treatment separately RPL model Non cost attributes normally distributed, cost follows constrained triangular, error component for hypothetical alternatives Individual-specific WTP values Across all treatments Pooled model (same as above) Scale for groups of choice sets, first set in T1 fixed to 1

14 RESULTS

15 Time spent on market introduction and / or ICS

16 Results SQ choices T1 No ICS T2 High Quality Low Price T3 High Quality High Price T4 Low Quality Low Price T5 Low Quality High Price Total N (observations) 147 (1764) 151 (1812) 150 (1798) 155 (1860) 151 (1812) SQchoices % SQ always * 41 34* * * % * Significantly different compared to T1 at 5%-level

17 Individual-specific WTP (median) Treatment 1 Treatment 2 Treatment 3 Treatment 4 Treatment 5 No ICS high quality / low prices high quality / high prices low quality/ low prices low quality / high prices n = 147 n = 151 n = 149 n = 155 n = 151 Untere Havel Untere Havel Untere Havel Obere Havel Obere Havel Obere Havel City Spree City Spree KopeSpree Kopespree Dahme Dahme Dahme * Note: bold italics => 1% level; italics => 5% level

18 Individual-specific WTPs plot

19 Scale factor Scale differences by treatments T1 - no ICS T2 T3 T4 1 T5 0.5 S1 S2 S3-4 S5-7 S8-10 S11-12 Choice task

20 Conclusions Frequency of status quo choices is significantly different in three treatments ICS have a significant (and different) impact on stated WTP estimates anchoring present Early learning effects are not obvious first choice remains the hardest However, treatment without ICS reaches highest scale ICS inclusion may do more harm than good

21 The End The survey was part of the project Nitrolimit ( Funding for this project was provided by the Federal Ministry of Education and Research in Germany and is gratefully acknowledged (Fkz 033L041D).

22 Treatment T2 High Quality Low Price T3 High Quality High Price T4 Low Quality Low Price T5 Low Quality High Price Expected effects Status quo choices: more than in T1 WTP compared to T1: on average lower Scale: higher scale factor in S1/S2 compared to T1 Status quo choices: less than in T1 WTP compared to T1: on average similar or higher Scale: higher scale factor in S1/S2 compared to T1 Status quo choices: similar than in T1 WTP compared to T1: on average lower Scale: higher scale factor in S1/S2 compared to T1 Status quo choices: more than in T1 WTP compared to T1: on average higher Scale: higher scale factor in S1/S2 compared to T1 X X X X X X X

23 Individual-specific WTPs plot

24 Individual-specific WTPs plot

25 Individual-specific WTPs plot

26 Individual-specific WTPs plot

27 Scale differences T1 T2 T3 T4 T5 Value Std. err Value Std. err Value Std. err Value Std. err Value Std. err S1 1 fixed S S S S S