The authors regret three errors in Hager and Krakowski (Reference Hager and Krakowski2022).

First, the conditional IV models in Table 3 on page 574 included controls in the first stage only. When adding controls to the second stage, too, Table 1 shows that the coefficient of interest becomes stronger and more precise for the sabotage outcome (compare the original model in column 6 to the updated specification in column 7), but weaker and no longer statistically significant for the protest outcome (compare columns 2 to 3). In the revised Online Appendix (https://doi.org/10.7910/DVN/JWTRSC), we implement the same adjustment to Tables A13 and A14. In both tables, relevant comparisons are between columns 2 and 3 for the protest outcome, and columns 6 and 7 for the sabotage outcome. In Table A13, the effect of surveillance on both outcomes is reduced but remains statistically significant at the 90% (protest) or 99% (sabotage) level. In Table A14, the analogous estimates are no longer statistically significant.

Table 1. Corrected Table 3: Surveillance and Resistance (IV)

Notes: The Table reports coefficients from regressions of the indicated resistance outcomes on the number of secret police officers instrumented by the number of corrupted Catholic priests. The time series data was collapsed by taking the average of secret police officers and the respective resistance outcome in any given community across the time periods where outcome data are available (years 1980-86 for the regressions of protests and years 1975-79 for the regressions of sabotage). All control variables are cross-sectional and refer to the period indicated in Table A1. Standard errors are given in parentheses. All outcomes and control variables are standardized. Fixed effects refer to voivodships.

Second, the manuscript states that the models in published Tables 3, A11, A13, and A14 use the same controls as in Table A6. In reality, we used slightly different controls. (Comparing columns 3 vs 4 and 7 vs 8 of Table 1 indicates which controls were omitted.) Additionally, the industrialization index was created by averaging across minerals and coal variables, while in Table A6 we only used the minerals variable. In Table 1, we provide the updated estimates when using the controls from Table A6. There are no substantive changes. In the revised Online Appendix (https://doi.org/10.7910/DVN/JWTRSC), we implement analogous corrections to Tables A11 (columns 3 and 6), A13 (columns 4 and 8), and A14 (columns 4 and 8). The results remain substantively unchanged, except for the effect of surveillance on protest in Table A13 which is no longer statistically significant (column 4).

Third, the table notes in the manuscript were unclear on how we aggregated the panel data in the cross-sectional analyses. In the revised notes to published Table 3 (Table 1) and updated Tables A13 and A14, we clarify that the protest, sabotage, and surveillance variables are the mean levels of these variables in any given community across the time periods where outcome data are available (years 1980-86 for protests and years 1975-79 for the sabotage).

Overall, the effect of surveillance on sabotage remains unchanged when correcting the mistakes. By contrast, the effect of surveillance on protest is attenuated in the corrected conditional IV model (and no longer statistically significant). We should mention, however, that the conditional IV model is only one of several pieces of evidence in support of this relationship. Moreover, the corrected analyses rely on a small number of observations and therefore suffer from low statistical power. Our multi-method approach thus highlights the value of using multiple types of analyses when there are data limitations.