Is culture related to strong science? An empirical investigation

National culture is among those societal factors which could influence research and innovation activities. In this study, we investigated the associations of two national culture models with citation impact of nations (measured by the proportion of papers belonging to the 10% and 1% most cited papers in the corresponding fields, PPtop 10% and PPtop 1%). Bivariate statistical analyses showed that of six Hofstede’s national culture dimensions (HNCD), three dimensions of power distance, individualism, and uncertainty avoidance had statically significant associations with citation impact of nations. The study also revealed that of two Inglehart-Welzel cultural values (IWCV), the value survival versus selfexpression is statistically significantly related to citation impact indicators. We additionally calculated multiple regression analyses controlling for the possible effects of confounding factors including international migrant stock, investments in research and development, number of researchers, international co-authorships, and national self-citations. The results revealed that the statistically significant associations of HNCD with citation impact indicators disappeared. But the statistically significant relationship between survivals versus self-expression values and citation impact indicators remained stable even after controlling for the confounding variables. Thus, the freedom of expression and trust in society might contribute to better scholarly communication systems, higher level of international collaborations, and further quality research.

National culture is among those societal factors which could influence research and innovation activities (Couto & Vieira, 2004;Jones & Davis, 2000;Yair, 2020) and productivity (Kedia, Keller, & Jullan, 1992). We found only three studies which focused on the correlation between national culture and citation impact, although citation impact is one of the most important metrics to measure the research performance of individuals, groups of researchers, institutions or world nations. For example, the study by Abraham (2020) showed that individualist and low-hierarchy nations are more likely to have higher citation impact comparing with collectivist nations in which the hierarchical societal systems are prevalent. The three studies provided some insights into the relationship of national culture and citation impact, but are concerned by several limitations. For example, the citation impact of nations may be resultant to many factors such as allocated expenditures to research and development, number of researchers of a country, and the level of national self-citations (Waltman, 2016). These factors should be controlled in the statistical analyses of the relationship between impact and culture. Other limitations refer to the absence of robustness tests and the use of inappropriate indicators for measuring citation impact.
In this study, therefore, we applied two separate cultural models and the relevant datasets, that is, Hofstede's national culture dimensions (HNCD) and Inglehart-Welzel cultural values (IWCV). We linked the models with advanced citation impact indicators. We estimated the associations of national culture dimensions with citation impact of nations after controlling for country-level data on openness of nations (international migrant stock), academic openness (international coauthorships), national self-citations, number of researchers, and investments in research and development.

Literature overview
In our search of the literature, we found only a few studies that have focused on the correlation between cultural values and national performance. These studies have been published in recent years.
Andrijauskienė and Dumčiuvienė (2017) studied the associations of national culture (based on HNCD) and the nations' innovativeness (based on data extracted from the Summary Innovation Index -a sub-index of the European Innovation Scoreboard) for 27 EU member states. The authors showed that of the HNCD, the dimensions of indulgence and individualism were positively while power distance and uncertainty avoidance were negatively correlated with national innovation performance. Further regression analyses revealed that just two dimensions of power distance and indulgence were statistically significant predictors of nations' innovativeness.
The study by Allik, Lauk, and Realo (2020) showed that after controlling for variables of wealth, i.e. research and development expenditures, economic inequality, population, and the worldwide governance indicator (WGI), the "good governance" (measured by WGI) was a strong predictor of the scientific impact of nations. The WGI included six factors of responsibility and effectivity of government, degree of violence, the ability of the government to formulate and implement sound policies and regulations, the authority and influence of law in society, and the corruption level of nations. Allik et al. (2020) measured the scientific impact of nations by three indicators: 1) the nations' average citation rate per paper, 2) the proportion of papers of a nation within the top 1% most cited publications, and 3) the number of research categories in which a country is listed in the Essential Science Indicators (ESI). Abraham (2020) analyzed HNCD data and SCImago Journal & Country Rank data for 60 countries and showed that of six dimensions of HNCD, three dimensions of power distance, individualism, and indulgence had statistically significant correlations with the research impact of nations (measured by the number of published documents, citable documents, citations, self-citations, h index values, and citations per document for countries). After controlling for the wealth of countries, Abraham (2020) found that uncertainty avoidance was (positively) correlated with the research impact of nations, too.
While the three studies provide some evidence for the relationships of national culture dimensions/values and citation impact, the results are not inclusive. The studies are concerned by several limitations which we tried to avoid in this study. First, we used two different cultural approaches (HNCD and IWCV) to investigate the relationship of national culture and citation impact.
We wanted to test the robustness of the results. Second, we went beyond bivariate analyses and used advanced regression models to analyze the data. Third, we used two advanced citation impact indicators, i.e. the proportions of papers belonging to the 10% and 1% most frequently cited papers in the corresponding subject category and publication year (PPtop 10% and PPtop 1%). Fourth, we undertook parallel statistical analyses (i.e. correlation analysis, linear regression analysis, and fractional logistics regression analysis) to investigate the stability of associations between national culture dimensions/values and citation impact. Fifth, we controlled for the effects of many possible confounding variables: international migrant stock, investments in research and development, number of researchers, international co-authorships, and national self-citations.

