Before going further, we provide an example: the first publication of researcher A (a male in the ECON discipline—economics, econometrics, and finance who obtained his PhD in 1995, his habilitation in 2005, and his professorship in 2012) was, according to the Scopus database, published in 2015. The following assumption (based on selected global papers discussed above) is made in this regard: if 2015 is the beginning of an academic career for this scientist in the sense of beginning to publish (or becoming an active member of the academic profession), then his academic age is 0. Therefore, in 2017, he should have a biological age of 32. However, our “Observatory” database using his date of birth indicates 52 years as his biological age in 2017.
Thus, an inconsistency arises: using academic age as a proxy for biological age, scientist A is young and just beginning his scientific career (academic age = 2, career stage: beginning); whereas using biological age, we conclusively find that scientist A is older (52 years old, career stage: middle). Our task below is to estimate this mismatch between academic age and biological age on a large scale of the entire national science system, depending on selected parameters.
Correlation of biological and academic ages
In order to analyze the relationship between biological age (in the range of 30–70 years) on academic age (in the range of 0–40 years), a linear correlation analysis between these variables was performed (Table 3). Regardless of the approach of the correlations presented, a positive relationship was observed in each case. All Pearson’s linear correlation coefficients are significantly different from zero (at a significance level of α = 0.05). Another clearly visible pattern is the usually strong or very strong correlation observed for disciplines belonging to the STEMM cluster. For most STEMM disciplines, a stronger correlation was observed than for all disciplines together (i.e., Total). The scatter plots (Fig. 2) clearly indicate that the vast majority of individuals publishing in STEMM disciplines publish their first article at a relatively young age (the points shift to the right on the X axis). Moreover, the highest correlations were observed for STEMM disciplines such as chemistry (CHEM, r = 0.889), physics and astronomy (PHYS, r = 0.883), mathematics (MATH, r = 0.849) or medicine (MED, r = 0.750).
In contrast, disciplines such as business (BUS), management and accounting, arts and humanities (HUM), psychology (PSYCH), and social science (SOC) (representing the non-STEMM field)—relatively abundantly represented by Polish scholars—are characterized by a relatively low strength of the relationship between biological and academic ages. The value of the correlation coefficient does not exceed 0.5 in any case and remains in the range from 0.354 for humanities to 0.488 for psychology; the scatter analysis (Fig. 2) indicates a clear shift of points to the left on the X axis.
Thus, the results reveal that most representatives of non-STEMM disciplines, despite their relatively advanced biological age, are relatively young in terms of their academic age. Their first publication in the Scopus database—that is, their publication debut in the global scientific arena—clearly begins later than in the case of STEMM scientists.
This observation is consistent with common intuitions and previous surveys (Kwiek, 2015a, 2020), according to which it takes a shorter amount of time for representatives of the sciences in Poland to participate in the global circulation of science than for representatives of the humanities, social sciences, or economics (among other reasons due to the expansion of the private sector between 1990 and 2006 on which much of the energy of non-STEMM scholars had been focused, Kwiek & Szadkowski, 2019).
Smaller differences in correlations were observed for the other independent variables: university type in terms of research intensity (research-intensive IDUB institutions vs. the rest), academic position, and gender. For faculty at research-intensive universities (IDUB), the correlation is higher than for IDUB (r = 0.739 vs. r = 0.670), but for both, the correlation coefficient is close to the correlation for all observations (r = 0.691). Moreover, the correlation is higher for men than that for women (r = 0.697 vs. r = 0.655).
We have not focused in this paper on the year scientists and scholars have been granted their doctoral degrees as a proxy for an academic age for a simple reason: although in our dataset we have the date for every person, this biographical attribute is rarely available on a national scale in other systems. Our idea was to assess a proxy that is widely available through large bibliometric databases like Scopus and therefore our analyses are performed on the date of first publication. However, the results of a linear correlation analysis between the age of earning a PhD and academic age show important differences compared with the above analyses (see Table 5 in the Data Appendix). While correlations for STEMM disciplines are high and correlations for non-STEMM are low, there are large differences between correlations for male and female scientists and correlations for research intensive IDUB institutions and the rest, not observable in above analyses. Somehow surprisingly, the correlation coefficient for female scientists is almost twice as high as the one for male scientists, and the correlation coefficient for research intensive IDUB institutions is twice as high as the one for the rest (r = 0.676 vs. r = 0.375; and r = 0.752 vs. r = 0.371, respectively). These differences may result from possibly higher publication requirements for doctoral degrees in research-intensive institutions and from possibly higher productivity of women in the early years of employment.
