Abstract
Stoltz, Dustin S., and Marshall A. Taylor. 2019. “Textual Spanning: Finding Discursive Holes in Text Networks.” Socius 5. (Original doi: 10.1177/2378023119827674.)
This article has been revised and republished because of a coding error in the original publication. Namely, the R function listed in the Appendix (“Textual Spanning R Function”) does not implement the textual spanning equation listed in equation 1 in the article (p. 3). Equations 2 and 3 are correctly implemented in the function (i.e., the definition of the den and cSP objects). We have updated the textSpan function on GitHub (https://github.com/dustinstoltz/textSpan) and noted the error in the article’s replication repository (https://github.com/dustinstoltz/textual_spanning_socius). For a detailed, step-by-step guide to calculating textual spanning in R, see https://github.com/dustinstoltz/textSpan/blob/master/2020_spanning_step_by_step_guide.pdf.
We reran the simulations and empirical illustration. In general, the bivariate Pearson correlations between the textual spanning scores for the simulated graphs and the topic model network are positive and high, indicating that the guiding theoretical intuition of the measure remains well founded. These correlations are reported in Table A1. Three of the five correlations are high and positive, suggesting that we obtain similar results with the corrected function to those in the article.
Correlations between Textual Spanning Scores Reported in the Article and Textual Spanning Scores with Correct Function.
Note: Bivariate Pearson correlations are reported. All specifications for graph construction are the same as those reported in the original article.
There are two exceptions. First, the two measures on the simulated disconnected ring graph are highly negatively correlated. As we note in the original article, however, textual spanning is currently specified for fully connected text graphs. The second is the core-periphery graph (r = –.350). This suggests that network topology matters when considering the most appropriate measure of novelty, and further research is needed to determine when textual spanning is the appropriate measure. The corrected function also contrasts similarly to the other common centrality metrics on the topic model graph, suggesting that, as argued in the original article, “textual spanning is identifying structural features of the discursive field that standard centrality measures miss” (p. 7).
Socius decided to republish the paper because of the extensive number of edits that are needed to make the necessary revisions. A watermarked version of the original article (published February 8, 2019) is appended to this corrigendum for reference purposes.