1http://dx.doi.org/10.20396/bjos.v19i0.8656977 Volume 19 2020 e206977 Original Article 1 Department of Preventive and Social Dentistry, Federal University of Rio Grande do Sul, Porto Alegre, RS, Brazil. 2 Federal University of Pelotas, Pelotas, RS, Brazil. 3 Department of Stomatology, Federal University of de Santa Maria (UFSM), Santa Maria, RS, Brazil. 4 Dental Materials Laboratory, School of Dentistry, Federal University of Rio Grande do Sul, Porto Alegre, RS, Brazil. 5 Department of Periodontology, Federal University of Rio Grande do Sul, Porto Alegre, RS, Brazil. Corresponding author: Roger Keller Celeste Department of Preventive and Social Dentistry, Federal University of Rio Grande do Sul, Porto Alegre, RS, Brazil. Email: roger.keller@ufrgs.br Received: October 06, 2019 Accepted: December 02, 2019 Brazilian dentistry research productivity: state level socioeconomic, educational and structural factors Roger Keller Celeste1,* , Francisco Wilker Mustafa Gomes Muniz2, Thiago Machado Ardenghi3, Fabrício Mezzomo Collares4, Cassiano Kuchenbecker Rösing5 Aim: To explore socioeconomic, educational and research factors associated with dental research productivity at the state level in Brazil. Methods: The authors used the Scopus database to identify dental articles published from 2006 to 2016 associated with Brazilian universities at the state level. Several social, economic, educational and research structure variables were obtained from the census and National Research Council to predict the rate of articles per 100 thousand inhabitants among the 27 Brazilian states. Rates were fitted in linear weighted least-squared regression with stepwise technique. Twenty-two variables were grouped in six blocks (social, economic, general education, dental education, research workforce and structure). Results: A total of 21189 articles were published, and the state of São Paulo accounted for 46%, followed by Rio Grande do Sul with 9.4%; four states did not publish any articles. There were an average (± standard deviation) of 2.6 (±1.98) published articles per 100 researchers and 13.4 (±9.6) articles per 100 thousand inhabitants. Research structure and workforce explained 92.4% and 87.2% of state variability, respectively, while the final model explained 94.5%. One extra PhD and one extra undergraduate researcher per 100 thousand inhabitants were associated with 11.3 more and 3.5 fewer articles, respectively, while every 10 points (range 0-100) on the Human Development Index (Education Component) was associated with 3.3 more articles. Conclusion: State scientific output has several associated factors, but research workforce and general education variables seem to be good predictors. Large disparities among state research outputs have been described and must be addressed by research and development policies. Keywords: Bibliometrics. Dentistry. Research. Science. https://orcid.org/0000-0002-2468-6655 2 Celeste et al. Introduction Sustainable development goals in relation to educational processes highlight the necessity of expansion in higher education and scientific programmes, especially in less developed countries1. One of the reasons is that investment in science and tech- nology has long been accepted as a way to generate knowledge and a cornerstone of social and economic development2. Regarding investment in research and devel- opment (R&D), Brazil, an upper middle-income country, has fallen behind the average for upper income countries, with only 1.2% of gross domestic product (GDP) expen- ditures and 698 full-time researchers per million inhabitants3. On the other hand, esti- mates related to dental research in Brazil have been recognized as outstanding, and current Brazilian research productivity is higher than many high-income countries4-7 as it has a strong increment since late 1990’s8. Research productivity as an output of investments has been assessed based on the number of scientific articles. The main determinants have been studied at individual (researcher) and institutional/organizational levels9. For example, male researchers of higher rank and those awarded large