Small‐area population forecasting: borrowing strength across space and time

Abstract

Demographic forecasting techniques do not perform particularly well for small areas. In this study we propose a spatio‐temporal regression approach for small‐area population forecasting that borrows strength from what has happened nearby and what has happened in the past. In particular, a regression model incorporating temporally lagged neighbour growth and neighbour characteristics is applied to examine population change at the minor civil division (MCD) governmental level in Wisconsin, USA since 1960. For each MCD, the population growth rate for 1980–1990 is regressed on its growth rate for 1970–1980, its various characteristics in 1980, neighbour growth rates for 1970–1980, and neighbour characteristics in 1980. The estimated regression coefficients are then used for projecting population in 2000. Accuracy of the forecasts is measured against the 2000 Census counts. The state's official MCD projections for 2000, which were based on an extrapolation projection – the most often used traditional population forecasting approach for small geographic areas – are taken as the ‘gold standard’ against which improvements through the regression formulations are sought. The projection evaluations reveal mixed results and do not suggest unambiguous preference for the spatio‐temporal regression approach or the extrapolation projection. We discuss several reasons for the inability of our spatio‐temporal forecasting model to outshine the simple extrapolation forecast. Although this is disappointing, the proposed approach is more solidly theoretically grounded and provides useful information to policy and decision makers at the community level regarding the consequences of various developmental strategies being adopted by themselves as well as by their neighbours. Copyright © 2010 John Wiley & Sons, Ltd.

Small-area population forecasting in an urban setting: a spatial regression approach

“…First, the city of Milwaukee is within the state of Wisconsin, which makes the findings of this study comparable to the findings of Chi (2009) and Chi and Voss (2011) because both are conducted at the MCD level in Wisconsin. Second, the knowledge-based spatial regression approach is argued to work better in smaller geographic areas where population change relies more on migratory factors than births and deaths (Chi and Voss 2011). In Wisconsin, census tracts on average are similar in size to MCDs, which average 76.53 km 2 .…”

“…Grid-cell-based population forecasting often uses remote sensing data (Riahi and Nakicenovic 2007). However, these three types of models require significantly extensive data capacity and well-grounded expertise in spatial analysis and statistics, and they are expensive to develop and implement (Chi and Voss 2011). There is a large literature on these models; see Chi (2009) and Wilson and Rees (2005) for discussion.…”

“…However, the findings indicate that the knowledge-based approach does not outperform simple extrapolation which is currently the most commonly used method for small-area forecasting. Furthermore, Chi and Voss (2011) proposed to add spatial process effects into the forecasting process by considering the growth and characteristics of neighbouring areas. While their method is theoretically sound, their projections are not unambiguously superior to simple extrapolations.…”

“…In this study, we select the census tracts of the city of Milwaukee, Wisconsin, USA, to test whether the knowledge-based spatial regression approach developed by Chi and Voss (2011) performs well for small-area population forecasting in an urban setting. Milwaukee is selected for two reasons.…”

“…The next section describes the knowledge-based spatial regression approach developed by Chi and Voss (2011). The data and analytical approach are then introduced.…”