The demand for voluntary carbon offsets:

Size: px
Start display at page:

Download "The demand for voluntary carbon offsets:"

Transcription

1 The demand for voluntary carbon offsets: Field experimental evidence from the long-distance bus market in Germany Martin Kesternich (Centre for European Economic Research, ZEW) Andreas Löschel Daniel Römer 37 th IAEE International Conference New York City, USA June 18, 2014

2 Frankfurt Munich 72 USD Munich Frankfurt

3 Frankfurt Munich 72 USD Munich Frankfurt Your bus trip causes 41.4 kg of CO 2 emissions. These emissions can be offset at a cost of 1 USD. Do you want to offset your emissions at a cost of 1 USD? Additional information Yes No

4 Share of compensated bookings in %

5 Frankfurt Munich 72 USD Munich Frankfurt Your bus trip causes 41.4 kg of CO 2 emissions. These emissions can be offset at a cost of 1 USD. Due to a campaign, we support your contribution to climate protection during this trip and we will cover half of the related offsetting cost. For each USD of your contribution we will reimburse 0.50 USD. As a result, you will pay 0.50 USD instead of 1 USD. Do you want to take part in the campaign and offset your emissions at a cost of 0.50 USD? Additional information Yes No

6 Share of compensated bookings in %

7 Frankfurt Munich 72 USD Munich Frankfurt Your bus trip causes 41.4 kg of CO 2 emissions. These emissions can be offset at a cost of 1 USD. Due to a campaign, we support your contribution to climate protection during this trip and we will increase the amount of emissions being offset by 100%. As a result, you will offset 82.8 kg of CO 2. Do you want to take part in the campaign and offset your emissions at a cost of 1 USD? Additional information Yes No

8 Share of compensated bookings in %

9 This presentation contribute to the understanding of preferences for individual (indirect) CO 2 emission reductions focus on (long-term) impacts of financial stimuli on decision behavior

10 FIELD EXPERIMENTAL DESIGN

11 introduced as an additional booking step at the company s website no advertisement, neutral wording, no additional passenger data randomization strategy treatment assignment based on a randomly generated number being saved as a browser cookie during the first booking returning customers reassigned to same treatment offsetting activity offsetting costs linear in distance: 11ct/100pkm (47g CO 2 /pkm incl. LCA, USD/tCO 2 ) project: improved cookers for Ghana (VER, WWF Gold Standard)

12 fall 2013 n = 11,258 online booking decisions, ~ 1,600 observations per treatment 53.7% of all customers are female average age: 26.2 years (median: 23)* high share of university students (57.9%)* rather low income (56% earn less than 1000 Euro per month)* for 33.8% of all customers we observe multiple bookings * Survey data collected after the experiment among customers (bus rides/online), survey participants did not necessarily take part in the experiment

13 price rebates matching grants USD/t CO 2 p -25% 1/3: USD/t CO USD/t CO 2 p -50% 1: USD/t CO USD/t CO 2 p -75% 3: USD/t CO 2 baseline USD/t CO kg CO 2 for 1 USD

14 price rebates matching grants USD/t CO 2 p -25% 1/3: USD/t CO USD/t CO 2 p -50% 1: USD/t CO USD/t CO 2 p -75% 3: USD/t CO 2 baseline USD/t CO kg CO 2 for 1 USD p-50%: 41.4 kg CO 2 for 0.50 USD

15 price rebates matching grants USD/t CO 2 p -25% 1/3: USD/t CO USD/t CO 2 p -50% 1: USD/t CO USD/t CO 2 p -75% 3: USD/t CO 2 baseline USD/t CO kg CO 2 for 1 USD p-50%: 41.4 kg CO 2 for 0.50 USD 1:1: 82.8 kg CO 2 for 1 USD

16 RESULTS

17 Descriptive analysis (bookings) 1.5 trips per booking distance: 273 km CO 2 emissions: 12.8 kg hypothetical offsetting costs: 0.42 USD (including rebates) share of compensated bookings: 29.9% additional information on offsetting program: 1.4%

18 Treatment effects (only first booking): participation rates (Statistical evidence: Mann-Whitney U tests, series of logit regression models) Share of compensated bookings in % Observation 1 Price reductions of 25% or more and a 1:1 matching scheme increase the share of compensated bookings in contrast to the control group.

19 Treatment effects (only first booking): participation rates (Statistical evidence: Mann-Whitney U tests, series of logit regression models) Share of compensated bookings in % Observation 2 Given equal carbon prices (in /t CO 2 ) price rebates lead to higher participation rates than matching grants.

