In this section, we investigate the factorsFootnote 6 that determine modelers’ promotion and time to promotion. We will first assess whether early signals of job market attractiveness (i.e., what was known at the time of first hire) are associated with future promotion. Next, we will use all known information on modelers to predict promotion and time to promotion. Finally, we will discuss the value of different research portfolios in different hiring departments and discuss how membership in coauthorship networks and prestige substantially improve the probability of promotion.
Our approach to analysis is as follows. We model the probability of promotion using logistic regression analysis and time to promotion using survival analysis. For the latter analysis, we employ a proportional hazard Cox regression  in which non-censored observations are those modelers who were promoted and right-censored observations are those modelers who were not. Thus, our failure variable is promotion, and positive estimates or hazard ratios imply an increase in “risk” of being promoted and should be judged as positive factors towards promotion. Tied uncensored observations are resolved using the Breslow method .
Does Early Job Market Attractiveness Predict Promotion and Time to Promotion?
An implicit assumption in the entry-level job market is that attractive job candidates remain attractive after being hired. Signals of early job market attractiveness include whether the degree-granting department is prestigious (i.e. top 30), as this suggests future research productivity  and exceptional productivity in the leading journals, an important criterion for promotion  that may help hiring departments bolster or solidify their rankings .
We test this implicit assumption by relating promotion and time to promotion to the signals observable at the time of first hire. Thus, the objective of these models is to determine whether signals observed in the entry-level job market can be related to future promotion and time to promotion. Note that, for this reason, information on which hiring department the modeler started at cannot be incorporated into the model as this information is revealed afterwards. Similarly, because a candidate’s social connections may not be developed enough at this stage, we also omit coauthorship community variables from this analysis. Finally, because only one top tier 4 publication was observed prior to hiring, we do not divide tier 4 publications into “top” and “other” for this analysis. Results are shown in Table 5.
The first set of results in Table 5 shows the influence of modelers’ early job market signals on promotion.Footnote 7 For each model, two sets of estimates are shown. The first set includes regression coefficients, while the second set includes odds ratios (for promotion models) or hazard ratios (for time to promotion models). These ratios measure the increase in odds of being promoted when the variable in question takes the value of 1, as compared to the case where it takes the value of 0 , and can be interpreted as the number of times the modeler with the variable with the value of 1 is more likely to be promoted as compared to the modeler with the value of 0, everything else constant. Standard errors are shown in parentheses.
In general, we find that early job market attractiveness is not a very strong predictor of future promotion or time to promotion. Indeed, the predictive powerFootnote 8 of both models is quite low, with the promotion model (model 1) exhibiting an R
2 of 9.49 % and the time to promotion model (model 2) yielding an R
2 of 11.17 %. As to the factors that influence our dependent variables, we find that the average number of authors in modelers’ publications before graduation, and being male, increase the probability of promotion. Of particular interest are critical factors such as early research productivity in tier 1 journals and top 30 status. Despite being important determinants of early job market attractiveness and hiring department utility in the marketing job market , we find that these factors do not influence the probability that a modeler may achieve promotion after being hired. However, we find that tier 1 publications do influence time to promotion, albeit marginally so.
Promotion and Time to Promotion Analysis Using Full Information
We now assess the determinants of promotion and time to promotion with all available information. For promotion, we assess the probability of a modeler being promoted at the employment level. This means that each data point in our dataset represents a particular hiring department where each modeler worked—a unique scholar-hiring department combination. Because a number of modelers in the promotion dataset worked in more than one department, the number of observations increases from 128 to 204.
An interesting case to consider when assembling data at the employment level is when a modeler moves from department A (without promotion) to department B (with promotion). For each of these “promotion movements,” two data points are included in our promotion dataset. The first data point includes the modeler’s research productivity at department A and the characteristics of such department, with the dependent variable indicating no promotion; the second data point includes the same research productivity as in the case above, and the characteristics of department B, with the dependent variable indicating promotion.Footnote 9
Research productivity when considering data at the employment level is measured with the number of articles published when employed at each particular department.Footnote 10 Dummy variables are used to indicate the type of hiring department modelers were employed in. In both specifications, the coauthorship community variables are also included as dummy variables. Therefore, the estimates of the community membership dummy variables should be interpreted as the effect of belonging to these communities on promotion and time to promotion with respect to modelers who belong to other, smaller communities, or those who do not belong to any.
For the analysis of time to promotion, we retain the original formulation used in model 2. That is, we analyze the data at the candidate level and use 128 observations. The reason to continue using this original formulation is that the inclusion of “promotion moves” can lead to an important confound when examining time to promotion. Specifically, a modeler may have moved from one department to another because of a promotion. Whereas in the promotion model, one can classify the first department as a “no promotion” and the second as a promotion, then keep the modeler’s publication portfolio equal at both departments; for time to promotion, it is unclear what time value to assign to the second move. Results of the promotion (model 3) and time to promotion (model 4) analyses with full information are also shown in Table 5.
