Bivariate analysis: academic behaviors and attitudes
The first question in this section is whether the working habits of top performers are different from those of the remaining 90 % of research-involved academics. The second question is whether top performers are more research-oriented (both consistent with the research literature on research productivity, see especially Fox 1992; Bentley and Kyvik 2013; Shin and Cummings 2010).
Academic behaviors: working time distribution
We explore here the five dimensions of academic work which were captured by the CAP/EUROAC datasets: teaching, research, service, administration, and “other” academic activities. The mean for the annualized total working time differential between top performers and the rest of academics is 6.2 h, ranging from 2.2 h in Italy to 9.4 h in Norway and 10.2 h in Germany (see the details by type of academic activity and by country in Table 12 in the ESM). In other words, for example, German top performers, when compared with the rest of (research-involved, as in the whole paper) German academics, spend on average an additional 66.3 full working days in academia per year (10.2 h times 52 weeks divided by 8 h per day), and Norwegian top performers spend on average an additional 61.1 full working days.
We know from previous research productivity studies that longer working hours, and especially research hours, substantially contribute to high productivity: Our study shows (with powerful results: p value <0.001) what exactly “longer hours” mean for the upper 10 % of highly productive academics, and shows it from a comparative cross-national perspective. A ticket to enter the class of national top performers differs from country to country, though, as systems are not equally competitive: In more competitive systems, top performers work much longer hours than in less competitive systems when compared to average academics.
We are interested in the differences in the means of total working hours, and especially the means of research hours, between the two subpopulations in each country and the significance of the results (Table 3). Our results are based on two-sided tests assuming equal differences in arithmetic means with a significance level α = 0.05. For each pair with a mean difference significantly different from zero, the symbol of the larger category (“Top” for top performers or “Rest” for the rest of academics) appears in the column. Tests are adjusted for all pairwise comparisons within a row for each innermost subtable using the Bonferroni correction. The t tests for the equality of two arithmetic means (Top vs. Rest) were performed for each country for each of the five types of academic activities studied.
As clearly shown in Table 3, longer research hours for top performers are statistically significant for a pool of seven countries (“Top” symbols in the line of “research,” the only exception being Switzerland). But also for a pool of seven countries, longer administration hours for top performers are statistically significant (“Top” symbols in the line of “administration”). The same applies to service hours (three countries) and hours spent on “other” academic activities (four countries). Not surprisingly, the same also applies to total working hours in all the countries studied. In three countries (Austria, Norway, and Switzerland), their longer teaching hours are also statistically significant. Further details are given in Table 12 in the ESM: In the column “group with significantly larger mean,” top performers appear in respect of almost all countries, that is, they work longer hours in almost all the categories studied. There is a standard working pattern for top performers in most of the countries studied. “Science takes time,” and much more scientific production takes much more time. Top performers work (much) longer hours. Their longer total working time is statistically significant for all countries.
Academic attitudes: teaching and research orientation
The results of the z test for the equality of fractions performed for all countries (Table 4) are based on two-sided tests with a significance level of α = 0.05. Tests are adjusted for all pairwise comparisons within a row for each innermost subtable using the Bonferroni correction.
The z tests for the equality of fractions (Top vs. Rest) were performed for each country for each of the four categories of teaching and research orientation. Correspondingly, as before, for each pair with a fraction difference significantly different from zero, the symbol for the larger category (“Top” for research top performers or “Rest” for the rest of academics) appears in the column.
As clearly shown in Table 4, the research role orientation (answer 3) among top performers is statistically significant in a pool of eight countries (“Top” symbols in the line for “in both, but leaning toward research”, with no exceptions). Additionally, in a pool of five countries, the strong research role orientation (answer 4) for top performers is also statistically significant, again with no exceptions. The division in role orientation between top performers and the rest of academics is clear (and all differences are statistically significant): In all the systems studied, top performers are more research-oriented than the rest of academics (see the column “group with a larger mean” and answers 3 and 4 in Table 13 in the ESM). Being interested “primarily in teaching” virtually excludes such European academics from the class of research top performers: Their share attains a maximum of 2 % in Ireland but in the majority of them it is 0 %. In addition, being interested “in both, but leaning toward teaching” again almost excludes such European academics from the same class: Their share is about 3 % in the United Kingdom and 5–9 % in the other countries with only two exceptions: Poland (17.4 %) and Portugal (21.7 %) where it is substantially higher (both are teaching-oriented systems, though, Kwiek 2012, 2013a). Our results show that a research role orientation is a powerful indicator of belonging to the class of the European research elite: Being research-oriented is virtually a must for European academics and being teaching-oriented virtually excludes them from this class.