20 Treatment effects (only first booking): participation rates (Statistical evidence: Mann-Whitney U tests, series of logit regression models) Share of compensated bookings in % Observation 3 An equal split between passenger and bus company (1:1 or p-50%) leads to higher participation rates than a weaker intervention (1/3:1 or p-25%) and is statistically equivalent to a stronger intervention (3:1 or p-75%).

21 Treatment effects (only first booking): net contributions (Statistical evidence: Mann-Whitney U tests, series of logit regression models) Average net contributions in USD Observation 4 1:1 matching is the only incentive scheme effectively increasing customers net contributions in contrast to the baseline scenario. Price rebates decrease average donations.

22 Beyond treatment effects: further explanatory variables dependent variable: participation y/n Explanatory variables dy/dx [95% Conf. Intervall] p-25% p-50% 0.071*** p-75% 0.085*** /3: : *** : ticketprice distance trips X booking group ** CO 2 (inkl. match) voucher *** info 0.394*** female ** booking# *** Average marginal and discrete probability effects are calculated after ML estimation, all bookings, *** p<0.01, ** p<0.05 Observation 5 Factors associated with a smaller likelihood to offset carbon emissions from the underlying bus trip are group purchases, female sex (for matching schemes), and repeated bookings.

23 Conclusion extension to green tickets matching does not affect participation as much as rebates do weak price elasticity of demand first insights on the dynamics of the stimuli decreasing effects for returning customers over time except for 1:1 matching efficiency analysis average donations per passenger only increase in 1:1 matching all other treatments do not pay off

24 Appendix

25 Long-term effects (all bookings): participation rates Explanatory variables Participation (all bookings) Participation (without stage 1 bookers) p-25% 0.512** (0.200) 0.394* (0.213) p-50% 0.907*** (0.199) 0.887*** (0.209) p-75% 0.609*** (0.184) 0.759*** (0.212) 1/3: (0.194) (0.204) 1: (0.184) (0.209) 3: ** (0.219) 0.433* (0.231) control X booking # (0.100) (0.112) p-25% X booking # *** (0.118) (0.124) p-50 X booking # *** (0.118) *** (0.121) p-75% X booking # (0.093) ** (0.124) 1/3:1 X booking # (0.105) (0.107) 1:1 X booking # (0.088) 0.209* (0.112) 3:1 X booking # * (0.117) * (0.122) further controls yes yes constant *** *** (0.146) (0.157) log likelihood Observations 11,229 10,642 (Preliminary) Observation 6 Maximum likelihood estimates in binary logit models, all bookings, *** p<0.01, ** p<0.05 Long-term effects for price rebates schemes suggest a decreasing impact over time for returning customers. While large match ratios (i.e. 3:1) are not suitable to maintain the share of compensated bookings in the long term an equal sharing rule (1:1) may be beneficial in counteracting downward trends for repeated bookings.

26 Long-term effects: participation rates after intervention (Preliminary) Observation 7 The two minimal interventions (1/3:1 and p-25%) lead to slightly lower or equal participation in the long-run, while the remaining four treatments are followed by higher participation than the control group. We get strongest results for the 1:1 match, yielding the highest point estimate and marginal significance (p<0.10).

27 Table 1: Summary statistics of the experimental design All bookings Treatment control 1:1 p-50% 3:1 p-75% 1/3:1 p-25% all Bookings per ID Trips per booking Distance per booking (km) Emissions per booking (kg CO 2 ) Ticketprice (excl. offsetting cost) Offsetting cost (% of tot booking) Voucher (% ) Info button request (%) Group ticket (%) Female Carbon price ( /t CO 2 ) Observations 1,652 1,571 1,590 1,634 1,592 1,637 1,582 11,258

28 Table 2: Summary statistics of the experimental results Only first booking Treatment control 1:1 p-50% 3:1 p-75% 1/3:1 p-25% all Compensated bookings (%) CO 2 compensation: - without match (kg) - including match (kg) Offsetting payments - total payment (cents) - paid by customer (cents) Observations 1,323 1,245 1,298 1,323 1,273 1,312 1,278 9,052 All bookings Treatment control 1:1 p-50% 3:1 p-75% 1/3:1 p-25% all Compensated bookings (%) CO 2 compensation (kg): - without match - including match Offsetting payments - total payment (cents) - paid by customer (cents) Observations 1,652 1,571 1,590 1,634 1,592 1,637 1,582 11,

29 Table 3: Mann-Whitney U (MW-U) test on differences between treatments: participation rates Only first booking Treatment control 1:1 p-50% 3:1 p-75% 1/3:1 p-25% control (0.0345) (0.0000) (0.2105) (0.0000) (0.5952) (0.0466) 1: (0.0088) (0.3782) (0.0044) (0.0085) (0.8915) p-50% (0.0004) (0.8049) (0.0000) (0.0055) 3: (0.0002) (0.0750) (0.4541) p-75% (0.0000) (0.0026) 1/3: (0.0120) p-25% Note: According to a MW-U test, the null hypothesis states that the median of two independent observations is equal. The first booking per ID serves as an independent observation. We compare rows with columns and report both z statistics and p values (in parentheses).