Cultural datasets
This research used two nation-level cultural datasets to investigate the relationships of national culture and the citation impact of nations. The first (Hofstede, Hofstede, & Minkov, 2010) perceives national culture as a stable construct which does not change in short term periods (or it changes in very long time periods), and the second (Inglehart, 1997) assumes the dynamics of national culture during time. Recent research (Beugelsdijk, Maseland, & Van Hoorn, 2015;Beugelsdijk & Welzel, 2018) showed that both approaches are complementary. Although young generations seem to become more individualistic (which supports the dynamic approach), one can assume that nearly half of national culture values are stable.
Hofstede's national culture dimensions (HNCD) Hofstede et al. (2010) categorized the world countries into six bi-polar dimensions of national culture: individualism versus collectivism, small versus large power distance, low versus high uncertainty avoidance, indulgence versus restraint, low versus high uncertainty avoidance, masculinity versus femininity, and short-versus long-term orientations. As Hofstede (2011) explains, the individualism-collectivism dimension is the extent of group orientations among people in a country. In individualist nations, the people are supposed to take care of themselves and their close family members, while in collectivist nations, the people are bounded by in-group loyalties.
The power distance shows the degree of acceptance of power inequality among the individuals in lower classes and it is manifested in interactions of people in family (parent-children), school (teacher-students), working place (manager-employee), and country (authorities-citizens). In power distant nations (those which have large power distance), the education system is hierarchical and injective. That means the educators are regarded as unquestioned authorities (Hofstede, 2011). The uncertainty avoidance refers to the capacity of individuals in a country to recover quickly from vague or ambiguous situations. The masculinity-femininity shows the degree in which nations have overlapped or separated gender roles and values. The perceptions of people towards the time and their following behaviors shape the time orientation dimension. Finally, the indulgence versus restraint dimension refers to the degree of freeness or regulation of gratification in a country. Inglehart andWelzel (1984-2015)  According to Inglehart and Welzel (2005, p. 5), national culture is a dynamic phenomenon and subject to change, and the socioeconomic development plays a major role in the change process. Inglehart and Welzel (2005, p. 5) claimed that industrial and postindustrial modernization episodes change the cultural values in a certain way: a move from conventional to secular-rational beliefs during the industrial revolution, and a move from survival to self-expression beliefs during the postindustrial period which contributed to increased independence from authorities. This thinking in episodes is different from Hofstede's approach which assumes national culture as a stable construct.