research grants were reported to have higher productivity in the USA10, although younger researchers supervising graduate stu- dents tend to publish more in Saudi Arabia11. Also, higher education institutions (HEI) have pivotal roles, as they account for 23.7% of R&D expenditure and 64.3% of all scientific publications2. An important fraction of research from HEI comes from grad- uate programs; hence, it is not surprising that the ratio of graduate students to staff has been described as an important factor for departmental productivity12. Few stud- ies have assessed the impact of undergraduate students in these figures; however results show some effect on faculty productivity13. On a macro-level, GDP has also been associated with performance14 and productivity, and the Human Development Index (HDI) an important predictor of country rate of article publication7. Although it could be hypothesised that socioeconomic factors are associated with research productivity in Brazil, it is unclear if this association will remain after controlling for direct factors such as the rate of researchers per inhabitants. In addition, no study has described this phenomenon in the Brazilian context, which will be important in explaining state differences. In the mid-1990s, Brazil implemented an evaluation system for graduate study pro- grams with a strong emphasis on scientific productivity in high-impact journals. This policy was proposed from a national perspective with few incentives for state-level research agencies, with some exceptions. Variability among Brazilian states has not been described or explained to the best of our knowledge. However, analysing such variability will foster equitable development and should be pursued, as the capacity to produce contextualized knowledge at local levels is key for sustainable develop- ment. In addition, taking the continental size of Brazil into consideration, understand- ing local/state conditions may shed some light on possible contextual predictors affecting scientific productivity elsewhere. Therefore, the objectives of this study are to explore socioeconomic, educational and specific research factors associated with state-level output for Brazilian dental research. 3 Celeste et al. Materials and methods This is an ecological study in which the units of observation were all 27 Brazilian states. The number of published articles (dependent variable) was obtained from the Scopus database, and the other 20 potential predictors were obtained from several sources. Scopus was chosen because it is used by the Brazilian Higher Education Assessment Council (CAPES) to assess institutional proposals. It also allows the identification of authors’ institutional addresses, making it possible to count the number of articles per state. Outcome Variable The dependent variable was the rate of articles per 100 thousand inhabitants at the state level. This was established by dividing the total number of articles from each state in an 11-year-period (2006-2016) by their population according to the 2010 cen- sus. The number of publications was retrieved by combining a search strategy for dental articles with an additional search for universities with undergraduate programs in dentistry (using authors’ address identification filters at Scopus). The search strat- egy used to identify dental articles was obtained from a previous publication5 and is available only in the original publication. The search to identify universities was based on the names of 219 undergraduate programs available on the website of the Fed- eral Dental Council (CFO). Universities with more than one program in the same state counted as one, because such differences are not distinguished by Scopus. Independent Variables Seven social and economic variables were obtained from the Instituto Brasileiro de Geografia e Estatísica and Atlas de Desenvolvimento Humano do Brasil15. They are based on the 2010 census data: Gini coefficient of income inequality, proportion of individuals with inadequate sanitation, proportion of individuals living in urban areas, HDI, mean per capita income, proportion of