Download files
Access & Terms of Use
open access
Embargoed until 2016-09-30
Copyright: Zhang, Le
Embargoed until 2016-09-30
Copyright: Zhang, Le
Altmetric
Abstract
This thesis investigates the production and evaluation of evidence in experimental economics, and it consists of four essays. The first two essays are concerned with the question of how giving decisions are affected by different experimental design choices. We first present our replication of Engel's (2011) meta-study of dictator game experiments. Using his data and statistical models, we determine his results to be robust with one important exception. In Engel's report, the effects of the take-option on giving are statistically insignificant. By contrast, we find the negative effects to be economically and statistically significant, which is consistent with the relevant literature. We also find that the effects of the take-option disappear when the reference point is changed away from the original endowment in a data coding that we call normalisation.
In the second essay, we reflect on the meta-analysis method used in the first essay. We investigate the effects of various experimental design and implementation choices using our own data, including a comparison between Engel's dataset and our own dataset.
The aim of the third essay is to emphasise the importance of sample size design and appropriate statistical inferences. We demonstrate that it is a problematic procedure of statistical inference only to consider significance levels (type-I error) without taking into account type-II error. Making use of the marginal effects estimated in the meta-analysis from the first essay, we determine the statistical power of these studies to be notably low. We also discuss the consequences of low statistical power with specific examples.
In the fourth essay, we analyse participants' pro-social and antisocial preferences using both dictator and joy-of-destruction games. We examine the extent to which behaviours are context-dependent and consistent with between-subjects and within-subject experimental designs. We determine that for one-shot dictator and joy-of-destruction games, participants' social preferences depend on the choice set. We also observe certain inconsistent behaviours (exhibiting both altruism and nastiness) and find significant differences in the proportion of inconsistent behaviours across treatments. While most participants tend to be selfish, they choose to increase social welfare when given the opportunity.