Citation impact data
The citation impact data were exported from the Max Planck Society's in-house database which is based on the Web of Science (Clarivate Analytics). The data included the national proportions of papers (articles and reviews published in 2006 and 2010) belonging to the top 1% (PPtop 1%) and top 10% (PPtop 10%). PPtop 10% is the proportion of papers (published by a country) which belong to the 10% most frequently cited papers within their subject categories and publication years (correspondingly, PPtop 1% refers to the 1% most frequently cited papers) (Hicks, Wouters, Waltman, De Rijcke, & Rafols, 2015). We used fractional instead of full counting of papers: the national numbers of (highly cited) papers were weighted by the number of countries on a paper. For example, if there were three countries mentioned on a paper, each country received 1/3 of the paper.
The IWCV dataset was available for the year 2006, and the last update of HNCD dataset was for the year 2010. To align the citation impact data with the cultural datasets, years 2006 and 2010 were selected as reference points in this study. For citation impact indicators, only those nations which had published ≥500 articles and reviews are included in the analyses. The alignment of the two cultural datasets (IWCV and HNCD) with the citation impact indicators resulted in different samples in the reported statistical analyses of this study, that is, 36 for the IWCD dataset, and 58 for the HNCD dataset.

Confounding variables
In order to investigate the association of cultural dimensions/values and citation impact of nations in this study, we controlled for the possible effects of several confounding variables. These variables have been revealed as (possible) influencing factors of citations (on the national level) in the scientometric literature.
Previous studies have shown that the quantity (i.e. the number of publications) (Lee & Bozeman, 2005) and the quality (received citations) of academic contributions of researchers (Gazni & Didegah, 2011) are positively associated with co-authorships. For example, Li, Liao, and Yen (2013) analyzed the academic productions of Information Systems scholars (extracted from the Social Science Citation Index database, Clarivate Analytics). The authors showed that those scholars who were connectors in co-authorship networks were more likely to produce publications with higher citation counts. This co-authorship effect is also visible on the aggregated national level.
In a recent study, Wagner, Whetsell, Baas, and Jonkers (2018) studied the role of national openness (international co-authorships and mobility of scholars) in the context of national performance. The researchers operationalized performance as fractionally field-weighted citation impact (FWCI) of nations. Colledge (2014) defines the FWCI as "the ratio of citations received (by a nation) relative to the expected world average for the subject field, publication type, and publication year". Controlling for national investments in research and development in the statistical analyses, Wagner et al. (2018) showed that academic openness of nations was a strong predictor of national citation impact.
National investments in research and development have been reported as a predictor of the quantity of publications of nations (Baker et al., 2015;Wagner & Jonkers, 2017;Wagner et al., 2018).
However, the evidence on the effects of those investments on the quality (citation impact) of publications is limited. Recent studies (Wagner & Jonkers, 2017;Wagner et al., 2018) have shown that the national investments in research and development are not a statistically significant predictor of citation impact of nations, but the openness of nations (measured by international coauthorships and free mobility of scholars). In other words, the international "engagements" of nations attract the attention to their research system (Wagner & Jonkers, 2017).
Some studies indicated that institutional and national self-citations could bias the assessment of institutions and nations by using citation impact indicators. For instance, Aksnes (2003) analyzed 45,000 publications (articles, notes, reviews, and proceeding papers) of Norway (for the years 1981-1996) and revealed that self-citations on the author level constituted 36% of all received citations.
This finding was confirmed in a later study with a larger sample (Fowler & Aksnes, 2007). Glänzel, Thijs, and Schlemmer (2004) revealed that self-citations were prevalent for 50 studied countries (varied from 53.14% for the Ukraine to 22.10% for the United States). Highly developed nations such as the United States, Canada, and the United Kingdom had the lowest self-citation rate.
Based on the results of these previous studies, we considered several variables in the current study that might confound the relationship of cultural values and citation impact such as national selfcitations (McGarty, 2000), national openness (such as openness to migrants) (Wagner & Jonkers, 2017), national academic openness (such as global scholarly collaborations and international coauthorships) (Glänzel, 2001;Li et al., 2013;Wagner et al., 2018), and the size of the research sector in each nation (i.e. the number of researchers in research and development).
Aligning the existing data of confounding variables with the cultural and citation impact data reduced the final samples to 28 (IWCV dataset) and 44 (HNCD dataset) nations. The applied datasets, their descriptions, and original sources are summarized in Table 1.