Regarding the probability of promotion, we find that working in a department within the USA, as opposed to working overseas, implies that modelers are 79 % less likely to be promoted. This means that, conversely, modelers have a much higher likelihood of promotion in international departments. In addition, we find evidence (albeit at the 10 % significance level) that modelers are 2.08 times more likely to be promoted if they are employed in a private hiring department as compared to a public hiring department, on average. As to research productivity, we find evidence that tier 1, tier 2, and tier 4 publications increase promotion probability. Candidates with a tier 1 publication are twice as likely to be promoted than those who do not have one, and candidates with a tier 2 publication are 61 % more likely to be promoted. Interestingly, we find the effect of top tier 4 publications to be quite substantial. To be specific, candidates with a top tier 4 publication are 2.88 times more likely to be promoted as candidates who do not have such a publication. However, care must be exercised when interpreting this estimate, as the number of candidates with top tier 4 publications is relatively low.
As to the impact of coauthorship, we find that membership in one of the communities found using the Louvain community detection algorithm has a significant (at the 10 % level) effect on promotion: modelers who belong to coauthorship community 4 are 2.29 times more likely to be promoted as compared to modelers that do not belong to this community. Finally, when comparing the full-information promotion model (model 3) to the early job market promotion model (model 1), we observe an increase in predictive power, obtaining an R
2 of 31.07 %. This implies that as modelers’ academic career unfolds, valuable, additional signals that can aid predicting the probability of promotion emerge.
Regarding time to promotion, we find that few variables are statistically significant, as in our analysis of early job market attractiveness. Furthermore, predictive power remains quite poor, with an R
2 of 15.76 %. However, we find evidence (at the 10 % significance level) that tier 1 publications again accelerate time to promotion. For each tier 1 publication, the likelihood of promotion increases by 10 %. Furthermore, male modelers, on average, are promoted faster, implying a 62 % increase in probability of promotion. Finally, we find that belonging to coauthorship community 4 influences time to promotion as well, such that, similar to promotion, membership in this community almost doubles the chance of being promoted.
The Value of Research Portfolios at Different Hiring Departments
A modeler’s publication portfolio may be valued differently for promotion at different hiring departments. If so, modelers may need to adjust their research portfolio or target journals as they move from one type of department to another. Thus, we next investigate the impact of modelers’ publication record on the probability of promotion at departments that offer a Ph.D. program as compared to those who do not. Because of the low incidence of top tier 4 publications from modelers in hiring departments without a Ph.D. program, we do not divide tier 4 publications into top and other for this analysis. Results are shown in Table 6.
We find that the effect of modelers’ publication portfolios on promotion is substantially different at departments who have a Ph.D. program, as compared to others. For departments with a Ph.D. program, the qualitative nature of the results in Table 5 holds, although the magnitudes are different. Importantly, tier 4 publications now are observed to exceed the value of tier 1 publications. Notice that, counterintuitively, for departments without a Ph.D. program, only tier 1 publications influence promotion, this at the 10 % significance level. Finally, notice that the effect of coauthorship community 4 is also present (again, at the 10 % significance level), which suggests that being a member of this coauthorship social network is a robust predictor of promotion.
Predicting Modelers’ Probability of Promotion
The analysis shown so far is valuable in that it isolates the main factors associated with modelers’ promotion outcomes. However, it is also important to predict these outcomes given modelers’ characteristics as well as those of the departments they graduated from and those they work in. As such, we present a brief predictive analysis of promotion scenarios by focusing on the role of research productivity, prestige, and coauthorship on modelers’ predicted probability of promotion. Note that given the low predictive power of the time to promotion models shown in Table 5, we focus on predicting the probability of promotion only.
Consider a modeler who does not come from a top 30 degree-granting department, is unconnected with the four major coauthorship communities discussed earlier, and has no publications in his portfolio. Furthermore, given the characteristics of the average modeler, assume this modeler is male, from a private degree-granting department, and was an international student (not from the US). We call this modeler the “benchmark” modeler. Given the results shown in Tables 5 and 6, a counterfactual modeler with more publications, prestige, or member of a coauthorship network should be expected to be more likely to be promoted than the benchmark modeler.
Utilizing the regression results from model 5, we estimated modelers’ probability of promotion at a Ph.D. granting, private university, which is a common first place of employment among modelers. We find that the predicted promotion probability for the benchmark modeler is 13.24 %. If the benchmark modeler, instead, studied at a top 30 degree-granting department, the predicted probability of promotion increases to 19.65 %. If, in addition, the modeler were also a member of coauthorship community 4 (which we call a “prestigious and connected” modeler), the predicted probability of promotion exhibits a further increase to 38.10 %. This highlights the fact that, in hiring departments with a Ph.D. granting department, the advantages determined by modelers’ degree-granting department prestige and embeddedness into particular coauthorship networks can potentially increase the predicted probability of promotion by more than 20 %.
We also investigated the impact of a high research productivity on the probability of promotion. To be specific, we deem a high research productivity to be four published tier 1 articles, as the average across all promoted modelers was 3.45 tier 1 publications, as shown in Table 1. With such a publication portfolio, the probability of promotion for a benchmark modeler is 75.41 %; for a prestigious modeler, it is 83.09 %; and for a prestigious and connected modeler, it is 92.51 %. These results highlight that the advantages enjoyed by modelers from prestigious degree-granting departments or who are well connected, persist even when their peers exhibit the same level of research productivity.