Mitigating Catastrophic Risk

2024-03-15

2024-12-19

Survival rate: percentage of games where climate catastrophe is avoided.

We will measure at the group level the fraction of games where the climate catastrophe was avoided, i.e. where the group reached the final round of the game.
This outcome will depend both on participant choices as well as externally set parameters (and these will interact), so we will also compare the actual survival rate in any given treatment with counterfactuals based on the same parameters but with choices as in the other treatments.

- Decisions to invest in mitigation
- Fraction of groups reaching “doom probability threshold”

Decisions to invest in mitigation: the decisions of participants to invest or not to invest in mitigation.
Fraction of groups reaching the “doom probability threshold”: This threshold is defined as p=(1+gamma)/gamma * Delta N. It is the probability threshold that if reached means that the probability of disaster will keep increasing even under maximal mitigation efforts.

In our game participants receive an endowment of 20 tokens and play for 15 periods. In each round there is a probability p that the game ends and the endowment is lost. In our game p increases round-by-round using the following formula
p^{t+1}=(1+\gamma) p^{t}.
Participants can invest into reducing the probability. Specifically participants can pay in each round 1 GBP to reduce the probability by \Delta.

Our treatments will vary
- the group size (either N=3 or N=12)
- the growth path of p.
There are three exogenous parameters affecting the growth-path: the initial probability (p^{0}), the round-on-round percent increase (\gamma) and the effectiveness of individual action (\Delta). Variation in those “visible” parameters then induces variation in “hidden” parameters such as the doom probability threshold or the number of stages needed to reach that threshold. We will vary these parameters across sessions.

Participants will play this 15-period game eight times with random re-matching in between. They will be paid for one randomly selected round.

Prior to the experiment, participants are asked to complete an online survey. This survey consists of four tasks: one to measures risk aversion, one to measure cooperative preferences, one to measure participants’ understanding of exponential growth, and one to measure participants understanding of compound lotteries. In addition, we obtain participants’ age, gender, nationality, ethnicity, and student status.

Not available

Treatments are assigned to sessions by researchers randomly in an office by a computer.

experimental sessions

Yes

36

6912 observations at the individual level (and correspondingly fewer at the group level in the N=3 and N=12 groups).

For each group size we plan to have 3456 observations or 18 clusters.

We will have 36 sessions with 24 participants each who will play for 8 rounds. This amounts to 6912 observations at the individual level. At the group level (main test) this will be 1152 observations in the small groups (N=3) and 288 observations for the big groups (N=12). We will test how our main (as well as secondary outcomes) depend on parameters related to the growth path (e.g. the percent increase in each round or the number of stages needed to reach the doom probability threshold), the effectiveness of individual action (\Delta) as well as group size. We will use OLS regressions to test for these differences. Standard errors will be clustered at the session level. Our sample sizes allow us to detect an increase or decrease of 5 (10) percentage points in the survival rate between the conditions with N=3 and N=12 with 80 percent power at the 5% level. (This is based on a t-test with hypothesized standard deviation of 0.25 (0.5) and a mean in the small groups of 0.5). There are no prior studies in this setting but this power analysis gives us confidence to detect economically meaningful differences in our main outcomes with our chosen sample sizes. We will focus both on mature behaviour after learning has occurred as well as behaviour in early stages. The latter is interesting in this context as in the context of the climate crisis learning is often not possible or very limited. When explaining individual donation decisions, some models will include demographics and the measures for risk preferences, cooperative preferences, understanding of exponential growth, and understanding of compound lotteries as controls. Furthermore, we will also explore which of these four measures best explain donation decisions, as all these four characteristics ay be relevant in this setting.