individuals living in poverty and GDP per capita. Those variables are considered distal contextual factors that may indirectly affect research productivity. We selected five variables that represent the general and dental educational con- text at the state level. The mean number of years of education at the age of 18 and the education component of the HDI were obtained from Atlas de Desenvolvimento Humano do Brasil. The rate of dental schools with undergraduate programs per mil- lion inhabitants was calculated using the number of programs available from the CFO website. Two other variables were obtained from official government data (Instituto Nacional de Estudos e Pesquisas Educacionais Anísio Teixeira [INEP]) from http:// inep.gov.br/enade: mean score on the national board for final-year dental students (years: 2007, 2010, 2013) and percentage of members of dental school teaching staff holding a PhD. Finally, we collated data to calculate eight variables concerning research structure and workforce. Based on information from the Research-Groups Census (2010) for differ- ent types of researchers at the National Research Council (CNPq) website (http://dgp. cnpq.br/planotabular/index.jsp), we created eight rates per state level inhabitants: rate http://inep.gov.br/enade http://inep.gov.br/enade http://dgp.cnpq.br/planotabular/index.jsp http://dgp.cnpq.br/planotabular/index.jsp 4 Celeste et al. of PhD researchers, rate of PhD students, rate of undergraduate students in research, rate of research groups and research lines. Data from the Research-Groups Census (2010) are provided by group leaders and certified by their institutions. Data about the number of graduate programs in dentistry (master and PhD levels) for each state was obtained from the CAPES website (https://sucupira.capes.gov.br/sucupira/) and overall, there were 101 active graduate programs. Statistical Analysis Descriptive data on the overall profiles of articles were presented by institution, journal and country of co-authors. Bivariate analyses were presented with categorized covari- ates, and differences were tested using the Kruskal-Wallis non-parametric rank test. Categorization was necessary for descriptive purposes, and states were grouped in tertiles in the case of a gradient or median/specific cut-off point if the bivariate rela- tion was non-linear. The rate of publication was modelled using linear regression with ordinary weighted least squares by population size16. All 20 variables were grouped in 6 blocks (see Table 2) and modelled using a within-block stepwise forward technique, taking p < 0.20 as a variable to enter into the model. Those significant variables within each block were transferred to the full model, also modelled with a stepwise forward technique with p < 0.10 to enter into the model. This two-step approach was needed due to the high degree of collinearity among variables; the final model was evaluated based on R-squared fit index, variance inflator factor (VIF), homoscedasticity (Cook-Weisberg test for heteroscedasticity) and normality of residuals (Shapiro-Wilk test). In multiple linear regression, variables were entered into the model as continuous and non-cate- gorized. All analyses were carried out using Stata 13.1. Results In an 11-year period, Brazilian dental researchers published 21189 articles indexed in the Scopus database. The state of São Paulo accounted for 46% of all articles (n = 11767), followed by Rio Grande do Sul with 9.4% (n = 2395). Four states (Acre, Roraima, Amapá and Rondônia) did not publish any articles. Overall, during the study period (2006-2016), there were an average (± standard deviation) of 2.6 (±1.98) pub- lished articles per 100 researchers and 13.4 (±9.6) articles per 100 thousand inhabi- tants. The state with the highest rate was São Paulo, with 28.5 articles per 100 thou- sand inhabitants, followed by Rio Grande do Sul (22.4), Rio Grande do Norte (15.0), Paraíba (14.9) and Paraná (14.8). Productivity per 100 researchers was highest in the Federal District with 17.1 articles per 100 researchers, followed by Mato Grosso (6.1), Ceará (3.6), Sergipe (3.3) and five states (Maranhão, Santa Catarina, Goias, Rio de Janeiro and São Paulo) with 3.1. Articles are concentrated in a limited number of journals, institutions and co-autors’ countries. The top 10 journals accounted for 24.5% of all articles published by Bra- zilian researchers (Table 1), and Brazilian journals summed 6 of the top 10. While the whole list reaches almost 1000 journals, about 150 journals published 80% of all papers. The top 10 institutions were associated with 80.4% of all published articles: Universidade de São Paulo published 28.2% (the three campuses cannot be distin- https://sucupira.capes.gov.br/sucupira/ 5 Celeste et al. guished in Scopus), followed by UNESP (the same as Universidade de São Paulo, n = 3) with 14.7% and UNICAMP with 12.8%. The top 10 co-author countries summed Table 1. Percentages of articles published (n=21189) with at least one Brazilian author among journals indexed in Scopus between 2006 and 2016. Publication Institution N % Universidade de Sao Paulo - USP 5966 28.2% Universidade Estadual Paulista – UNESP 3114 14.7% Universidade Estadual de Campinas 2720 12.8% Universidade Federal do Rio de Janeiro 1029 4.9% Universidade Federal de Minas Gerais 969 4.6% Universidade Federal do Rio Grande do Sul 867 4.1% Universidade Federal de Pelotas 656 3.1% Universidade Federal de Santa Catarina 614 2.9% Universidade Federal Fluminense 560 2.6% Universidade Federal de Sao Paulo 546 2.6% Subtotal 17041 80.4% Total 21189 100.00% Country United States 2229 10.5% Italy 492 2.3% United Kingdom 455 2.1% Canada 437 2.1% Germany 266 1.3% Switzerland 255 1.2% Spain 249 1.2% Netherlands 203 1.0% Sweden 169 0.8% France 155 0.7% Subtotal 4910 23.2% Total 21189 100.00% Journals Brazilian Dental Journal 870 4.1% Journal Of Applied Oral Science 709 3.3% Dental Press Journal Of Orthodontics 625 2.9% Journal Of Endodontics 617 2.9% Brazilian Oral Research 569 2.7% Pesquisa Brasileira Em Odontopediatria E Clinica Integrada 401 1.9% Brazilian Journal Of Oral Sciences 373 1.8% American Journal Of Orthodontics And Dentofacial Orthopedics 346 1.6% Journal Of Periodontology 344 1.6% Operative Dentistry 344 1.6% Subtotal 5198 24.5% Total 21189 100.00% 6 Celeste et al. Table 2. Mean rate of articles per 100 thousand inhabitant in the period of 2006-2016 among Brazilian states (n=27). Mean Std. Dev. n p-value* Social Indicators Gini Coefficient Lower level 14.47 6.93 5 0.15 Middle level 16.92 10.35 12 Higher level 5.12 2.94 10 Proportion of people with inadequate sanitation up to 10% 15.99 9.58 15 0.03 11% or more 5.54 3.83 12 Urbanization Lower half 5.49 4.23 13 0.07 Upper half 16.69 9.35 14 IDH Lower half 5.79 3.84 13 0.48 Upper half 17.47 9.31 14 Economic Indicators Mean individual income Lower half 5.81 3.82 13 0.51 Upper half 17.44 9.34 14 Proportion of poverty up to 10% 17.81 9.16 11 0.02 11% or more 5.60 3.87 16 GDP per capita Lower tertile 5.29 3.19 9 0.10 Middle tertitle 8.64 4.05 9 Highest tertile 20.03 9.02 9 General Education Mean number of year of education Lower tertile 7.14 4.52 9 0.18 Middle tertitle 8.34 4.52 9 Highest tertile 19.88 9.95 9 HDI education component Lower tertile 5.06 3.75 9 0.01 Middle tertitle 5.99 4.02 9 Highest tertile 18.49 8.84 9 Dental Education % of undergrad lectures with PhD Lower level 5.15 4.27 9 <0.01 Middle level 8.21 4.21 9 Higher level 20.18 9.28 9 Rate of dental schools per 1 million inhabit <=1 school 5.67 3.09 9 0.76 >1 school 16.93 9.48 18 Mean ENAD score 2007-2013 <=2 points 2.40 2.20 5 <0.01 >2 points 13.85 9.52 22 Research structure Graduate programs per million inhabit Lower tertile 6.24 5.06 9 <0.01 Middle tertitle 4.92 1.45 9 Highest tertile 19.34 8.45 9 Research groups per 100 thousand inhabit Lower tertile 4.39 3.70 9 <0.01 Middle tertitle 6.88 3.79 9 Highest tertile 19.35 9.05 9 Research lines per 100 thousand inhabit Lower tertile 4.65 3.78 9 <0.01 Middle tertitle 7.65 3.92 9 Highest tertile 21.45 8.53 9 Research workforce PhD Researchers per 100 thousand inhabit Lower tertile 4.35 3.51 9 <0.01 Middle tertitle 6.28 3.15 9 Highest tertile 20.03 8.23 9 Graduate Student researchers per 100 thousand