Abstract High‐resolution gridded population estimates increasingly support research. They allow study at a finer spatial scale than the usual survey or administrative data (spatialization) and higher frequency than typical decadal census data (temporal interpolation). However, little is known about how accurately these gridded data follow actual changes in population. Therefore, we use China's census data for 2000, 2010, and 2020 to test the predictive accuracy of four popular gridded population data products, conducting our tests at three spatial levels (county/district, prefectural city, and province). Although gridded estimates understate inequality in the population spatial distribution, they fairly accurately predict cross‐sectionally at all three spatial levels, with under 5 percent of variation unexplained. They far less accurately predict temporal changes in population, especially for disaggregated spatial units (counties and districts), with less than one‐fifth of the variation in population changes predicted by the gridded data. Predictive accuracy of gridded data for population changes has fallen substantially in the last decade. We show how these inaccurate predictions could distort analyses that rely on population estimates to examine trends in spatial inequality. Overall, our results suggest that caution is required in using these gridded data products as proxies for the actual changes in local population.
Zhang et al. (2025) studied this question.