Furthermore, the differences in academic positions reveal (Fig. 3) that although the correlation coefficients are significantly lower, the differences in the slope of the regression curve clearly show that scientists and scholars in lower positions are generally academically younger than their colleagues in higher positions (shift of points to the left on the X-axis), which is probably due to the correlation of position with biological age. Simultaneously, as the academic seniority increases, the strength of the correlation clearly decreases (from r = 0.583 for assistant professors to only r = 0.339 for full professors).
These results clearly show the diversity of scientists and scholars in terms of when they begin to publish: current full professors, the oldest and highest in the academic hierarchy, are the latest to begin; and the youngest and at the beginning of their academic career, scientists only with a doctorate (assistant professors), find international publishing more natural. These findings confirm the results of earlier surveys and interviews that reveal radical generational differences in Polish science (Kwiek, 2015b) and survey results linking “internationalists” and “locals” in research to age and academic generations (Kwiek, 2020). Remarkably, an earlier survey research (conducted on 4000 returned questionnaires) is strongly corroborated by the detailed large-sample research presented here.
Intergenerational differences in international publishing are due to distinctly different starting conditions for different groups of researchers: different opportunities for international collaboration and different institutional requirements at successive stages of academic career development, particularly growing after 2010, when two series of structural higher education reforms (2010–2012 and 2016–2018) were initiated. Faculty at the level of assistant professors and, thus, also predominantly young faculty, have been embarking on their academic careers for a decade now, under radically better financial conditions and changed political and social realities than their colleagues who are currently associate or full professors. The inter-cohort differences clearly reveal the evolution of the Polish system of science.
Analyzing the distribution of biological age by individual academic ages for all scientists and scholars—regardless of discipline, institutional type, academic position, and gender—there is an increasing but fading trend of the medians of biological age and a decreasing variation of biological age with increasing academic age (Fig. 4). For gender and institutional type (IDUB vs. rest), no significant differences noted, while disciplines (Fig. 5) and academic position (Fig. 6) have a strong influence on the shape of the biological age distribution.
For the STEMM disciplines, the trend of medians in many cases (e.g., BIO, CHEM, COMP, EARTH, ENG, ENVIR, MATER, MATH, MED, or PHYS) appears not to be fading but largely linear, once again highlighting the strong association between academic age and biological age in the STEMM field. Although academic careers evolve with the age of the scientist, for the non-STEMM field, it appears that the trend does not occur at all in certain cases (particularly in large disciplines such as ECON, HUM, or PSYCH) or is barely noticeable (as in BUS and SOC). Similarly, the variability in the distribution of biological age, which decreases with academic age for STEMM disciplines, appears to be unchanged or changing in a disordered manner. In terms of academic positions, the distribution of biological age for individual years of academic age appears to be similar to the general pattern (i.e., somewhat resembling a logistic curve—fading growth) for assistant professors and associate professors, but behaves rather differently for full professors. Among full professors, there is a clear dominance of scientists and scholars who are older in biological age for almost every year of academic age.
The interquartile ranges (Fig. 7) decrease as academic age increases, and they vary between 11 and 13 years for the low academic ages to 2–3 years for the oldest ages. In other words, variability decreases as academic age increases. While for the younger cohorts the deviation from the median age for the middle 50% of academics is approximately ± 6 years, it is only ± 2 years for the older cohorts.
The conclusions obtained from the correlation analysis are confirmed by the contingency analysis of academic career stages (four stages) with academic position (three positions) and age groups (four age groups). We allocated all scientists and scholars to four career stages or years of academic publishing (academic age brackets): beginning (B, less than five years of academic experience since the first publication in Scopus), early career (E, 5–14 years), middle career (M, 15–29 years), and late career (L, 30 and over). In very approximate terms, if academic age is 0 years (beginning of publication career in the sense of first publication = 30 years), then those in the beginning stage are aged 30–34 years, those in the early career stage are aged 35–44 years, those in the middle career stage are aged 45–59 years, and those in the late career stage are aged 60 years and over. However, in all analyses, we use a well-defined and strictly determined academic age range of 0–40 years for each researcher.
