Later, we will extend &(cid:19)(cid:20)(cid:4) to an RD polynomial using the distance to the Danube River as the running variable. We use a two-stage least squares (2SLS) procedure. Thus, we extend our difference-in-differences approach to a fuzzy spatial regression discontinuity (RD) design that allows us to address estimation biases due to local spill-overs and measurement errors. This strategy is rather new in a panel framework. So far, studies have applied a spatial fuzzy RD framework in the cross-section only, e.g., Eugster et al. (2011) and Basten and Betz (2013). Fuzzy RD is equivalent to an instrumental variable approach (Angrist and Pischke 2010). Our instrument is the location of a municipality with respect to the Danube River, conditional on

The instrument is highly relevant, as already indicated by the graphical inspection of Figure 1. As we will discuss below, first-stage F-statistics confirm this impression. We also conduct an RD plot, shown in Figure 7, that documents a sharp and significant spatial discontinuity in Turkish pillages across the Danube River in our West sample. The probability of pillages was approximately 0.6 for municipalities south of the Danube River (negative range in Figure 7); municipalities in the adjacent northern region (positive range) were only marginally exposed to Turkish pillages. 30 FIGURE 7. SPATIAL DISCONTINUITY IN TURKISH PILLAGES 8 . 0 6 . 0 s e g a l l i p h s i k r u T 4 . 0 2 . 0 0 . 0 100 50

0 Distance to Danube River 50 100 Notes: The figure shows an RDD plot of the Turkish pillages using the distance in kilometers to the Danube River as the running variable. The red dashed vertical line indicates the Danube River. The green lines show the local averages of municipalities likelihood of being pillaged based on a quadratic polynomial fit. Negative values of the distance to the Danube River refer to municipalities south of the Danube River; positive values refer to municipalities north of the Danube River. The gray lines indicate 95% confidence bands. For illustrative reasons, the number of bins equals 11 for each side of the threshold. We also have good reasons to believe that our instrument is excludable. First, and most importantly, the main reason for the sharp spatial discontinuity in pillages is the arguably exogenous location of the Battle of Bisamberg. Turkish troops stopped expanding north of the Danube

River west of Vienna because of their defeat. Second, even today, the Danube River west of Vienna is not a border between administrative districts in East Austria. Bridges and ferries allow crossings of the Danube River quite easily in the present day. Third, we find no evidence that the Danube River might have become an issue after 2005 but not in the years before. We do not observe changes in spatial discontinuities in covariates across the Danube River west of Vienna from the pre-campaigning to the campaigning period. Column (3) in Table 15 in the supple31 mentary material shows that the FP vote share is the sole exception, considering both its magnitude and its statistical significance. Furthermore, east of Vienna (longitude > 16.5), the Danube River is not a geographic threshold for long-gone exposure to Turkish atrocities, and neither do the FP vote shares show any differences from the pre-campaigning to the campaigning