A large literature on historical persistence finds that many modern outcomes strongly reflect characteristics of the same places in the distant past. However, the explanatory power of these regressions frequently rises when the persistence variable is replaced with synthetic spatial noise, indicating that these findings may be less straightforward than they appear. We show that although spatial standard error corrections are too fragile to be useful in empirical research, trustworthy estimates are easily obtained by adding a thin plate spline in longitude and latitude to the regression. For 40 studies in major journals, splines typically cause the explanatory power of the historical variable to fall substantially, suggesting that many of the best known persistence results are spurious artefacts of matching low frequency spatial patterns in the data.