However, a study of multidimensional relationships requires a model approach with a number of dependent variables, including research hours and research orientation, among several others.
Logistic regression analysis
We have developed an analytical model to study research productivity based on the research literature, especially Fox (1992: 295–297), Ramsden (1994: 211–212), and Teodorescu (2000: 207). Following Ramsden (1994), we have assumed that “any sensible explanation of research output must take into account personal (individual) and structural (environmental) factors, and preferably also the interaction between them.” Independent variables are grouped as “individual” and “institutional” characteristics in eight clusters (see Table 5).
There are two questions related to the overall research approach taken. The first question is why estimating a regression model for each of the 11 countries rather than pooling the sample and control for country. The argument for the choice of 10 % top performers per country (and per major academic field cluster) is that the approach of selecting merely the upper 10 % of academics, regardless of the country, does not fit the purpose of highlighting cross-national differences among top performers. The factors important in predicting high research productivity in some countries might be irrelevant in other countries. However, we have also developed a single model controlling for country fixed effects, and the two models will be compared briefly in the “Discussion” section. The second question is why the regression model is not controlled for academic discipline as a potentially important source of variation: Unfortunately, the number of observations per discipline was too small in many cases.
In this multivariate analysis, we have dichotomized all category variables through a recoding procedure. We started with 42 personal and institutional characteristics, grouped into eight clusters (see Table 10 in the ESM). We then conducted Pearson Rho’s correlation tests to find significantly correlated predictors of the dependent variable. The predictors were entered into a four-stage logistic regression model (as in Cummings and Finkelstein 2012). Multicollinearity was tested using an inverse correlation matrix, and no independent variables strongly correlated with others were found. The predictive power of the fourth model (as measured by Nagelkerke’s R
2) was the highest for Portugal (0.54), the United Kingdom (0.40), Norway, Ireland, Switzerland, and Finland (about 0.30–0.32). The total average variance demonstrated for the 11 countries studied is about 32 %. The predictive power of the models of research productivity estimated by other researchers is not substantially higher (the average variance in Drennan et al. 2013: 129 is about 30 %, and about 30 % for 10 globally studied countries in Teodorescu 2000: 212). A model fit defined via the percentage of cases predicted correctly is generally in the 80–90 % range. In Table 6, we present the results of the final, fourth model.
Statistically significant individual variables
The collection of individual variables emerges as more important than the collection of institutional variables, both in terms of the frequency of occurrence and the size of regression coefficients.
In the first block of individual predictors (“personal/demographics”), being a female academic entered the equation in two countries only: It is a strong predictor of not becoming a top performer in Italy, where the odds ratio value indicates that female academics are about half as likely as male academics to be a top performer, and in the United Kingdom, where they are only about one-third as likely. But in all other countries, being a male academic is not a predictor of becoming a top performer. While the finding for Italy is consistent with the analysis of Italian “star scientists” in Abramo et al. (2009), overall, our findings are clearly different from the findings from linear regression analyses in which being a female academic has traditionally been negatively correlated with research productivity.
While in most single-nation and cross-national studies age is not a statistically significant variable, our model shows that “age” is a powerful predictor of high research performance in four countries. A one-unit increase (i.e., 1 year) in Ireland and Switzerland increases the odds of becoming a top performer by about 3.5 % on average (ceteris paribus) and in Portugal by 10.5 %.
