Corrigendum to ‘Not every coopetitor is the same: The impact of technological, market and geographical overlap with coopetitors on firms’ breakthrough inventions’ [Long Range Plan 53 (2020) 101873] (Long Range Planning (2020) 53(1), (S0024630118301031), (10.1016/j.lrp.2019.02.006))

The authors have self-identified errors in the above published article. In this corrigendum, the authors provide an overview of the identified errors, report the results after correcting for these errors and discuss the implications of the corrected results. The first author of this paper, who executed the data collection and data analysis, voluntarily acknowledged these errors. After investigating the data sampling, operationalization of variables and data analyses, the second and third authors have concluded that the errors were not of an intentional nature. In the published article, Tables 2–4 mentioned in the notes that the number of observations was 2582. Although this number represented the size of the initial sample, which was created for this research project, it does not represent the actual 214 observations in the regression models. The difference was caused by (i) use of a three-year window for the dependent variable (n = 2582–110 = 2472), (ii) missing values in the control variables R&D intensity and firm size (n = 2472–1632 = 840) and (iii) use of fixed effects negative binomial model that excludes observations with all zero outcomes or firms with only one observation (n = 840–622 – 4 = 214). An in-depth assessment of the initial sampling of the data revealed that (i) some observations before 1995 or after 2015 were included in the initial sample, which did not fit within the panel of 1995–2015; (ii) the existence of all alliances was assumed to be three years, which should have been 5 years as mentioned in the published article; and (iii) matching of firms with the Compustat database based on company name was incomplete. In the corrected results, we have excluded observations before 1995 and after 2015, adjusted the existence of all alliances to five years, and searched all public firms in the Compustat database and added available data for R&D intensity and firm size. These corrections generate a corrected initial sample of 2292 observations. In the new sample, there are 81.63% of our observations collaborating with one coopetitor in a particular year. When running regression analysis, the final sample is reduced to 237 observations from 28 groups, because of (i) use of a three-year window for the dependent variable (n = 2292–731 = 1561), (ii) missing values in R&D intensity and firm size (n = 1561–929 = 632), and (iii) use of fixed effects negative binomial model that excludes firms with all zero outcomes or firms with only one observation (n = 632–382 – 13 = 237). In the corrected results, we report the number of observations per model in Tables 2–4 In Table 3, Models (1)–(3) have 250 observations from 30 groups because the observations with all zero outcomes are reduced when the 95% percentile, instead of the 97% percentile is used to calculate breakthrough inventions (n = 237 + 13 = 250). In Table 3, Models (10)–(12) have 632 observations from 116 groups because a Tobit model is used. Similarly, the first stage endogeneity test in Table 4, Model (1) has 632 observations because a linear model is used. We have found out that, in the original results, an error was made in the standardization of the variable geographical overlap. To generate the correct results, we have carefully reconstructed all variables, including technological, market and geographic overlap variables, from the raw data. When reconstructing these overlap variables, we further found that some missing values for the core independent variables were mistakenly transformed into zeros and we corrected for it. In the published article, while the mean and standard deviation of all other variables reported in Table 1 were based on 2472 observations after calculating the dependent variable in a three-year time window, the mean and standard deviation of seven variables were based on all of 2582 observations. In the corrected results, we consistently report descriptive statistics for all variables based on the corrected initial sample of 2292 observations, and clearly indicate the corresponding number of observations per variable. In line with the core analyses, the correlations in Table 1 are based on the final sample of 237 observations. In Table 4, Model (3) of the published article, patent class dummies were absent in the analysis. In the corrected results, patent class dummies are added in all models. Because of changes in the sample, all models now include 29 patent class dummies. The corrected results are reported in Tables 1–4 below. Differences in the descriptive statistics are caused by (i) removing observations that did not fit in the 1995–2015 panel, (ii) correcting the overlap variables, and (iii) optimizing the matching with the Compustat database. As Table 2 illustrates, there is still evidence in support of Hypothesis 1 and Hypothesis 2. Moreover, Table 3 shows that the hypothesized interaction effect between technological overlap and market overlap is robust across different specifications. In Table 3, Model (12), this interaction effect slightly exceeds the significance threshold (p = 0.120). The interpretation of this interaction effect remains similar. The non-linear relationship becomes steeper when market overlap increases. Consistent with what was reported in the published article, the evidence for Hypothesis 3 is not robust. In the published article, this lack of robustness was mainly driven by the variables considered in the models. In the corrected results, this lack of robustness is mainly driven by the modelling approach. The authors would like to apologise for any inconvenience caused.