The majority of empirical research in economics ignores the potential benefits of nonparametric methods, while the majority of advances in nonparametric theory ignore the problems faced in applied econometrics. A new book by IZA fellow Daniel J. Henderson (University of Alabama) and Christopher F. Parmeter (University of Miami) helps bridge this gap. In an interview with IZA Newsroom, the authors explain what it’s all about.
What are nonparametric methods? Sounds like pretty dry stuff…
It’s all in the pitch. Nonparametric regression is simply a special case of classic linear regression. Anyone who understands weighted least-squares can understand nonparametric regression. For example, local-linear regression is simply linear regression local to a point of interest. That is, instead of weighting via say heteroskedastic errors, we give more weight to observations near the point of interest. When this is repeated over a grid of points, we end up with a smooth function that does not require a priori restrictions and hence better fits the data.
Why is this particularly relevant for labor economists?
Labor economists recognize that the impact of a policy is likely heterogeneous across a population. Some individuals may benefit more than others while some may actually be worsened. These differences often result from underlying nonlinearities in the relationships between variables policy makers have control over and the observed outcomes. Typically the underlying relationship is not known to the researcher/policy maker and hence many labor economists are turning to nonparametric methods.
Can you give an example from the real world?
In a recent IZA Discussion Paper, Ozabaci, Henderson and Su look at the relationship between child care use by single mothers in the United States and their children’s subsequent test scores. The choice of when/whether a woman should return to work is an important question and it is also relevant whether or not the government should subsidize such care. The use of nonparametric methods here confirms many of the findings in the literature (e.g., negative returns to child care when mothers have higher levels of education), but as the estimators allow for heterogeneity both across and within groups, they are able to contradict many findings in the literature. Specifically, they are able to show that it is the amount (more child care leads to more negative returns) and not type (formal versus informal) of child care that matters.
What was your motivation for writing this book?
We believe the majority of empirical research in economics still ignores the potential benefits of nonparametric methods. Applied economists do not necessarily dismiss these methods because they do not like them. We believe a major reason many do not employ them is because they do not understand how to use them. Our book helps bridge the gap between applied economists and nonparametric econometricians/statisticians by teaching the methods in terms that someone with one year of graduate econometrics can understand.