Will the Long Term Historical Database at the Census Tract Level Improve the Accuracy of the Long Term Population Projection for Census Tracts?

S. Simon Choi, Southern California Association of Governments
Jung Won Son, University of California, Los Angeles

Population projection for census tracts (or transportation analysis zones) plays an important role in diverse areas of planning, both public and private. More precise projection makes possible more accurate regional transportation planning, local economic development planning, and regional air quality management among many. Inaccurate population projection might result in overinvestment or underinvestment that in turn brings about waste of resource or lowered quality of life for residents. However, it has been difficult to assess the accuracy of the long term population projection for census tracts due to data constraints. The purpose of this study is to assess if the long-term historical database improves the accuracy of population projection for census tracts using recently created historical database. Unlike the state and county boundary, the census tracts change their boundary depending on the growth of population causing difficulties in research; The conversion of old boundary to new boundary takes enormous efforts and time. The quality of population data based on converted boundary is not guaranteed. The historical database for census tracts is not readily available. The quality of the historical database is questionable. Fortunately, there are efforts to convert the historical census data with different boundary. As a result, researchers are able to access a relatively long historical database at the census tract level. Then, the question is, would the availability of the long term historical database at the census tract level improve the accuracy of the long term population projection? Using a simple linear extrapolation method, Smith and Sinich (1990) found that the long term historical database at the state level does not improve the accuracy of the population accuracy of the population projection. They conclude that the recent 10 year historical database is good enough to achieve the reasonable accuracy of the next 10 or 20 year projection of population at the state level. However, Isserman (1993) found that the long term historical trend at the county level improves the accuracy of the population projection. To answer this question, the proposed research focuses on Los Angeles County using 1940-1990 Los Angeles Census Tract database developed by Prof. Ethington et al at USC. The research procedure would be as following: first, we review the quality of the database. Tracts with notably outlying growth pattern will be checked. Second, we review the stability of the growth patterns of census tracts. If the growth pattern shows an unstable and fluctuating pattern, we prepare the synthetic database using the smoothing technique. Third, we prepare the long term population projection for census tracts by using the best method among available extrapolation or ratio methods. Two constraints will be introduced during the projection process. The first constraint is that the limited supply of land caused by the existing land use zoning will be introduced as a ceiling for the future population projection. The second constraint is that preliminary population projection for census tracts will be normalized to the county population projection. The simple raking method will be used to adjust the census tract population. The best method is selected by comparing the accuracy measures such as mean error (ME), mean squared error (MSE), or mean absolute percentage error (MAPE). Fourth, different base year periods are used to develop the population projection for census tracts. The projected population will be evaluated by comparing with the actual population for the base period.