Abstract
This article has been corrected since it was initially published OnlineFirst. Incorporating Greene’s (2010) suggestions for interpreting marginal effects, the article was revised to reflect a more precise interpretation of the interaction effects proposed in hypotheses 2–5 by removing the statistical tests of the marginal effects presented in tables 3–6. The marginal effects (tables 3–6) and interaction graphs (figures 1–4) are presented for illustrative purposes to assist the reader in evaluating the implications of the interaction results presented in models 3–6 in table 2. Text in the Method and Discussion sections was revised to correspond to the changes to the tables. The corrections to the article prevent readers from potentially misinterpreting the partial interaction effects; however, these corrections do not materially change the results of the study nor the inferences derived from the marginal effects analysis.
The following changes were made to the online version before the article was included in the June 2016 print issue of the Administrative Science Quarterly, 61: 254–290:
On page 21, in the paragraph under the heading Relationship Depth, “Thus to properly interpret our results” was deleted and the sentence begins with “We computed…”; in the next line, “and z-statistic” was deleted and a new sentence was inserted after the period: “While the calculated marginal effects of logit model interaction terms should not be the primary basis for hypothesis testing, they do provide an informative interpretation of the implications of the estimated model (Greene, 2010).” In the following sentence, “the interaction effect was statistically significant” was changed to “the calculated marginal effect appears to remain at nonzero values,” and “the sign of the interaction value” was changed to “sign of the calculated marginal effect.” “Tables 3–6 list the value and significance” was changed to “Tables 3–6 list the calculated marginal effect. . . .” In tables 3 and 4, the columns headed “z-statistic” have been deleted, as have bullets denoting the significance of the marginal effects at p < .01.
On page 22, the sentence beginning “Additionally, the true interaction effect is not significant” has been changed to “Additionally, the true interaction effect does not appear to be different from zero.”
On page 24, in tables 5 and 6, the columns headed “z-statistic” have been deleted, as have bullets denoting the significance of the marginal effects at p < .01.
On page 25, the first complete sentence, “We found support for our hypothesis only in the computed true interaction results,” has been deleted; the sentence, “The analysis of the true interaction effect, however, provides support for our hypothesis, as shown in table 6 and the plot in figure 4,” has been changed to “The analysis of the true interaction effect, as shown in table 6 and the plot in figure 4, implies that there may be a small marginal difference in the likely dissolution of shared versus unshared client ties.”
On page 28, “Lastly, we found that sharing an exchange relationship also affects the likelihood of dissolution, though only slightly,” has been changed to “Lastly, the results from our models imply that sharing an exchange relationship may affect the likelihood of dissolution, though only slightly.” The sentences, “When lobbyists leave their firms, however, they are more likely to lose a shared client relationship than a solely served client relationship. This finding highlights how the effects. . . .” have been changed to “When lobbyists leave their firms, however, they may be more likely to lose a shared client relationship than a solely served client relationship. This implication highlights how the effects. . . .”
On page 33, the following reference was added: Greene, W., 2010 “Testing hypotheses about interaction terms in nonlinear models.” Economics Letters, 107: 291–296.