30 Table 5: Mann-Whitney U (MW-U) tests on treatment differences (contribution levels) Only first booking Treatment control 1:1 p-50% 3:1 p-75% 1/3:1 p-25% control (0.0436) (0.1788) (0.1916) (0.9965) (0.6159) (0.6213) 1: (0.2655) (0.4783) (0.0087) (0.0122) (0.0841) p-50% (0.8246) (0.0001) (0.054) (0.6503) 3: (0.1116) (0.0753) (0.3531) p-75% (0.5068) (0.0594) 1/3:1 Note: According to a MW-U test, the null hypothesis states that the median of two independent observations is equal. The first booking per ID serves as an independent observation. We compare rows with columns and report both z statistics and p values (in parentheses) (0.3076)

31 Table 10: Maximum likelihood estimates in binary logit models, treatment effects and determinants on offsetting behavior, only first booking, dependent variable: compensated Explanatory variables (1) (2) (3) (4) (5) (6) p-25% 0.199** 0.201** 0.199** 0.196** 0.199** 0.198** (0.0881) (0.0882) (0.0881) (0.0881) (0.0881) (0.0882) p-50% 0.430*** 0.431*** 0.429*** 0.428*** 0.430*** 0.430*** (0.0861) (0.0861) (0.0860) (0.0860) (0.0860) (0.0861) p-75% 0.461*** 0.462*** 0.460*** 0.459*** 0.460*** 0.461*** (0.0865) (0.0865) (0.0865) (0.0864) (0.0865) (0.0865) 1/3: (0.0894) (0.0894) (0.0894) (0.0894) (0.0896) (0.0900) 1: ** 0.225** 0.224** 0.221** 0.281*** 0.256*** (0.0882) (0.0881) (0.0881) (0.0881) (0.0899) (0.0935) 3: *** 0.229* (0.0878) (0.0878) (0.0877) (0.0877) (0.101) (0.128) ticketprice *** ( ) ( ) distance *** ( ) ( ) trips X booking ** (0.0300) (0.0663) group *** ** (0.0822) (0.106) CO 2 (inkl. match) *** ( ) ( ) voucher (0.175) (0.175) (0.175) (0.175) (0.175) (0.177) info 1.890*** 1.890*** 1.878*** 1.872*** 1.897*** 1.898*** (0.186) (0.185) (0.184) (0.183) (0.186) (0.186) female ** * * ** * ** (0.0462) (0.0462) (0.0462) (0.0462) (0.0462) (0.0464) Constant *** *** *** *** *** *** (0.0769) (0.0747) (0.0793) (0.0677) (0.0693) (0.0864) log likelihood Observations 9,024 9,024 9,024 9,024 9,024 9,024 Omitted treatment category is control, robust standard errors in parentheses, *** p<0.01, ** p<0.05, * p<0.1

32 Table 8: Postestimation Wald tests on selected explanatory variables after binary logit regression on participation decision Column (6) Coefficient p-50% p-75% 1/3:1 1:1 3:1 p-25% <*** <*** >** < < p-50% < >*** >* > p-75% >*** >** >* 1/3:1 <* <** 1:1 < Note: We compare rows with columns. *** p<0.01, ** p<0.05, * p<0.1, We test whether the difference of two estimated coefficients is significantly different from zero, e.q. we test whether b[p-25%] = b[p-50%].

33 Table 7: Maximum likelihood estimates in binary logit models, treatment effects and determinants on offsetting behavior, only first booking, dependent variable: compensation Explanatory variables (1) (2) (3) (4) (5) (6) p-25% (0.0128) (0.0126) (0.0128) (0.0130) (0.0128) (0.0126) p-50% (0.0118) (0.0117) (0.0118) (0.0121) (0.0118) (0.0117) p-75% * ** * * ** (0.0114) (0.0113) (0.0115) (0.0117) (0.0114) (0.0113) 1/3: (0.0134) (0.0133) (0.0134) (0.0137) (0.0135) (0.0134) 1: ** ** ** ** ** (0.0134) (0.0133) (0.0136) (0.0139) (0.0136) (0.0141) 3: *** (0.0132) (0.0130) (0.0132) (0.0134) (0.0196) (0.0219) ticketprice *** ( ) ( ) Distance *** *** (3.30e-05) (6.11e-05) trips X booking *** ( ) (0.0153) group *** ** (0.0147) (0.0170) CO 2 (inkl. match) *** -3.98e-05 ( ) ( ) voucher (0.0262) (0.0255) (0.0260) (0.0268) (0.0260) (0.0263) info 0.245*** 0.241*** 0.247*** 0.257*** 0.243*** 0.241*** (0.0227) (0.0231) (0.0257) (0.0282) (0.0231) (0.0222) female * * * * ** ( ) ( ) ( ) ( ) ( ) ( ) Constant *** *** *** *** *** *** (0.0177) (0.0154) (0.0162) (0.0121) (0.0128) (0.0160) log likelihood Observations 9,024 9,024 9,024 9,024 9,024 9,024 Omitted treatment category is control, robust standard errors in parentheses, *** p<0.01, ** p<0.05, * p<0.1

