RETRACTED: Psychological processes linking authentic leadership to follower behaviors

This article has been retracted: please see Elsevier Policy on Article Withdrawal (http://www.elsevier.com/locate/withdrawalpolicy).This article has been retracted at the request of the Senior Editor.After concerns were raised about possible problems of reporting in this paper, the Senior Editor consulted with the two previous Senior Editors of The Leadership Quarterly and a methodologist (M1) (not the claimant) to assess the seriousness of the allegations and to make a preliminary determination concerning the allegations’ merits. All concurred that there were serious problems in this paper. The methodologist (M1) prepared a report outlining the problems and this report was forwarded to a second methodologist (M2) to confirm the correctness of methods used by the first methodologist to detect the problems. The second methodologist attested to the correctness of the first methodologist's analyses. The Senior Editor then contacted the authors to inform them of the problems identified in the paper. The authors were asked to respond to concerns raised and encouraged to send the original data from this paper to the Senior Editor for reanalysis.The authors did not provide the original data, but rather sent a letter replying to the methodology report, along with new analytic results. These new results were reviewed by a third methodologist (M3) as well as the methodologist who prepared the report (M1). Both agreed that the reanalysis failed to replicate the results that were originally reported and further supported concerns about poor model fits, serious reporting errors, model misspecifications, and methodological misstatements in the published article.The Senior Editor has concluded that the published paper included misreported findings with respect to model fit statistics, as well as a problem in that the model as stated in the published paper was not tested. This compromised the scientific review process. More specifically, the paper reported on page 906, that a higher-order model consisting of four factors had a better fit to the data than did a first-order model, which is not possible. In addition, on the basis of what would be expected given the information reported in the paper, the dfs reported for the higher-order and first-order models were incorrect and the chi-square of the higher-order model (i.e., 201.76) was reported to be lower than that of the first-order model (i.e., 259.67), which is not possible because the higher-order model employs more constraints, thus has more dfs and necessarily a higher discrepancy (chi-square) statistic. Moreover, the RMSEA of the higher-order model was reported to be .05 in the paper; based on the information reported in the published paper, it should have been .09 (thus, this result made the higher-order model look better than originally reported). The RMSEA of the first order model was reported to be .07, though it should have been .11 (also making this model look better than conveyed); according to the authors’ response, the discrepancies with respect to the RMSEAs are due to the fact that the authors had “erred by entering a textual error and that the analysis was actually done at the individual level (N = 387).” Yet, in the article they had expressly noted that they did the CFAs using “group-level data because we conceptualized authentic leadership as a group-level construct in our analysis” (p. 906). Irregularities can be seen on page 908; specifically there are several incorrectly reported fit statistics that make the alternative models look worse than actual. For example, for the first reported alternative model (having a χ2 of 6051.29), the approximate fit indexes were reported to be: TLI = .70, CFI = .71, and RMSEA = 09. Based on the information reported in the paper, these fit statistics should have been TLI = .91, CFI = .95, and RMSEA = .06. With respect to these irregular results, the authors responded that they may have switched from using a group level analysis to an individual level analysis prior to submitting the paper, which would have impacted how the RMSEA was reported for the alternative models. The authors, though, had to speculate as to whether this was the explanation. Additionally, they stated that they erred in reporting the CFI and TLI of the target model (note, what should have been reported for the CFIs and TLIs for the alternative models can be reconstructed using information from the target model, whose reported results have been acknowledged by the authors to have been incorrect). As for the discrepancies in the degrees of freedom for comparing the higher- and first-order models, the authors also acknowledged correlating disturbances in their models; also, that the df differences between the higher-order and first-order model was 1 instead of 2 because the authors stated that they “had inadvertently estimated 1 additional parameter (a random error covariance) in the higher order model.” The use of correlated disturbances was not specifically mentioned in the article making it impossible for other researchers to follow what they did or to replicate their findings. The practice of correlating disturbance terms of measurement items, on the basis of modification indexes as disclosed by the authors, capitalizes on chance, increases model fit statistics, and can compromise model estimates (cf. Brown, 2006; Gerbing & Anderson, 1984; Maccallum, Roznowski & Necowitz, 1992; Steiger, 1990). Furthermore, the authors were unable to reproduce their own results in part stating that they had not maintained records of which disturbance terms were correlated (e.g., in the case of alternative models, where they reported in their response that 19 disturbances had been correlated). Additionally, many of the models re-estimated showed inadmissible solutions in the statistical output submitted in their response (e.g., negative error variances or non-positive definite covariance matrixes). Inadmissible solutions can be indicative of model misspecification. Finally, models were estimated at the individual level, without declaring this fact in the review process, which could substantially bias estimates and model fit statistics (i.e., they did not use a cluster or “sandwich” correction for the chi-square statistic, on which all fit statistics depend, nor did they correct standard errors of the estimates for clustering, cf. Muthén & Satorra, 1995).As a consequence of the processes and concerns outlined above, the scientific trustworthiness and value of this work cannot be established. However, intentional wrongdoing should not be inferred.References:Brown, T. A. 2006. Confirmatory factor analysis for applied research. New York: Guilford Press. http://www.guilford.com/books/Confirmatory-Factor-Analysis-for-Applied-Research/Timothy-A-Brown/9781462515363Gerbing, D. W. & Anderson, J. C. 1984. On the meaning of within-factor correlated measurement errors. Journal of Consumer Research, 11(1): 572-580. http://www.jstor.org/stable/2489144Maccallum, R. C., Roznowski, M., & Necowitz, L. B. 1992. Model modification in covariance structure-analysis: The problem of capitalization on chance. Psychological Bulletin, 111(3), 490-504. doi: 10.1037/0033-2909.111.3.490Muthén, B. & Satorra, A. 1995. Complex sample data in structural equation modeling. In P. V. Marsden (Ed.), Sociological Methodology: 267-316. Washington, DC: American Sociological Association.Steiger, J. H. 1990. Structural model evaluation and modification: An interval estimation approach. Multivariate Behavioral Research, 25: 173-180. doi: 10.1207/s15327906mbr2502_4