Hi,
I'm dealing with an issue that parallels (I think) the econometric issue of gender or race gaps in wages and earnings and the like, and I have learned that many studies looking to examine the factors responsible for gender or race gaps (i.e., group differences) in wages, etc., essentially estimate group specific equations, then use a decomposition method such as the Oaxaca-Blinder technique or something similar. My question to the list is whether this approach makes sense for my situation:
I'm interested in estimating the effects of changes in the ratio of black to white poverty, unemployment rates, etc. on changes in the ratio of black to white violence rates. The expectation is that reductions in black/white differences in things like poverty will translate to reductions in black/white differences in violence (through some theoretical process of frustration or strain or the like), and that these effects (or this process) should be stronger in cities in which black/white gaps in economic situations have declined the most. I have annual data on violence from 1960-2000 and decennial data on the explanatory variables (e.g., poverty) for 1960-2000, and I have this information for about 200 of the largest U.S. cities. Given that I don't have annual data for the explanatory variables, I'm thinking about using the available information to linearly interpolate 5 year change scores (e.g., 1960 and 1970 poverty to measure the change in poverty between 60-65, 65-70) and to compute my dependent variables (black/white violence) as three year averages centered around the mid decade points (this reduces substantially the instability in annual estimates). So, I'll have about 8 time data points.
Essentially, I have two group-specific dependent variables (white violence rates, black violence rates) and am interested in evaluating how the gap between them has changed over time and how much changes in that gap are due to changes in black poverty rates, white poverty rates, black-white poverty rates (and other features such as graduation rates, employment rates, family structure, etc.). This seems very similar to the basic issue confronted by economists who study race or gender wage or earnings differentials. Any thoughts about this?
A secondary question is that I have all data elements for about 200 cities, and I am wondering whether economists have developed techniques for addressing the factors associated with group differences in wages or earnings (etc.) generally, and how these might vary across cities (or states, or countries). For instance, one might predict that gender differences in earnings have always been smaller and perhaps have declined more quickly in Scandinavian nations--one could piece this together from a series of individual country studies, but it sure would be nice to develop a technique that does so within a single study. Any thoughts on these issues, as well as suggestions for literature I can evaluate (the least technical would be best for me since I am not an economist, but do have training in statistical methods more generally), and advice on which of the many packages to use to estimate something along these lines would be much appreciated.
Cheers,
|