Statistics
This study used several statistical methods to explore the relationships of cultural dimensions (as independent variables) with citation impact indicators (as dependent variables) by controlling for five confounding variables: international migrant stock, investments in research and development, number of researchers, international co-authorships, and national self-citations. Since the dependent variables of this study (i.e. citation impact) are proportions (see above), we computed fractional logistic regression models using STATA 16.0. The fractional logistic regression (using the fracreg command) suits for a dependent variable (such as rates, proportions, and fractional data) that has values in the range of 0≤ values ≤1. Various models (probit, logit, or heteroskedastic probit) can be computed for the conditional mean (STATA, n.d.) which usually lead to similar results. In this study, we computed the logit model because of its relatively simple interpretation (using the fracreg logit command).
In order to investigate the stability of our results based on fractional logistic regression models, we used IBM SPSS (version 25) to perform linear regression analyses and Pearson's correlation analyses.
The results of the stability analyses are part of the supporting information that can be found at https://dataverse.no/dataverse/inn.

Results
The findings of the fractional regression analyses are presented in this section. The findings show the associations of national culture (measured by two separate datasets of HNCD and IWCV) and two citation impact indicators (PPtop 1% and PPtop 10%) as dependent variables, before and after controlling for confounding variables of international migrant stock, investments in research and development, number of researchers, international co-authorships, and national self-citations. Table 2 shows the results of the fractional logistic regression analysis for investigating the direct effects of HNCD on citation impact measured by PPtop 10%. As it is presented in Table 2 (Model 1), two dimensions of Power distance and Uncertainty Avoidance have statistically significant effects on PPtop 10%. The effects of those dimensions of HNCD on PPtop 10% are as follows. A 1 percentage point increase in the power distance score of nations is associated with a 2.9 percentage points decline in PPtop 10%; and a 1 percentage point increase in uncertainty avoidance score of nations is associated with a 3.2 percentage points decline in PPtop 10%. After controlling for the possible effects of the confounding variables (international migrant stock, investments in research and development, number of researchers, international co-authorships, and national self-citations) none of the HNCD dimensions had a statistically significant effect on PPtop 10% (see Table 2, Model 2).   .0003358

Constant
After controlling for the possible effects of confounding variables (international migrant stock, investments in research and development, number of researchers, international co-authorships, and national self-citations) ( Table 9, Section A and B), the statistically significant effect of selfexpression values on PPtop 1% remained constant as follows: a 1 percentage point increase in selfexpression values was associated with a .23 percentage points increase in PPtop 1%.

▪ Summary of the results
The summary of the statistical analyses is depicted in Table 4. This table also includes results of the linear regression analyses and Pearson's correlation analyses from the supporting information document. As the results in the table reveal, the associations of the cultural dimensions/values with citation impact vary depending on the variables included and methods used. Thus, the only inclusive result of these statistical analyses is that just one cultural dimension of survival versus self-expression values seems to be a predictor of national citation impact. The finding was stable (robust) before and after controlling for several confounding variables (Table 4).
To better understand the characteristics of this robust relationship, we drew the scatterplots of survival versus self-expression values and PPtop 10% and PPtop 1% (see Figure 1 and Figure 2). 1 As it is shown in Figure 1

