ABSTRACT
Introduction Since 1978, economists, psychologists, sociologists, and accountants have used experiments to investigate the determinants of tax compliance. Numerous experimental treatments have been tested, including economic incentive effects (such as changing the probability of audit, the severity of the fine for non-compliance, or the tax rate) and more psychological effects (for example, the use of “neutral” terminology instead of tax terminology, or appeals to morality). This chapter adds to the existing survey literature on tax compliance (see Andreoni et al., 1998, and Slemrod, 2007) by using a meta-analysis to test some of the traditional incentive-effect hypotheses regarding tax compliance. A compelling argument for a need to combine data from multiple studies is made by Goldfarb (1995). He argues that economists need a “methodology of Plausible Inference (MPI)”, and suggests that economists should use metaanalysis as that MPI. The advantages of meta-analysis are evident in the increasing number of such studies appearing in the economics literature.2 Meta-analysis allows for a large increase in power – an important consideration in experimental research. This increase in power comes from two sources: increased number of data points, and an increased variability in the dependent variables. Because data are relatively expensive to collect, experimentalists collect as little as possible, which means the power of any particular study will be small. By combining studies, meta-analysis increases the power of statistical tests to detect significant results. If reviewers use significance as a criterion, then many lowpower studies showing little or no significance will cause the reviewer to conclude no effect of the particular treatment. However, a properly done meta-analysis will be able to combine the results of these small samples into one large sample, and take advantage of the increase in power to test for the hypothesized effect. An additional advantage of a meta-analysis on experiments is that multiple treatment effects can be examined. For example, a typical experimental paper might examine the impact of having a fine rate of two or four, while a second paper might look at fine rates of three and five. Combining the data from these two studies increases the variability in the independent variables, and allows for more powerful statistical tests.
