Minimum detectable effect size for main outcomes (accounting for sample
design and clustering)
Main test: We are interested in testing whether different thresholds trigger different effort provision and risk taking. To do so, for the main test we define two kinds of threshold for each participant, based on the participant’s baseline effort provision: low thresholds and high thresholds. We will restrict the analysis to people who in Round 1 provided effort between 40 and 80. For this subsample, we will define:
- Low threshold: The thresholds of 20 and 40 tasks.
- High threshold: The thresholds of 80 and 100 tasks.
Test 1: We regress \Delta Effort, defined as the difference between effort choices in Rounds 9-20 and corresponding choices in Rounds 6-8, on dummy variables indicating whether the threshold is low or high. Controls will be gender, age, social class, ethnicity, and education. Standard errors will be clustered at the participant level.
Test 2: We regress \Delta Risk, defined as the difference between risk choices in Rounds 9-20 and corresponding choices in Rounds 6-8, on dummy variables indicating whether the threshold is low or high. Controls will be gender, age, social class, ethnicity, and education. Standard errors will be clustered at the participant level.
We performed a power analysis using a pilot with 34 participants. To do so, we performed simulations in which we bootstrapped the sample and assigned treatments (thresholds) randomly (code available upon request). We conclude that with a sample of 500 participants, we expect around 200 participants with Round 1 choices between 40 and 80 and in this case we have 80% power to detect at the 5% level an effect of high thresholds on effort of 2.5 tasks and on risk of 7 percentage points.
Secondary tests: To more closely test the model, we will define four types of thresholds:
- the low threshold to be the highest threshold weakly below the participant’s effort choice in the corresponding Round 6-8,
- the high threshold to be the lowest threshold strictly above the participant’s effort choice in the corresponding Round 6-8,
- the very high threshold to be any threshold at least 21 units but less than 40 units above the participant’s effort choice in the corresponding Round 6-8 and
- the extremely high threshold to be any threshold at least 41 units above the participant’s effort choice in the corresponding Round 6-8.
Note that any of these might be missing for some participants. Note also that we do not define very low and extremely low thresholds as we do not expect behavioural differences between those types of thresholds.
We will run the same regressions as defined above using these four thresholds and study to what extent they are in line with the model.
Tertiary tests:
- We are interested in exploring heterogeneity by other background variables, and in particular gender.
- We are interested in comparing effort choices in R2-5 with R1 depending on the difficulty of the threshold.