inhabit Lower tertile 4.10 3.84 9 <0.01 Middle tertitle 4.75 1.96 9 Highest tertile 18.61 8.44 9 Undergraduate researchers per 100 thousand inhabit Upper half 4.48 2.79 13 <0.01 Lower half 16.15 9.34 14 % of researcher >50 year-old Upper half 7.53 4.53 11 0.76 Lower half 16.16 10.07 12 % of male researchers Upper half 7.71 4.45 11 0.80 Lower half 16.14 10.15 12 Total 13.40 9.57 27 * Kruskal-Wallis ranking test 7 Celeste et al. 23.2% out of 133 countries; the USA accounted for 10.5%, followed by Italy with 2.3% and the UK/Canada with 2.1% each. In bivariate analysis, several variables were significantly related to state productiv- ity and showed high degrees of correlation (Table 2 and Table 3). Nonetheless, in the final regression model, only three variables explained 94.5% of state variability (Table 4): the educational component of the HDI, rate of undergraduate students in research and rate of PhD researchers. Every 10 points in the educational compo- nent of HDI was associated with 3.3 more articles per 100 thousand inhabitants (95% confidence interval – 1.0: 5.5), while every additional PhD researcher per 100 thousand inhabitants was associated with 11.3 more articles (95% confidence inter- val – 8.8: 13.8), and one additional undergraduate researcher was associated with 3.5 fewer articles (95% confidence interval – 6.2: -0.7). In the final model (Table 4), no variables were heteroscedastic; the highest VIF was associated with the rate of PhDs (VIF = 4.9). Graphic analysis of residuals showed that they were normally dis- tributed (Shapiro-Wilk test, p = 0.06). Discussion State scientific output has several associated social and economic factors, but three seem to be good predictors: rate of PhD researchers, rate of undergraduate students (involved in research) and general education level (HDI-education). In addition, our findings showed striking state disparities in total research output. Taking the size of Brazil into consideration, as this mirror other large disparities in social, economic and cultural aspects. The use of such associations is of interest to understand which fac- tors can predict better or worse research productivity rates. Only three variables remained in the final model, and the most influential was the rate of PhD researchers, confirming a previous study12, followed by the educational component of HDI and rate of undergraduate students involved in research. The rate of PhD researchers was highly correlated to other variables and may have affected some of them. For example, graduate programs educate PhDs and may be indirectly responsible for their scientific output. Furthermore, the presence of a graduate pro- gram is an interesting indicator of PhD students, research grants and other resources, such as laboratory infrastructure. States with lower levels of competitiveness may fall behind and try to offset with more undergraduate researchers than expected. In contrast to our results, another study showed that undergraduate students may increase overall productivity13. Nonetheless, such papers may be published in jour- nals not indexed by Scopus and thus did not appear in our work. To our knowledge, this is the first study to include social and educational indicators, with HDI-education showing a statistically significant effect. We speculate that it may have a direct effect on research productivity by boosting critical thinking in lower education, but it is also likely to be a general marker of social development and investments in education at basic and higher levels. State disparities were found in total research output, with São Paulo having 46% of all papers, as the University of São Paulo (USP) accounts for 28% of the whole country. USP’s superiority over other Brazilian institutions has also been confirmed in previ- 8 Celeste et al. Ta bl e 3. S pe ar m an c or re la tio n co effi ci en t m at rix in cl ud in g al l v ar ia bl es a m on g B ra zi lia n st at es (n =2 7) . V 1 V 2 V 3 V 4 V 5 V 6 V 7 V 8 V 9 V 10 V 11 V 12 V 13 V 14 V 15 V 16 V 17 V 18 V 19 V 20 V 21 V 22 V 1 A rt ic le R at e pe r 1 00 th ou sa nd /h ab 1 V 2 G in i -0 .3 7 1 V 3 In ad eq ua te S an ita tio n -0 .6 2 0. 