34 Table 11: Maximum likelihood estimates in binary logit models, treatment effects and determinants on offsetting behavior, all bookings, dependent variable: compensated Explanatory variables (1) (2) (3) (4) (5) (6) p-25% (0.0886) (0.0887) (0.0887) (0.0887) (0.0886) (0.0887) p-50% 0.346*** 0.348*** 0.345*** 0.345*** 0.346*** 0.347*** (0.0877) (0.0877) (0.0877) (0.0877) (0.0877) (0.0877) p-75% 0.412*** 0.413*** 0.411*** 0.412*** 0.411*** 0.413*** (0.0884) (0.0884) (0.0884) (0.0884) (0.0884) (0.0884) 1/3: (0.0913) (0.0913) (0.0913) (0.0912) (0.0913) (0.0918) 1: ** 0.219** 0.217** 0.216** 0.276*** 0.245*** (0.0901) (0.0901) (0.0901) (0.0901) (0.0917) (0.0951) 3: ** (0.0890) (0.0890) (0.0890) (0.0889) (0.102) (0.128) ticketprice *** ( ) ( ) Distance *** ( ) ( ) trips X booking *** (0.0315) (0.0682) Group *** ** (0.0800) (0.102) CO 2 (inkl. match) *** ( ) ( ) Voucher *** *** *** *** *** *** (0.168) (0.167) (0.167) (0.167) (0.168) (0.169) Info 1.924*** 1.926*** 1.914*** 1.904*** 1.931*** 1.930*** (0.182) (0.182) (0.181) (0.180) (0.183) (0.182) Female ** ** ** ** ** ** (0.0463) (0.0463) (0.0463) (0.0463) (0.0463) (0.0464) booking # *** *** *** *** *** *** (0.0475) (0.0476) (0.0475) (0.0474) (0.0475) (0.0474) Constant *** *** *** *** *** *** (0.0909) (0.0891) (0.0940) (0.0831) (0.0845) (0.101) log likelihood Observations 11,229 11,229 11,229 11,229 11,229 11,229 Note: Omitted treatment category is control, robust standard errors in parentheses, *** p<0.01, ** p<0.05, * p<0.1

35 Table 13: Maximum likelihood estimates in tobit models, treatment effects and determinants of offsetting behavior, all bookings, dependent variable: compensation Explanatory variables (1) (2) (3) (4) (5) (6) p-25% (0.0127) (0.0126) (0.0127) (0.0129) (0.0127) (0.0126) p-50% (0.0120) (0.0119) (0.0120) (0.0122) (0.0120) (0.0119) p-75% ** ** ** ** ** ** (0.0116) (0.0115) (0.0116) (0.0118) (0.0116) (0.0115) 1/3: (0.0136) (0.0135) (0.0135) (0.0139) (0.0137) (0.0135) 1: ** ** ** ** ** (0.0134) (0.0133) (0.0135) (0.0138) (0.0139) (0.0143) 3: *** (0.0132) (0.0130) (0.0132) (0.0134) (0.0192) (0.0219) ticketprice *** -8.95e-05 ( ) ( ) distance *** *** (3.15e-05) (6.16e-05) trips X booking *** ( ) (0.0162) group *** *** (0.0144) (0.0166) CO 2 (inkl. match) *** ( ) ( ) Voucher ** ** * ** ** (0.0240) (0.0236) (0.0239) (0.0246) (0.0238) (0.0242) Info 0.252*** 0.248*** 0.254*** 0.266*** 0.251*** 0.248*** (0.0220) (0.0224) (0.0247) (0.0269) (0.0226) (0.0217) Female ** ** ** ** ** *** ( ) ( ) ( ) ( ) ( ) ( ) booking # *** *** *** *** *** *** ( ) ( ) ( ) ( ) ( ) ( ) Constant *** *** *** *** *** *** (0.0181) (0.0162) (0.0173) (0.0133) (0.0139) (0.0185) log likelihood Observations 11,229 11,229 11,229 11,229 11,229 11,229 Note: Omitted treatment category is control, robust standard errors in parentheses, *** p<0.01, ** p<0.05, * p<0.1