Lectures on MethodsMeasuring Risk Attitudes for Prediction and Explanation: A Discussion of Some Issues
Nat Wilcox
University of Houston
2:00-3:30 pm, Thursday, June 25, 2009
Abstract
The measurement of risk attitudes, by means of some elicitation task(s), is ubiquitous in economics experiments. Sometimes risk attitude measurements are dependent variables in their own right. More frequently, risk attitudes are measured so that the experimenter can control for variations in risk attitudes in some other primary or target task (e.g. an auction or game). A similar purpose is to examine the consistency of risk attitudes inferred from two or more tasks. I will concentrate on some issues associated with these uses of measured risk attitudes, with special emphasis on two topics. First, I will consider the reliability of measured risk attitudes: How reliability is a product of the structure of elicitation tasks, and the consequences of typical levels of reliability for explanation and prediction. Second, I will consider variations in task context. I will show that when elicitation task context and target task context differ, there is no guarantee that risk attitudes measured in the two tasks will be consistent unless very specific estimation techniques are used. These techniques will be discussed as well.
Econometric Challenges in Experimental Economics
Joachim Winter
Department of Economics
University of Munich
3:45-5:15 pm, Thursday, June 25, 2009
Abstract
As experimental economics developed into a field of economics, its methodology was based on “using controlled experiments to learn about economic behavior” (quoted from the Economic Science Association’s mission statement). An important advantage of controlled experiments is that appropriate randomization facilitates inference on causal effects. Statistical analysis of the data can be based on simple, nonparametric tests, exploiting the random assignment of treatments. A newer trend in experimental economics is to analyze the data by estimating econometric models: Multiple regression and other more advanced econometric methods are used to sharpen statistical inference by controlling for random variation of covariates across treatments, or to learn about the effects of covariates that have not been set randomly by controlled treatments (such as subjects’ beliefs, behavior in past rounds of repeated interactions, or average behavior in a group of subjects). Econometric models are also used in preference elicitation (e.g., measurement of risk attitudes). Using econometric models has broadened the scope of issues experimental economists can investigate, but it has also created new challenges because (causal) inference might be distorted under certain circumstances. For instance, experimental economists now need consider problems such as endogeneity of explanatory variables. Small samples are also potentially problematic for estimating econometric models since many estimators are only consistent but not unbiased and may have poor small-sample properties. This lecture discusses the econometric challenges faced by experimental economists, in particular once the econometric model contains explanatory variables that are not set by random treatment.
 |
|
