Minimum detectable effect size for main outcomes (accounting for sample
design and clustering)
We base our optimal sample size calculations sizes using results from the related experimental study conducted by Löschel et al. (2021) using a Chinese sample. They report an extensive margin effect of -31% when turning from the local (i.e., Beijing, where 66% of the subjects contribute) to the global setting (i.e., Shenzen, 44% of the subjects contribute). To be able to detect a similar effect size, a power analysis with an underlying two-sample proportions (Pearson’s χ2) test (with α=0.05, p1=0.66, p2=0.44 indicates) that at least 150 experimental observations are needed to achieve a statistical power of 0.7. We were able to recruit 160 subjects for our experiment.
After the first wave of interventions, the following patterns emerge. Our preliminary results clearly indicate that the share of subjects that contribute to CO¬2 removal is larger than zero in both treatments (t-test, p=0.000). 65.0% of all subjects in our sample contribute a positive amount to the public good. While 70.0% of all subjects give a positive amount in S, this share decreases to 60.9% in CBS. This decrease is however not statistically significant at any conventional level (exact Fisher´s test, p=0.249).
In the following, we now describe subjects’ implicit willingness to pay (WTP) for CO2 removal. We denote the amount of money that a subject contributes to the reforestation project as its minimum WTP (WTPmin). Including all observations, the median WTPmin is 4.55 EUR/100kg CO2 removal and the mean WTPmin amounts to 6.33 EUR/100kg removal. Analyzing differences between treatments we find that both the median WTPmin (5 EUR in S vs. 2 EUR in CBS, MW-test, p=0.1449) and the mean WTPmin (7.18 EUR in S vs. 5.61 EUR in CBS, t-test, p=0.1663) does not differ significantly between S and CBS in our sample. In relative terms, average contributions amount attribute to 35.9% (in S) and 28.1% (in CBS) of the initial enumeration (20 EUR) and are therefore a little bit smaller than average contribution levels in conventional one-shot public goods experiments. Looking only at the contributions of positive contributors, we again find that both the median WTPmin (10 EUR in S vs. 10 EUR in CBS, MW-test, p= 0.3887) and the mean WTPmin (10.28 EUR in S vs. 9.21 EUR in CBS, t-test, p= 0.4200) does not differ significantly between S and CBS in our sample. Most notably, however, these first insights already indicate that a change in the treatments from a pure CO2 perspective to a scenario where local ancillary benefits from CO2 removal are explicitly stressed do not lead to a higher WTPmin. We can therefore not reject the Null-Hypothesis (H0: WTPCBS = WTPS) of H2.
Given our sample size calculations we are confident that this the Null result is not driven by an underpowered sample. We will run a further round of S and CBS interventions in 2021 to further increase the statistical power. In addition, we will control for potential income shocks caused by the COVID-19 pandemic which hit Germany after the first wave of interventions.