About the Hope Street Group Weightings

Each variable in the Economic Opportunity Index (EOI) was chosen because of its predictive power for economic opportunity, i.e. future real income. Therefore, by construction, all of the indicators in the EOI are somehow correlated with economic opportunity (either positively or negatively). However, the strength of the effect may vary considerably from one indicator to the next. For example, a one-percentage point increase in the labor force participation rate is probably more important for economic opportunity than a one-percentage point increase in the share of college graduates-if people are not in the labor force at all, that is much more of a hindrance to economic opportunity than the lack of a college degree.

Ideally, we could calculate the relative importance of each variable using purely statistical methods. Specifically, we could use the economist's traditional tool of regression analysis on historical data to isolate the impact of each indicator in the EOI on real income in later years. The larger the impact revealed via the regression analysis, the larger the weight of a particular indicator in the EOI.

Unfortunately, several statistical problems make this difficult. Two are particularly important. First, many of the variables in the index are correlated with one another (a simple example of this is that many rise over time). This makes it harder from a statistical standpoint to tease apart the individual impact of each. Second, and more subtly, many of the variables in the index actually have effects on each other. For example, education is a major driver of economic opportunity, but education also impacts health outcomes, which in turn affect economic opportunity as well. In this case it is very challenging to try to determine what portion of economic opportunity is impacted by education and what fraction is affected by health care. Third, given data limitations for many indicators in the EOI, we have only a limited span of historical data (back to 1980 in most cases) and a limited cross-section of different demographic groups. We also have only one country, at least in the current version of the EOI. More years, demographic groups, or countries would increase the usefulness of the regression approach.

As a result of these problems, we are unable to find statistical models that consistently predict the past behavior of real income growth. We therefore turned to an alternative approach - instead of applying regression analysis, we examined the impact of each variable individually on income and/or life expectancy through literature reviews. As a simplified example of our approach, consider the weighting of the college graduation rate. We found journal articles that measured the impact of college degrees on future income; e.g. Ashenfelter (1991)¹ finds a 40-50% increase in earnings between 12 and 16 years of education. As the literature generally has shown an increasing premium to education in recent years, we took the top end of this range (50%). Therefore, for every 1 percentage point increase in the number of college graduates, expected lifetime earnings rises roughly 1% * 50% = 0.5%. Every variable is benchmarked to the Index in this same manner and ultimately all of the variables are adjusted in accordance to their relative weights with one another to arrive at the final weights.²

----

¹ Ashenfelter, O. (1991). "How Convincing Is the Evidence Linking Education and Income?" ² Note that the "normalization" of our EOI indicators (conversion to a 0-100 scale for easy comparison from one to the next) complicates this process slightly. An indicator needs to change to have an impact on economic opportunity, so indicators with a large "raw" weight from the process above but relatively little variation over time will end up having a smaller weight than indicators with a similar "raw" weight but that show much more variation over time. This is an important reason why the macroeconomic and regulatory variables in our index have smaller weights (we have them only for the US as a whole, and we are not including other countries in our analysis-at least not yet-so we see less variation than we do in indicators that vary a lot from one race or gender to the other.