Academic journal article
By Brooks, Arthur C.; Lewis, Gregory B.
Journal of Public Administration Research and Theory , Vol. 12, No. 1
Policy analysts generally model policy outcomes as dependent on both policy tools and controls (variables that the policy maker can and cannot manipulate, respectively), then they test those hypotheses using some form of regression. Typically, they glean up to three pieces of information from the regression output: whether each variable has a statistically significant impact on the outcome, the direction of the impact, and the size of the impact. The emphasis on statistical significance, however, frequently crowds out seri-ous consideration of the size of the effect.
Exploratory analysis can squeeze additional information from regression results, providing a clearer sense of a variable's policy significance and allowing decision makers to find combinations of policy interventions that achieve particular objectives. In essence, exploratory analysis runs model-based studies in reverse: While traditional analysis uses a model to understand the effects of inputs on a particular output, exploratory analysis takes a user-defined output value and finds the range and combinations of inputs that achieve it.
This article provides a primer on exploratory analysis, employing a binary choice model on private charitable giving in Russia. (1) In this example, Russian policy makers ask how tax rates and transfer payments can be combined to achieve different levels of private charitable giving. The results show the policy significance of these tools, and the options that are open to policy makers. We should note that while it employs real data, the example is highly stylized, leaving out many possible regressors and using an especially simple functional form for the regression model. Our aim is not to perform a comprehensive analysis of the Russian philanthropy, but to clarify exploratory analysis.
AN EXAMPLE PROBLEM WITH A TYPICAL SOLUTION--AND TYPICAL UNRESOLVED QUESTIONS
As Russia moves past the Soviet system of government, the public sector can no longer be the sole provider of public goods and services (Davis 1996), and even its Constitution sets a goal of increasing charitable giving. (2) It has several tools to encourage private philanthropy directly and indirectly--including government subsidies to and nontaxation of charitable organizations, tax deductibility of contributions, income tax rates, and transfer payments to individuals. We focus on the last two tools. We estimate a logit model for the probability of giving charitably in Russia, using data from the World Bank's Household Expenditure and Income Data for Transitional Economies (HEIDE). (3) This database contains a sample of 5,915 Russian households in 1994 (Ackland et al. 1997). Tax data for Russia were taken from Martinez-Vazquez and Wallace (2000).
Our model includes two policy variables and two control variables. The policy variables are tax (the marginal tax rate), and income, measured as permanent income, the sum of all monthly expenditures (Prais and Houthakker 1971). Tax increases with income under progressive tax structures such as Russia's. Because donations to charities are fully tax deductible, the effective cost of donating a dollar to charity (1 - tax) declines with tax; higher taxes should increase the probability of giving (Steinberg 1990 and 1997). (4) Following Clotfelter (1985), we expect giving to increase with income, but at a decreasing rate; we allow for this by including income squared. The two control variables are family size, the number of members in the household, and urban, a dummy variable denoting whether the household is in an urban area.
Naturally, the tax and income policy handles would be fairly roundabout, inefficient ways to affect charitable giving. Realistically, impacts on giving from changes in tax rates and transfer payments would likely be secondary side effects to policy goals such as poverty reduction or economic growth. These side effects can be important to policy makers, however, and can change the cost-benefit calculation for the policy tool. …