66 1 V 4 U rb an iz at io n 0. 40 -0 .5 1 -0 .7 0 1 V 5 H D I 0. 42 -0 .6 3 -0 .8 3 0. 89 1 V 6 % o f P ov er ty -0 .4 8 0. 72 0. 86 -0 .8 3 -0 .9 6 1 V 7 M ea n in di vi du al in co m e 0. 40 -0 .6 1 -0 .8 1 0. 87 0. 98 -0 .9 7 1 V 8 G N P p er c ap ita 0. 38 -0 .5 4 -0 .7 7 0. 83 0. 95 -0 .9 2 0. 97 1 V 9 H D I-E du ca tio n 0. 42 -0 .6 3 -0 .8 1 0. 87 0. 98 -0 .9 2 0. 93 0. 89 1 V 10 M ea n Ye ar s of Ed uc at io n 0. 46 -0 .7 1 -0 .7 2 0. 62 0. 69 -0 .7 0 0. 61 0. 52 0. 73 1 V 11 M ea n EN A D s co re 0. 39 -0 .2 1 -0 .2 1 -0 .0 8 -0 .0 4 -0 .0 4 -0 .0 8 -0 .0 6 -0 .0 3 0. 24 1 V 12 R at e of D en ta l S ch oo ls -0 .2 2 -0 .0 9 -0 .1 4 0. 33 0. 39 -0 .3 4 0. 41 0. 40 0. 32 0. 19 -0 .5 9 1 V 13 % o f P hD in U nd er gr ad pr og ra m s 0. 67 -0 .0 4 -0 .3 1 0. 30 0. 20 -0 .2 4 0. 18 0. 15 0. 19 0. 30 0. 54 -0 .3 8 1 V 14 R at e of G ra du at e pr og ra m s 0. 72 -0 .3 5 -0 .3 8 0. 22 0. 12 -0 .2 0 0. 15 0. 16 0. 11 0. 25 0. 38 -0 .3 0 0. 56 1 V 15 R at e of R es ea rc h Li ne s 0. 82 -0 .2 6 -0 .4 2 0. 17 0. 15 -0 .2 1 0. 16 0. 19 0. 11 0. 23 0. 40 -0 .2 3 0. 52 0. 84 1 V 16 R at e of R es ea rc h G ro up s 0. 79 -0 .2 9 -0 .4 8 0. 27 0. 21 -0 .2 6 0. 21 0. 22 0. 15 0. 29 0. 40 -0 .1 8 0. 50 0. 77 0. 93 1 V 17 R at e of R es ea rc he rs (t ot al ) 0. 82 -0 .2 6 -0 .4 2 0. 17 0. 15 -0 .2 1 0. 16 0. 19 0. 11 0. 23 0. 40 -0 .2 3 0. 52 0. 84 1. 00 0. 93 1 V 18 R at e of P hD R es ea rc he rs 0. 86 -0 .3 9 -0 .5 1 0. 27 0. 24 -0 .3 4 0. 26 0. 26 0. 22 0. 30 0. 54 -0 .3 5 0. 66 0. 89 0. 94 0. 87 0. 94 1 V 19 R at e of U nd er gr ad ua te R es ea rc he rs 0. 76 -0 .1 5 -0 .3 2 0. 14 0. 01 -0 .1 1 0. 04 0. 02 -0 .0 4 0. 15 0. 46 -0 .3 1 0. 57 0. 77 0. 85 0. 90 0. 85 0. 84 1 V 20 R at e of P hD S tu de nt s 0. 85 -0 .4 7 -0 .5 6 0. 36 0. 34 -0 .4 0 0. 32 0. 29 0. 32 0. 41 0. 53 -0 .3 3 0. 61 0. 77 0. 87 0. 87 0. 87 0. 92 0. 84 1 V 21 % o f M al e R es ea rc he rs -0 .1 0 -0 .3 6 -0 .1 0 0. 09 -0 .0 1 -0 .0 1 -0 .0 2 0. 01 -0 .0 6 0. 07 -0 .0 4 -0 .1 0 -0 .1 6 0. 31 0. 03 0. 19 0. 03 0. 06 0. 15 0. 04 1 V 22 % o f S en io r R es ea rc he rs 0. 00 -0 .1 5 -0 .2 0 0. 24 0. 08 -0 .1 3 0. 12 0. 15 0. 06 -0 .0 3 0. 26 -0 .3 1 0. 20 0. 33 0. 24 0. 31 0. 24 0. 41 0. 30 0. 34 0. 39 1 N O TE : i n sh ad e co effi ci en ts o f v ar ia bl es re la te d to s am e th eo re tic al b lo ck 9 Celeste et al. Ta bl e 4. C oe ffi ci en ts fr om li ne ar re gr es si on m od el s of a rt ic le s pe r 1 00 0 00 /i nh ab ita nt s am on g B ra zi lia n St at es (n =2 7) , 2 00 6- 20 16 . B lo ck V ar ia bl e W ith in b lo ck s te pw is e* re gr es si on Fi na l M od el s te pw is e* re gr es si on co effi ci en t (I C 95 % ) ad ju st ed R 2 co effi ci en t (I C 95 % ) ad ju st ed R 2 So ci al In di ca to rs H D I* * (e ve ry 1 0 po in ts in cr ea se ) 14 .2 (9 .5 : 19 .0 ) 58 .6 % 94 .5 % Ec on om ic In di ca to rs M ea n in di vi du al in co m e (e ve ry R $1 00 0 in cr ea se ) 25 .8 (1 6. 4 : 3 5. 1) 54 .4 % G en er al E du ca tio n in di ca to rs H D I* * ed uc at io n co m po ne nt (e ve ry 1 0 po in ts in cr ea se ) 11 .9 (8 .2 : 15 .5 ) 62 .4 % 3. 3 (1 .0 : 5. 5) D en ta l E du ca tio n In di ca to rs % o f u nd er gr ad le ct ur es w ith P hD (e ve ry 1 0 pe rc en t p oi nt s in cr ea se ) 7. 1 (5 .1 - 9. 1) 76 .8 % R at e of d en ta l s ch oo ls p er 1 m ill io n in ha bi t ( >1 s ch oo l) 9. 2 (5 .2 : 13 .2 ) R es ea rc h st ru ct ur e in di ca to rs R es ea rc h gr ou ps p er 1 00 th ou sa nd in ha bi t ( ev er y on e m or e) 35 .0 (2 0. 4 - 4 9. 6) 87 .2 % G ra du at e pr og ra m s pe r 1 m ill io n in ha bi t ( ev er y on e m or e) 8. 4 (1 .1 : 15 .8 ) R es ea rc h w or kf or ce P hD R es ea rc he rs p er 1 00 th ou sa nd in ha bi t ( ev er y on e m or e) 14 .2 (1 2. 1 : 1 6. 2) 92 .4 % 11 .3 (8 .8 : 13 .8 ) U nd er gr ad ua te re se ar ch er s pe r 1 00 th ou sa nd in ha bi t ( ev er y on e m or e) -5 .7 (- 8. 5 : - 2. 9) -3 .5 (- 6. 2 : - 0. 7) * fo rw ar d st ep w is e re gr es si on w ith e nt ry v al ue p <0 .1 0 (o rd in ar y w ei gh te d le as t- sq ua re s by p op ul at io n si ze ) ** H D I= H um an D ev el op m en t I nd ex v ar ie s fr om 0 to 1 00 . 