|
Field
Last Published
|
Before
October 20, 2022 11:25 AM
|
After
October 26, 2022 12:09 PM
|
|
Field
Planned Number of Observations
|
Before
6 treatments, 175 individuals per treatment, 20 observations from each individual = 21,000 observations
|
After
6 treatments, 175-185 individuals per treatment, 20 observations from each individual = >21,000 observations
|
|
Field
Sample size (or number of clusters) by treatment arms
|
Before
175 individuals per treatment
|
After
185 individuals per treatment in computerised environments
|
|
Field
Power calculation: Minimum Detectable Effect Size for Main Outcomes
|
Before
The study is powered for equivalence tests of bidder surplus, auctioneer's revenue, and welfare (all logged values), between the Product-Mix and the sequential auction format in the computerised bidding environment, respectively. The hypothesied effect is therefore zero. The maximum pooled standard deviation for the outcomes in questions is 0.956 (logged utility). I choose a significance level of 0.05 and equivalence bounds of 20%. Power calulations (sample sizes) are presented in the pre-analysis plan for a power of 0.8 and 0.9 and for various scenarios of within-subject correlation. Choosing a sample size of 175 subjects per treatment conservatively allows for a serial correlation between rounds of up to 0.4. This is assumed to be a strict upper bound based on related experimental designs in the literature. The treatments with only human bidders are for exploratory analysis only, hence the sample size will be lower.
|
After
The study is powered for pairwise equivalence tests between all considered auction formats in the computerised bidding environment, concerning the outcomes of bidder surplus, auctioneer's revenue, and welfare (all absolute values), respectively. The hypothesied effect is therefore zero. I choose a significance level of 0.05 and equivalence bounds of +/- 20% of the pooled standard deviation corresponding to each pairwise comparison. Power calulations (sample sizes) are presented in the pre-analysis plan for a power of 0.8 and 0.9 and for various scenarios of within-subject correlation. Choosing a sample size of 185 subjects per treatment conservatively allows for a serial correlation between rounds of up to 0.4. This is assumed to be a strict upper bound based on related experimental designs in the literature. The treatments with only human bidders are for exploratory analysis only, hence the sample size will be lower.
|