Corrigendum to “FDI and Income Inequality in Tax-Haven Countries: The Relevance of Tax Pressure” [Econ. Syst. 48 (2024) 101172] (Economic Systems (2024) 48(1), (S0939362523001115), (10.1016/j.ecosys.2023.101172))

The authors regret some minor changes to the following tables: Table 1. The new replicated results are reported in the revised Table 1 below. Revised Table 1: Descriptive Statistics [Table presented] The FDI stock variable was previously identified as logged in Table 1 of the published paper. The revised Table 1 presents the non-log-transformed data, now including the median value of these variables as customary. The decision to exclusively present the non-logged data for the FDI stock variable in the revised table aligns with standard research practices. This approach simplifies interpretation for a broad audience, as the non-logged values are more intuitive and directly reflect the actual magnitudes of the variables. Emphasizing the non-logged data in the table ensures a clear and straightforward representation of the FDI stock variable, prioritizing transparency and accessibility in conveying research outcomes. Also, incorporating the mean value alongside the non-logged FDI stock variable in the current table offers additional insights into the average magnitude of the data. The mean provides a measure of central tendency that complements the median data, offering a different perspective on the distribution of the variable, facilitating a more comprehensive understanding of the data's typical value and variability, which can be particularly useful for researchers looking to assess the overall trend or average performance. Table 2. The new replicated results are reported in the revised Table 2 below. Revised Table 2: FGLS Estimation [Table presented] Notes: Standard errors are reported in parenthesis. (*) denote de significance level for every estimate being (*) significant at 10 %, (**) at 5 % and (***) significant at 1 %. F-statistics for the joint hypothesis that all coefficients are different from zero is not reported as all models have provided significant evidence at 1 % that the joint significance of the estimates are different from zero.In the estimations of the HDI model in the published paper, Political Stability (PolSt) variable was non-significant, whereas in the replication, it was found to be significant at the 1 % level. The reason for this discrepancy remains uncertain, but it may be attributed to differences in model tuning, such as variations in estimation parameters or updates to the statistical software since the initial analysis was conducted with an older version. Another plausible explanation is the transposition of estimates during the replication process, which could impact on the statistical significance and interpretation of the results. Therefore, the divergence in findings likely reflects technical or procedural differences rather than substantive changes in the underlying relationships. Table 3. The new replicated results are reported in the revised Table 3 below. Revised Table 3: Tax revenue and FDI Kuznets curve estimation [Table presented] Minor differences in the magnitude of the initial estimates were observed during the replication process. It was identified that Model 3 had inadvertently omitted fixed effects, a crucial component that controls unobserved, time-invariant factors within the data. This omission has likely contributed to the discrepancies and has since been corrected in the latest version, ensuring more accurate and reliable estimates. Additionally, as previously stated, differences in the versions of the statistical software used between the initial estimation and the replication may also have influenced the results, as updates often include changes to estimation routines, default settings, and numerical algorithms, which can affect final outcomes. Table A3. The new replicated results are reported in the revised Table A3 below. Revised Table A3: Model Estimation [Table presented] Note: In the panel models, the subscripts i and t, respectively, denote the country and time dimension of the panel, and ε_it is the individual-level effects. The model in the last column is the random model feasible generalized least squares where variables are transformed using cross-section weights to correct for cross-section heteroskedasticity. The fixed-effect is an F-test of the null that the cross-country fixed effects are redundant. The Hausman procedure tests the null of orthogonality of the unobserved individual effects and the regressors under which both estimators are consistent with no systematic differences in their parameter estimates. Breusch Pagan LM test and Pesaran CD tests are cross-sectional dependence/ contemporaneous correlation test. The null hypothesis asserts that the residuals across entities are not correlated, therefore not having cross-sectional dependence. Wu-Hausman test assesses the presence of no endogeneity issues in the regression, concluding the exogenous nature of our variables. For the 2SLS regression, the instruments selected were FDI flows taken from the balance of payments (UNCTAD database), population (WB database), electricity consumption (EIA database), control of corruption (WB database), and expected years of schooling (UN database), plus lagged values of FDI stock, GDP growth rate, openness, urbanization, political stability, mean years of schooling and conflict. Weak instruments diagnosis proved strong for all instruments selected to the 1 % significance level. Hausman test asserts both 2SLS and OLS to be equally consistent, deeming appropriate OLS as it is more efficient than the former.The initial calculations for cross-section, time fixed effects, and heteroskedasticity tests were inadvertently conducted using incorrect procedures, resulting in initially reported invalid estimates. Specifically, the initial estimation relied on the F-test for individual and time effects to determine the presence of fixed effects, the Hausman test to compare fixed and random effects models, and the Breusch-Pagan Lagrange Multiplier test to assess time-fixed effects. However, the proper estimation approach should have incorporated the Breusch-Godfrey test for serial correlation and the Breusch-Pagan test for heteroskedasticity. These corrections have since been applied, and the revised estimates are presented in revised Table A3. Additionally, discrepancies between results may also arise from differences in software versions, as updates often change estimation algorithms and default settings. The authors would like to apologise for any inconveniences this caused.