Table 5:

Entry and Research Direction

Panel A	Cardinal Measure		Ordinal Measure
	Intllct. Proximate Articles	Intllct. Distant Articles	Intllct. Proximate Articles	Intllct. Distant Articles
After Death	0.091** (0.030)	0.028 (0.035)	0.117** (0.028)	−0.024 (0.037)
Nb. of Investigators	6,228	6,099	6,260	6,017
Nb. of Fields	33,375	32,232	34,218	31,712
Nb. of Field-Year Obs.	1,228,157	1,186,589	1,259,176	1,167,423
Log Likelihood	−1,628,374	−1,816,449	−1,893,982	−1,628,170
Panel B	In-field vs. Out-of-field References		Backward Citations to the Star’sBibliome
	w/ in-field references	w/o in-field references	w/ references to the star	w/o references to the star
After Death	−0.023 (0.041)	0.128** (0.031)	0.078* (0.036)	0.152** (0.034)
Nb. of Investigators	6,195	6,260	6,247	6,259
Nb. of Fields	32,721	34,218	34,179	34,147
Nb. of Field-Year Obs.	1,204,315	1,259,176	1,257,747	1,256,576
Log Likelihood	−792,803	−2,510,350	−1,914,447	−1,767,579
Panel C	Vintage of Cited References		Vintage of 2-way MeSH term combinations
	Young	Old	Young	Old
After Death	0.071* (0.035)	−0.010 (0.034)	0.090** (0.033)	0.029 (0.036)
Nb. of Investigators	6,260	6,260	6,258	6,260
Nb. of Fields	34,218	34,214	34,206	34,210
Nb. of Field-Year Obs.	1,259,176	1,259,044	1,258,732	1,258,906
Log Likelihood	−2,124,598	−1,613,454	−1,853,064	−1,784,279

Note: Estimates stem from conditional (subfield) fixed effects Poisson specifications. In Panel A, the dependent variable is the total number of publications by non-collaborators in a subfield in a particular year, where these publications can either be proximate in intellectual space to the star’s source publication, or more distant (in the PMRA sense). Since PMRA generates both a cardinal and an ordinal measure of intellectual proximity, we parse the related articles using both measures, yielding a total of four different specifications. For the cardinal measure, a related article is deemed proximate if its similarity score is above .58, which corresponds to the median of relatedness in the sample. For the ordinal measure, a related article is deemed proximate if its similarity rank is below 90, which also corresponds to the median of similarity in the sample. In Panel B, we focus on whether the content of entrants’ contributions in the subfield change after the superstar passes away. Each cited reference in a related article can either belong to the subfield, or fall outside of it; it can cite a publication of the star scientist associated with the subfield, or fail to cite any of the star’s past contributions. In Panel C, the dependent variable is the total number of publications by non-collaborators in a subfield in a particular year, where these publications can either be “fresh” (citing young references, or being annotated by MeSH terms of recent vintage) or stale (citing old references, or being annotated by MeSH terms of distant vintage). All models incorporate a full suite of year effects and subfield age effects, as well as a term common to both treated and control subfields that switches from zero to one after the death of the star. Exponentiating the coefficients and differencing from one yield numbers interpretable as elasticities. For example, the estimates in the first column of Panel A imply that treated subfields see an increase in the number of PMRA-proximate contributions by non-collaborators after the superstar passes away—a statistically significant 100×(exp[0.091]-1)=9.53%. Robust standard errors in parentheses, clustered at the level of the star scientist.

^†p < 0.10,

^*p < 0.05,

^**p < 0.01.