Finally, being a “professor” (or academic seniority) emerged as the single most important variable in the model, with statistical significance in six countries. In four of them (Finland, Germany, Ireland, and Norway), being faculty at senior ranks increases the odds of becoming a top performer more than three times, in the Netherlands slightly less than three times, and in Poland almost twice (see Kwiek and Antonowicz 2015). This finding confirms the conclusions from previous productivity studies, although certainly academics in European higher education are more likely to be promoted to higher ranks if they are highly productive. Productivity affects being a professor and the relationship may be “reciprocal” (Teodorescu 2000: 214). But as Ramsden (1994: 223) argued, “identifying correlates of high productivity does not mean that we have identified causal relations.”
In the block of individual predictors, “socialization,” to great surprise, especially in the context of the US literature, both variables are either statistically insignificant or, as in two countries (Poland and Italy), they actually decrease the odds of becoming top performers. It could be that in an “academic oligarchy” types of systems (Kwiek 2013b, 2015a), doctoral students receive faculty guidance more by working for senior faculty, possibly as a cheap academic labor force, rather than independently working with them (and later productivity is substantially influenced by the early recognition of research work). The block of “internationalization and collaboration” emerges as the single most important grouping in predicting high research productivity: Each of the four variables at least doubles the odds of becoming a top performer. The four variables are as follows: “collaborating internationally,” “collaborating domestically,” “publishing in a foreign country,” and “research international in scope or orientation.” These variables enter the equation in all countries except one (Finland).
Domestic collaboration, as opposed to international collaboration, does not influence high research productivity in any country except for the United Kingdom. “Publishing in a foreign country” emerged as a powerful predictor in four smaller higher education systems: Ireland, Poland, Switzerland, and Norway, as with small academic markets it makes it more necessary for prolific academics to publish internationally. Also, “research international in scope or orientation” increases the odds in three countries. The atypical case of Germany where this variable actually decreases by half the odds of being a top performer could be explained by the large size of the national publishing academic market.
In the block of “academic behaviors,” contrary to previous research conclusions from linear regression models (most recently in Cummings and Finkelstein 2012: 58; Shin and Cummings 2010: 590; Drennan et al. 2013: 127), annualized mean weekly research hours emerged as determinative predictors only in three countries (Germany, Norway, and the United Kingdom): A unit increase of 1 h (in annualized research hours per week) increases the odds of being a top performer by a 2.6–3.7 % on average (ceteris paribus). In all the other countries, a high research time investment is not a determinative predictor of becoming a top performer.
Again, in the block of “academic attitudes and role orientation,” contrary to the findings from previous linear regression models, research orientation emerged as a powerful predictor of research productivity in only two countries, with Exp(B) = 3.141 for Ireland and Exp(B) = 1.51 for Poland. In all other countries, it was not a determinative predictor.
Surprisingly, while in simple descriptive statistics (both here and in Postiglione and Jisun 2013) and in inferential analyses presented above, both long research hours and high research orientation emerge as important characteristics of top performers, following the almost universal findings in the research productivity literature, here, a multidimensional model approach supports these findings in selected countries only.
Statistically significant institutional variables
The importance of variables differs from country to country, but the overall determinative power of individual-level predictors is much stronger than those of institutional-level predictors, consistent with previous research on productivity (Ramsden 1994: 220; Shin and Cummings 2010: 588; Teodorescu 2000: 212; Cummings and Finkelstein 2012: 59). As Drennan et al. (2013: 128) concluded, “institutional factors were found to have very little impact on research productivity.” This finding is also consistent with the conclusion about the American professoriate that “intrinsic motivations” rather than “institutional incentive structures” (Finkelstein 1984: 97–98, Teodorescu 2000: 217) stimulate research productivity. In general, the institutional-level predictors are statistically significant in only two cases in two countries (Switzerland and the United Kingdom). Surprisingly in the context of previous research (Fox 1983), two institutional predictors are not statistically significant in any of the countries studied: “availability of research funds” and “supportive attitude of administration.” This might mean that, generally, neither institutional policies nor institutional support substantially matter in becoming a top performer.
Interestingly, while the conclusions from linear regression models indicate that institutional-level predictors of research productivity are weak, in our logistic regression model the conclusions indicate that they are actually statistically insignificant. In particular, research funds and academic climate (good academic–administration relationships) do not enter the equations in any country in the model (on collegiality across Europe, see Kwiek 2015a). Also, the strong performance orientation of institutions is statistically insignificant in all countries except Switzerland.