10 Celeste et al. ous studies5,17, and it has been estimated to contribute more than 20% in all research areas18. Indeed, São Paulo is the only state in Brazil where the dentistry field is the most productive in all research fields in Brazil18; therefore, state differences are likely to be larger in dentistry than other areas. A similar concentration of publications in a few places has been reported in the African continent, where Nigeria and South Africa account for over two-thirds of all oral health-related research19. On one hand, São Paulo has the highest percentage of investments in R&D regarding GDP20, the São Paulo Research Foundation (FAPESP) plays an important traditional role21. On the other hand, there seems to be a trend to decentralize researchers, relocating them to other areas of Brazil from São Paulo22. Although the role of national R&D agencies in compensating regional disparities is not clear from our study, the concentration of graduate programmes in the southeast declined from 73% to 51% between 1980 and 2010 as part of the CAPES policy23. That policy increased the number of graduate pro- grams in other regions, decreasing the share of programs in already developed areas. One important aspect is that the regulatory system of evaluating higher education in Brazil likely triggers of the development of scientific communication and dissem- ination; the field of dentistry is an example. Our analysis shows a steep increase in publications in high-impact journals (data not shown). Over the period observed, the increase was not uniform countrywide; social, economic and cultural variables prob- ably accounted for the differences. The increase in funding in R&D must also explain part of the increase in productivity in the last decades. The present study confirms the virtuous cycle of investment and development output. In addition, the association between the increase in PhDs among teaching staff in dental education and better research output should be highlighted. The results encountered herein should encom- pass the increase in dental programs in states where there were few or none. A limitation of this study is the use of university names as surrogates for states. There may be other institutions contributing scientific output that were not included, although we have no reason to think our conclusions would be different in that case, as very few papers would be lost. A second point concerns the quality of the data, a common issue in ecological studies, as validity and reliability are usually lower when information is not designed for scientific purposes. Data from CNPq and other sources are administrative in nature with some degree of measurement error. None- theless, we believe that such measurement errors are likely to be random and do not invalidate our findings. Another limitation is that our study cannot identify inter-state institutional collaboration; this would require a different approach to find all authors’ addresses. The strengths of this work include the large geographical coverage and a search strategy with good sensitivity and specificity5. The generalization of our results is limited to Brazilian scenarios but may hold true for other countries with similar contexts. In conclusion, this study demonstrates the role of important factors associated with dental research productivity at the state level in Brazil. The rate of researchers, the most influential variable, is likely to be a consequence of other structural determinants of research productivity. State disparities were found not only in total output but also in per capita productivity. This research may assist agencies and researchers to better understand macro-determinants of scientific research and foster future policies. 11 Celeste et al. Acknowledgements RKC, TMA, FMC and CKR hold CNPq PQ Felowship. References 1. United Nations. The Sustainable Development Goals Report. United Nations; 2016. 2. Pastor Monsálvez JM, Serrano L. The determinants of the research output of universities: specialization, quality and inefficiencies. Scientometrics. 