Discussion
The purpose of this study was twofold: 1) to investigate the associations of national culture values/dimensions with citation impact, and 2) to see if the links of cultural values/dimensions with citation impact is stable after controlling for confounding variables of international migrant stock, investments in research and development, number of researchers, international co-authorships, and national self-citations. The results showed that after controlling for the effects of the confounding factors, the associations of HNCD with citation impact disappeared. Only one national culture dimension/value was a stable predictor of citation impact: survival versus self-expression values. The self-expressed nations were more likely to have higher citation impact than the survival nations (and vice versa). This finding was stable before and after controlling for confounding variables based on various statistical analyses.
The stable associations of survival versus self-expression values with citation impact are in accordance with previous research. Inglehart and Welzel (2005) revealed that the "sense of existential security" (i.e. the religion becomes less central when the people reach economic security) and "individual agency" (i.e. people acts on his/her own behalf) are mostly prevalent in nations with secular-rational values and self-expression values. Welzel and Inglehart (2010) further showed that emancipative values -a building block of self-expression values (defined as liberty and equal opportunities for all citizens) -contribute to the general human empowerment process. The possibility to exercise freedom (i.e. democracy) is a strong component of human empowerment (Welzel, Inglehart, & Kligemann, 2003). In democratic societies, there is higher trust among citizens (Inglehart & Oyserman, 2004), and this trust could serve as a positive accelerator of academic productivity (Chalker & Loosemore, 2016;Dwivedi, 1985), innovativeness (Bavec, 2007), international research collaborations (Bagshaw, Lepp, & Zorn, 2007), and the quality of relationships (Wong & Sohal, 2002). One could expect, therefore, that citizens have more trust (in general) in selfexpressed nations which could result in high quality relationships, eagerness to participate in group projects and diverse team groups, and higher quality publications (measured in terms of citation impact).
Although Wagner et al. (2018) showed that academic openness has a strong relationship with national citation impact, it is possible that academic openness is not the main predictor of citation impact. It might act as a mediator in this process, and other national/societal dimensions such as national cultural values affect the degree of openness of nations, and accordingly, citation impact. Allik et al. (2020) showed that "good governance" plays an important role in scientific impact of nations. Good governance (i.e. responsible government, low violence in society, governance of laws, transparent systems, and low corruption) is the main characteristic of self-expressed nations, too.
For example, Inglehart and Oyserman (2004) showed that the low level of trust and tolerance was more common in nations with survival values in which the economic and physical securities are on central focus, while societies with self-express values were characterized by the freedom of expression and the personal choice.
Trust is a stimuli for collaborations and innovations of individuals and nations (Doh & Acs, 2010). A low level of trust is associated with a higher level of uncertainty (Möllering, 2005) or uncertainty intolerance (Frederiksen, 2014) of individuals, and a high level of uncertainty is correlated with low productivity (Lazear & Spletzer, 2012). This uncertainty might negatively affect the motives for scholarly communications, too. Furthermore, other research has shown that when the uncertainty level in a country is high, the people are more likely to "develop low cost strategies to manage their worries and concerns" (Alaszewski & Coxon, 2009). Thus, one could speculate that researchers in countries with a low level of trust (among community members) might have pragmatic approaches focusing on success on the first try rather than targeting top journals in relevant fields to publish their papers. Focusing on targeting top journals comes with higher degree of uncertainty for researchers regarding peer review results.
The empirical research described in this paper has several limitations that should be considered in the interpretation of the results: (1) In this study, we used nations as 'units of analysis' and explored the links of two cultural dimensions/values with citation impact. We controlled for the effects of five most relevant confounding factors of international migrant stock, investments in research and development, number of researchers of nations, proportions of co-authored papers of a nation with international researchers, and national self-citations. However, the citation impact of nations might be resultant to (many) other possible factors that are not considered here.
(2) Because of their wide disciplinary applicability, we used two nation-level cultural frameworks and related datasets (HNCD and IWCV) in this study. However, there are general limitations of such dimensionalized cultural values. As Taras, Rowney, and Steel (2009, p. 362) emphasized, culture is a "complex" construct. While the number of quantitative cultural dimensions have been increasing from four dimensions in early studies of Hofstede to eighteen dimensions in later (GLOBE) studies, "it is still too early to claim that every aspect of culture is captured by any single model or even by all existing models taken together". In order to replicate the findings of this study, or to find the links of other dimensions of culture with research impact, we suggest that future research uses other nation-level cultural datasets such as the seven Schwartz cultural value orientation scores (Schwartz, 2008). These scores are freely available for 80 countries. Another alternative is GLOBE's cultural dataset (House, Hanges, Javidan, Dorfman, & Gupta, 2004) which includes data for 66 nations.
(3) The citation impact indicators of this study are based on datasets extracted from the Web of Science database, which mostly focuses on English publications. Furthermore, only articles and reviews have been considered (but not books and other document types). The results of this study on the national level might change without these limitations concerning language and document type.

Supporting information
The supporting information for this study is freely available from the Inland Norway University (INN) Open Research Data, see https://dataverse.no/privateurl.xhtml?token=b9d8d455-667e-4473-922c-51083477172a