This paper re-examines Lilien's sectoral shifts hypothesis for U.S. unemployment. We employ a monthly panel that spans from 1990:01 to 2011:12 for 48 U.S. states. Panel unit root tests that allow for crosssectional dependence reveal the stationarity of unemployment. Within a framework that takes into account dynamics, parameter heterogeneity and cross-sectional dependence in the panel, we show that sectoral reallocation is significant not only at the aggregate level but also at the state level. The magnitude and the statistical significance of the latter as measured by Lilien's index increases when both heterogeneity and cross-sectional dependence are taken into account.