An example of the differences between STEMM and non-STEMM disciplines is very well illustrated in the comparison between CHEM (chemistry) and HUM (arts and humanities) disciplines, both relatively populous (the upper panel in Fig. 8 presents age groups and the lower panel presents academic positions). In chemistry, the vast majority of assistant professors (62.5%) are in the early stage of their careers. On the other hand, a majority of associate professors (65.7%) are in the middle stage, while a majority of full professors (64.3%) are in the late stage. It is evident that the correlation of successive promotions and stages of scientific career with the advancement of own scientific work is understood as a bracket of academic age. Chemistry is an excellent example of a discipline in which academic positions correspond with the publication trajectories of scientists and scholars: full professors mostly publish much longer than other categories of scientists, as is expected. The contingency analyses by academic position (bottom panel) and by age group (top panel) are similar in this case, with the highest percentage of young scientists under the age of 40 (81.2%) being in the early career stage; this is consistent with the idea that younger generations entering the academic profession have been well internationalized in research.
In contrast, this division is not clear-cut in the humanities. We can observe a shift in the academic career stage (defined by the date of first publication) toward a younger age (Fig. 8, upper panel) and earlier academic career stage (Fig. 7, lower panel). Thus, among assistant professors at HUM, two-thirds (65.0%) are at the initial (beginning) stage of their career rather than the early stage, as was expected; but most importantly, the subsequent positions do not indicate a clear distinction among career stages—that is, one cannot identify a stage clearly dominated by associate professors or full professors. Scholars in these positions are mainly in the beginning and early stages of their careers, rather than in middle and late career stages, as suggested in the model (confirmed for CHEM above).
This implies that in the non-STEMM disciplines such as HUM arts and humanities (as in BUS business, management, and accounting; ECON economics, econometrics, and finance; PSYCH psychology; and SOC social sciences), Polish scholars began publishing their articles in the international circulation radically later than those in chemistry, which is compared here, but also in disciplines such as BIO and PHYS. It also implies that in the humanities, it is mainly the youngest scholars and those at the earliest stage of their scientific careers who publish internationally.
Analyzing the contingency between academic position and academic career stages (Fig. 8, lower panel), the evident pattern is that scholars in almost all non-STEMM disciplines are delayed by at least one academic career stage compared to STEMM disciplines. Moreover, in non-STEMM disciplines, associate professors and full professors do not dominate at all in the late stages of their careers, which implies that they publish internationally for shorter periods of time than one might assume.
An analogous pattern is observed for the contingency of age groups and academic careers (Fig. 8, upper panel): the STEMM disciplines clearly show advancement to later academic career groups with increasing age group, while for the non-STEMM disciplines this relationship is not obvious. We observe a significant shift over time for non-STEMM disciplines—the first articles of non-STEMM scholars are published late in their careers, often only in the phase of working as a full professor. This has one implication: the academic age in the case of non-STEMM disciplines does not keep up with the biological age, and full professors are often at the same stage of academic age as assistant professors. Thus, the inference of biological age of scientists in STEMM disciplines proves to be an adequate approximation of reality, while the same inference in non-STEMM disciplines proves to lead to erroneous conclusions and distorted results.
The later date of the first publication for non-STEMM authors can be linked to both external and internal factors. External factors include, first, the differentiated representativeness of Polish research outputs in Scopus by discipline, with low representativeness of outputs in non-STEMM disciplines; and, second, the language coverage of Scopus, with a limited number of Polish-language journals. Weaker representation of non-STEMM journals compared with STEMM journals in Scopus (especially in social sciences, and even more so, in humanities) was often discussed in literature (see Aksnes & Sivertsen, 2019; Singh et al., 2021; Harzing, 2019). And internal factors include, first, historically consistently weaker focus on international publishing in social sciences and humanities, at least until the two waves of higher education reforms in 2010s, compared with strong focus on publishing in national journals (both prior to the collapse of the communist regime in 1989 and in the post-communist period); and, second, weaker focus on publishing in English, compared with strong focus on publishing in Polish. Poland has a shorter history in international academic publishing generally, with less research resources such as funding and infrastructures (see Mongeon & Hus, 2016). The reforms introducing new types of research assessment exercise in 2011 and 2017, with new internationally-oriented individual and institutional publishing requirements, slowly change the publishing practice of scientists and scholars.
Researchers in social sciences and humanities represent high levels of what was termed “multilingual publishing”: a recent study shows that 8.1 percent of them publish only in English (the lowest rate in a sample of 7 European countries); and more importantly, 48.3 percent of them publish only in Polish (the highest rate). In more general terms, 44.1 percent of researchers publish in English and 88.4 percent publish in Polish (Kulczycki et al. 2020: 1375). However, publication patterns in social sciences and humanities in the Central and Eastern European countries are becoming increasingly similar to those in Western Europe and the Nordics, as a recent study of five European countries shows (Petr et al., 2021). As a national case study of Norway highlights, Scopus covers 89 percent of the total Norwegian scientific and scholarly output in medicine and health and 85 percent in natural sciences and technology, compared with merely 48 percent in social sciences and 27 percent in humanities (Aksnes & Sivertsen, 2019: 1). The later date of the first publication in non-STEMM disciplines makes inferring biological age from individual publication histories much less reliable—but the factors are both external (Scopus-related) and internal (Polish publishing patterns).
