This paper examines the local power of unit root tests for panel data, focusing on the bias-adjusted $t$-statistics proposed by Levin and Lin (1993) and Im, Pesaran, and Shin (1997). The study reveals that the local power of these tests is influenced by two main terms: the asymptotic effect of the detrending method on the bias and the location parameter of the limiting distribution under the sequence of local alternatives. It is shown that these two terms can offset each other, leading to a lack of power against the sequence of local alternatives. To address this issue, the paper suggests a class of $t$-statistics that do not require bias correction, based on transformed variables from a simple least-squares regression. Monte Carlo experiments demonstrate that avoiding bias correction can significantly improve the power of the test. The paper also discusses the asymptotic properties of the tests, including the limiting distribution and the impact of the number of time periods and cross-section units on the test's performance. Finally, it highlights the sensitivity of the tests to the specification of deterministic terms and the choice of augmentation lag, suggesting that a test for a common deterministic trend against individual-specific time trends is more desirable in practice.This paper examines the local power of unit root tests for panel data, focusing on the bias-adjusted $t$-statistics proposed by Levin and Lin (1993) and Im, Pesaran, and Shin (1997). The study reveals that the local power of these tests is influenced by two main terms: the asymptotic effect of the detrending method on the bias and the location parameter of the limiting distribution under the sequence of local alternatives. It is shown that these two terms can offset each other, leading to a lack of power against the sequence of local alternatives. To address this issue, the paper suggests a class of $t$-statistics that do not require bias correction, based on transformed variables from a simple least-squares regression. Monte Carlo experiments demonstrate that avoiding bias correction can significantly improve the power of the test. The paper also discusses the asymptotic properties of the tests, including the limiting distribution and the impact of the number of time periods and cross-section units on the test's performance. Finally, it highlights the sensitivity of the tests to the specification of deterministic terms and the choice of augmentation lag, suggesting that a test for a common deterministic trend against individual-specific time trends is more desirable in practice.