This paper examines issues in the estimation of time-series cross-section (TSCS) models, challenging the conclusions of many published studies, especially in comparative political economy. The Parks method for generalized least squares (GLS) produces standard errors that lead to extreme overconfidence, often underestimating variability by 50% or more. An alternative estimator of standard errors is proposed, which is correct when error structures are complex. Monte Carlo analysis shows that "panel-corrected standard errors" perform well. The utility of this approach is demonstrated through a reanalysis of a "social democratic corporatist" model. The Parks method, while based on GLS, is problematic because it assumes knowledge of the error process, which is rarely available. Instead, feasible generalized least squares (FGLS) is used, which estimates the error process. However, the FGLS formula for standard errors assumes the error process is known, not estimated. In TSCS models, this leads to underestimation of true variability. The paper provides evidence from Monte Carlo experiments showing that Parks standard errors can underestimate variability by 50% to 300%. The paper advocates a simpler method for estimating TSCS models, combining ordinary least squares (OLS) parameter estimates with panel-corrected standard errors. Monte Carlo analysis shows that these new estimates of sampling variability are accurate, even with complex panel error structures. The paper details the problems with the Parks method, the structure of TSCS models, and why OLS is problematic. It also explains the flaws of the Parks method, including its overconfidence in results. The paper presents a Monte Carlo analysis showing that the Parks method leads to extreme overconfidence in standard errors. It also shows that panel-corrected standard errors perform well, even with complex error structures. The paper reanalyzes Hicks and Swank's study, finding that many of their conclusions are artifacts of the Parks method. It also reanalyzes other studies, finding that their results may be due to numerical inaccuracies or inappropriate use of the Parks method. The paper concludes that panel-corrected standard errors should replace OLS standard errors for TSCS data.This paper examines issues in the estimation of time-series cross-section (TSCS) models, challenging the conclusions of many published studies, especially in comparative political economy. The Parks method for generalized least squares (GLS) produces standard errors that lead to extreme overconfidence, often underestimating variability by 50% or more. An alternative estimator of standard errors is proposed, which is correct when error structures are complex. Monte Carlo analysis shows that "panel-corrected standard errors" perform well. The utility of this approach is demonstrated through a reanalysis of a "social democratic corporatist" model. The Parks method, while based on GLS, is problematic because it assumes knowledge of the error process, which is rarely available. Instead, feasible generalized least squares (FGLS) is used, which estimates the error process. However, the FGLS formula for standard errors assumes the error process is known, not estimated. In TSCS models, this leads to underestimation of true variability. The paper provides evidence from Monte Carlo experiments showing that Parks standard errors can underestimate variability by 50% to 300%. The paper advocates a simpler method for estimating TSCS models, combining ordinary least squares (OLS) parameter estimates with panel-corrected standard errors. Monte Carlo analysis shows that these new estimates of sampling variability are accurate, even with complex panel error structures. The paper details the problems with the Parks method, the structure of TSCS models, and why OLS is problematic. It also explains the flaws of the Parks method, including its overconfidence in results. The paper presents a Monte Carlo analysis showing that the Parks method leads to extreme overconfidence in standard errors. It also shows that panel-corrected standard errors perform well, even with complex error structures. The paper reanalyzes Hicks and Swank's study, finding that many of their conclusions are artifacts of the Parks method. It also reanalyzes other studies, finding that their results may be due to numerical inaccuracies or inappropriate use of the Parks method. The paper concludes that panel-corrected standard errors should replace OLS standard errors for TSCS data.