Modeling approach: linear regression model
The correlation analysis of academic age and biological age with particular independent variables in the disjointed (two-dimensional) approach presented above leads to the observation of interesting patterns, but only the combined (multivariate) impact of all of the variables on biological age enables the provision of a full picture of the studied phenomenon.
In order to conduct multivariate analysis, a linear multivariate regression model was created, where the dependent variable was biological age and the independent variables were (1) academic age, (2) gender, (3) institutional type, (4) academic discipline, and (5) academic position. The resulting model explained 61.5% of the variation in biological age and the standard error was 6.59 years (i.e., when determining an age from the estimated model, we are off the mark by an average of 6.59 years; the relative error was 14%). All interpretations are subject to the ceteris paribus assumption and the significance level used was α = 0.05 (Table 4).
According to the model, if academic age increases by one year, biological age increases by 0.6 years on average. This is also the variable with the strongest influence on the dependent variable (the standardized coefficient is 0.523, which is the largest among all independent variables). The characteristic with the second strongest effect is academic position. The position of full professor has a positive impact compared to the position of assistant professor (reference category; the increase in age is on average 11 years greater); simultaneously, the position of associate professor is also characterized by a high impact (increase on average 5.4 years). This relationship applies only to the studied population of scientists and scholars publishing in the Scopus database, and not all Polish scientists and scholars. Moreover, the predictor associated with working at universities other than the 10 research universities (reference category) positively influences biological age (on average by 1.3 years). Importantly, gender does not significantly affect the prediction of biological age.
A few rather interesting findings in the model come from the analysis of the effect of individual academic disciplines on predicted biological age. Scientists and scholars publishing in BUS, DEC, and ENER have similar biological age as those publishing in HUM (which is a reference category). Only assignment to the HEALTH discipline has a significantly positive effect on biological age (by 2 years on average). Publishing in the vast majority of disciplines has a negative effect on biological age compared to HUM (except for the small HEALTH discipline, the small DEC discipline, and the large BUS and ENER disciplines). Assignment to the PHYS, CHEM, IMMU, BIO, PHARM, MATH, MATER, COMP, and MED disciplines has the strongest negative effect on biological age (4–6 years on average). These disciplines belong to the traditional STEMM field, which clearly indicates an earlier start to the academic career (as measured by academic age). The six other STEMM disciplines (i.e., AGRI, ENVIR, CHEMENG, DENT, ENG, EARTH) also have a negative impact compared to HUM by an average of 2–2.5 years. In contrast, the non-STEMM domains (particularly the largest, such as ECON, PSYCH, SOC) also show a negative impact, but clearly smaller than the others (1–1.5 years). This analysis demonstrates the overwhelming supremacy of STEMM disciplines in internationally visible scholarly production.
The analysis of standardized coefficients reveals that the most important predictor of biological age is academic age, the value for which was as high as 0.523 and it was much higher than the second most influential factor—position (associate professor had a value of 0.234, while full professor 0.357). Therefore, these two characteristics are definitely the strongest determinants of biological age. Among the disciplines, CHEM, BIO, MED, PHYS, MATER, and MATH (0.13, 0.13,0.12, 0.11, 0.11, 0.09 respectively) were characterized by relatively high, although much lower than previously mentioned, values of standardized coefficients, which indicates a strong influence of STEMM disciplines on age. The remaining disciplines were characterized by the value of the standardized coefficient significantly below 0.1. Moreover, IDUB did not turn out to be a strong predictor, with a value of only 0.06. The phenomenon of collinearity (significant correlation of the vector of independent variables) did not occur in the model—the values of VIF coefficients in almost every case were lower than 2 (with 4 as a value allowing to state the occurrence of significant interdependence).
In addition, three models were estimated for the age of obtaining PhD, habilitation, and full professorship using the same independent variables. However, these models had a negligible goodness of fit to the empirical data (R2 ranged from 0.005 for the model for the age of PhD to 0.09 for the age of professorship). In fact, the only variable that showed a significant association with each model was academic age, but the magnitude of the coefficient for this variable did not exceed 0.07, thereby implying that an increase in academic age by one year only marginally affects the increase in age at degree completion. In addition, other characteristics such as gender and field significantly affected the age of degree attainment only for the models for habilitation and professorship. These models are not analyzed in this paper.