Toward a Health Care Satellite Account

Ana Aizcorbe, BEA chief economist, noted that BEA does not gather data the way the U.S. Census Bureau does, but is engaged in an accounting or measurement of expenditures and other economic activities. BEA conducts research on their measures—health care is an area of recent focus. The goal is to improve measures of health care in the national accounts. Aizcorbe explained some of the challenges associated with this goal.

One challenge is how to best measure the services provided by the health care sector and the extent to which health care expenditures are increasing or decreasing. To illustrate the challenge—and the ways of looking at health care expenditures—Aizcorbe used the example of depression. The typical Bureau of Labor Statistics (BLS) approach would be to measure expenditures for specific services or treatments, while health economists tend to focus on expenditures by disease. In the case of depression, treatment was once limited primarily to the costly option of “talk therapy,” but when drug therapy became an option, people tended to substitute that less-expensive option. Thus, from the perspective of spending by treatment, the cost of treating depression has increased (i.e., the cost of talk therapy has increased), but from the “treatment of disease” perspective, the cost has decreased because of the substitution of the less-expensive treatment.

Therefore, one has to consider whether the treatments available for specific diseases have changed recently, and whether the total number of people being treated has changed (i.e., with lower cost options, more people might seek treatment). BEA is trying to identify diseases where spending by treatment vs. spending by disease is important. The work is still preliminary and further complicated by the tendency for some diseases to occur in groups. Still, some of the early data indicate treatment-based measures suggest higher increases in health care costs than do the disease-based measures. Findings so far—based on an extensive list of diseases—suggest treatment of disease–based measures usually show lower growth in health care costs.

Population Estimates Testing and Evaluation Research

Victoria Velkoff of the U.S. Census Bureau described the bureau’s plans for evaluating their population estimates against the results of the 2010 Census (a program they call Estimates Evaluation E2). The objective is to evaluate not only the methods they currently use, but a number of alternative methods.

Currently, the U.S. Census Bureau uses the administrative records method (ADREC) for county estimates and the housing unit method (HU) for subcounty estimates. Velkoff recalled the Housing Unit Based Estimates Research Team (HUBERT) project undertaken in response to recommendations that they consider HU methods for the county estimates. The HUBERT results found that the ADREC method has produced more accurate estimates than HU methods in most counties. However, some HU method proponents are concerned that ADREC showed no advantage with respect to bias, so the bureau is taking a further look at HU methods. The objective is to document the ultimate decision following the 2010 Census.

Phase one of the evaluation is to develop estimation principles and select accuracy measures. Phase two is to select alternative methods to test, develop official and alternative estimates, and carry out the evaluation. The timeline calls for the completion of phase one in April 2009 (already done), the determination of alternative methods by July 2009, the production of evaluation estimates by December 2010, the completion of evaluations by October 2011, and decisions on post-2010 methodology by spring 2012. Evaluations will cover total population and demographic characteristics, and the U.S. Census Bureau will look to external researchers to help evaluate specific methods, such as ratio correlation.

Jason Devine of the U.S. Census Bureau described the underlying principles of the estimates—items not subject to change. Among these are that the population estimates will reflect the census concept of usual residence, the most recent census counts will serve as the estimates base, and priority is given to a lack of bias because of the use in funds distribution. The estimates must be produced within deadlines determined by U.S. Census Bureau staff, estimates within a vintage must sum to others of that vintage, and each vintage must include a time series from the last census.

Methodological principles call for soundness (solid reasoning), accountability (understandable by many parties), availability of data (for all areas of the United States), availability of resources, robustness (insensitive to small departures from assumptions), comparability, adaptability, parsimony, and reasonableness (in terms of accuracy, demographic appropriateness, and external comparisons). Accuracy is defined as the degree of closeness to the 2010 Census values. Devine listed four properties of “good” estimates, drawn from a 1970s review of the census estimates program by the Committee on National Statistics. These include low average numeric error, low average percent error, few extreme percent errors, and the absence of bias for subgroups. Devine also identified the following five selected measures of accuracy, chosen after a review of 19 measures:

1. Average numeric error (root mean square error)
2. Average percent error (MAPE)
3. Extreme percent error (N greater than an established threshold)
4. Bias (MALPE)
5. Accuracy of share of total population (absolute error of shares)

Table 1 shows measures for 2000 county population estimates for ADREC and HU methods.

Devine concluded by identifying key points for the upcoming evaluations. These include focusing on the most promising alternatives, using production requirements and underlying principles to guide decisions, providing the most comprehensive evaluation possible with limited resources, and providing data sets that will allow others to assess the accuracy of the estimates.