2016 Nov 20;109(2):1255-81. doi: 10.1007/s11192-016-2102-3. 3. World Bank. Databank: World Development Indicators 2016 [Internet]. Whashington: International Bank for Reconstruction and Development/The World Bank; 2016 [cited 2019 Jun 10]. Available from: http://databank.worldbank.org. 4. Soares CJ, Bönecker MJS, Dias KRHC. [Area Document: Area 18 – Dentistry]. Brasilia: Coordination for the Improvement of Higher Education Personnel; 2016. p. 49. Portuguese. 5. Celeste RK, Broadbent JM, Moyses SJ. Half-century of Dental Public Health research: bibliometric analysis of world scientific trends. Community Dent Oral Epidemiol. 2016 Dec;44(6):557-563. doi: 10.1111/cdoe.12249. 6. Nadanovsky P. Growth in Brazilian scientific output in public health dentistry. Cad Saude Publica. 2006 May;22(5):887, 886. 7. Allareddy V, Allareddy V, Rampa S, Nalliah RP, Elangovan S. Global Dental Research Productivity and Its Association With Human Development, Gross National Income, and Political Stability. J Evid Based Dent Pract. 2015 Sep;15(3):90-6. doi: 10.1016/j.jebdp.2015.01.004. 8. Pulgar R, Jiménez-Fernández I, Jiménez-Contreras E, Torres-Salinas D, Lucena-Martín C. Trends in World Dental Research: An overview of the last three decades using the Web of Science. Clin Oral Investig. 2013 Sep;17(7):1773-83. doi: 10.1007/s00784-012-0862-6. 9. Daizen T. Research productivity. In: Teichler U, Arimoto A, Cummings WK, editors. The Changing Academic Profession in Japan. New York: Springer; 2015. p. 149–67. 10. Lee S, Bozeman B. The impact of research collaboration on scientific productivity. Soc Stud Sci. 2005;35:673-702. doi: 10.1177/0306312705052359. 11. Alghanim SA, Alhamali RM. Research productivity among faculty members at medical and health schools in Saudi Arabia: Prevalence, obstacles, and associated factors. Saudi Med J. 2011 Dec;32(12):1297-303. 12. Dundar H, Lewis DR. Determinants of research productivity in Higher Education. Res High Educ. 1998;39(6):607-31. 13. Morales DX, Grineski SE, Collins TW. Increasing research productivity in undergraduate research experiences: Exploring predictors of collaborative faculty-student publications. CBE Life Sci Educ. 2017 Fall;16(3). pii: ar42. doi: 10.1187/cbe.16-11-0326. 14. King DA. The scientific impact of nations. Nature. 2004 Nov 4;432(7013):8. doi: 10.1038/430311a. 15. United Nations Development Programme Institute for Applied Economic Research João Pinheiro Foundation. Atlas of Human Development in Brazil. 2013 [cited 2019 Aug 10]. Available from: http://atlasbrasil.org.br/2013/en. 16. Morgenstern H. Uses of ecologic analysis in epidemiologic research. Am J Public Health. 1982 Dec;72(12):1336-44. 12 Celeste et al. 17. Celeste RK, Warmling CM. [Brazilian bibliographical output on public oral health in public health and dentistry journals]. Cien Saude Colet. 2014 Jun;19(6):1921-32. doi: 10.1590/1413-81232014196.04932013. Portuguese. 18. Cross D, Thomson S, Sinclair A. Research in Brazil: a report for CAPES (InCites). Clarivate Analytics; 2018. p. 73. 19. Kanoute A, Faye D, Bourgeois D. Current status of oral health research in Africa: an overview. Int Dent J. 2012 Dec;62(6):301-7. doi: 10.1111/j.1875-595X.2012.00123x. 20. Correa RL, Nascimento DE. [State and regional disparities in science, technology and innovation in Brazil]. In: Proceedings of the 1st National Congress of Professional Masters in Public Administration. Curitiba; 2016 [cited 2018 Sep 6]. 12p. Available from: http://www.profiap.org.br/ profiap/eventos/2016/i-congresso-nacional-de-mestrados-profissionais-em-administracao-publica/ anais-do-congresso/40671.pdf. Portuguese. 21. Borges Neto M. [State foundations supporting research and development of science, technology and innovation in Brazil]. Rev USP. 2011;(89):174-89. doi: 10.11606/issn.2316-9036.v0i89p174-189. Portuguese. 22. Avellar SOC. [Internal migration of master and doctoral degree holders in Brazil: some considerations]. Rev Bras Pos-Grad. 2015;11(24):429-57. Portuguese. 23. Avellar SOC. [Spacial mobility of másters and doctors in Brazil: 1975-2010] [thesi]. Institute of Philosophy and Human Sciences, The University of Campinas; 2015.