In most countries, cabinets – a handful of individuals wielding overall responsibility for governing a state – are the nucleus of political power. Their actions can significantly influence the lives of millions. Yet, systematized information on cabinet members across extensive periods of time and world regions remains limited.
We present a new dataset, ‘Paths to Power’ (PtP), featuring systematized information on the social profiles of cabinets and governing elites in 141 countries. These include all countries, globally, with populations exceeding 10 million; all OECD and EU countries; and several other smaller countries. We collect information – whenever identifiable – on 44,789 cabinet members holding office across 1966-2021, regarding their level, type, and location of education, place of birth, family- and class origins, prior occupation, and political background. These are widely considered important variables for a host of research questions pertaining to the study of politicians (Krcmaric, Nelson and Roberts Reference Krcmaric, Nelson and Roberts2020), and thus politics, more generally.
Research on political elites was central to political science until the 1980s, but dwindled towards the turn of the millennium (Ricart-Huguet Reference Ricart-Huguet2019). In recent years, we have seen a resurgence of studies focusing on political elites, especially quantitative comparative scholarship, with a growing number of datasets and empirical analyses on the topic. Yet, these datasets and studies all have ‘blind spots’ in the sense that they are limited in their geographic or temporal coverage, the types of governing elites included, or the individual characteristics coded. This is especially the case for countries outside Western Europe and North America, which tend to be understudied in political science (Wilson and Knutsen Reference Wilson and Knutsen2022). This gap presents a special problem, since these understudied countries are, on average, less democratic, with institutions typically providing fewer constraints on governing elites. Existing results based on analysis of governing elites in ‘Western democracies’ may simply not be generalizable. PtP could therefore contribute to filling significant gaps in the study of political elites by enabling social scientists to derive new insights on, for example, global and temporal patterns of representation in political elites or how government members’ backgrounds relate to policies or outcomes such as climate impact, economic inequalities, or armed conflict. In addition to providing a rich and useful data source for scholars in various disciplines, our dataset is of relevance to journalists, policymakers, international organizations, and the wider public.
In the next section, we review existing datasets and highlight PtP’s unique contributions. We detail the variables contained in PtP and the data collection process before we address validity and reliability issues and limitations to the collected data. Then, we give an overview of the composition of political elites worldwide by providing maps showing the distributions on different variables included in PtP. Before concluding, we demonstrate how the data can be used to gain unique insights into six different fields of political science, namely comparative politics, political sociology, gender studies, public administration, international relations, and American and British politics.
Still, the many and varied applications contained in this article are only a limited subset of the potential applications of PtP. We hope that scholars studying various aspects, causes, or consequences of political elite behaviour and representation – across regions and political regime types – will use this new resource to address important social science questions that, until now, have been difficult to answer fully.
Datasets on the Social Profile of Governments
PtP is not the first dataset to register information on the social profiles of political elites. Yet, reflecting a multi-year coding effort involving more than thirty research assistants and researchers across the globe, it is the most comprehensive one in terms of combined geographical, temporal, and variable coverage. Table 1 provides a non-exhaustive summary of existing datasets on cabinet ministers and leaders published since 2010, and compares them with PtP on relevant dimensions.
Table 1. Available datasets on government social profiles published since 2010
Several of the datasets include other types of variables not listed here, such as gender or date of birth. Only datasets on cabinet members or leaders are included.
Many datasets provide rich information for one region or subset of countries, such as Africa (Raleigh and Wigmore-Shepherd Reference Raleigh and Wigmore-Shepherd2022; Ricart-Huguet Reference Ricart-Huguet2021), Western Europe (Alexiadou and Gunaydin Reference Alexiadou and Gunaydin2019), or OECD countries (Alexiadou Reference Alexiadou2015; Hallerberg and Wehner Reference Hallerberg and Wehner2020). Other datasets provide global coverage, but include only a limited number of years or variables (for example, Braun and Raddatz Reference Braun and Raddatz2010; Faccio Reference Faccio2006). The Global Leadership Project has global coverage for 2010-13 and 2017-19 and codes education, occupation, and demographic variables such as language and religion for an impressive number of political leaders in the 2010s (Gerring et al. Reference Gerring, Oncel, Morrison and Pemstein2019; Gerring, Jerzak and Öncel Reference Gerring, Jerzak and Öncel2023). PtP, in comparison, has yearly data from 1966 onward, thus covering several additional decades and allowing for the study of change over a long stretch of time. Furthermore, PtP consistently codes all members of governments for all countries. Goemans, Gleditsch and Chiozza (Reference Goemans, Gleditsch and Chiozza2009), Ellis, Horowitz and Stam (Reference Ellis, Horowitz and Stam2015), Baturo (Reference Baturo2016), and Coppedge et al. (Reference Coppedge, Gerring, Knutsen, Linberg, Teorell, Altman, Angiolillo, Bernhard, Borella, Cornell, Fish, Fox, Gastaldi, Gjerløw, Glynn, God, Grahn, Hicken, Kinzelbach, Krusell, Marquardt, McMann, Mechkova, Medzihorsky, Natsika, Neundorf, Paxton, Permstein, Pernes, Rydén, von Römer, Seim, Sigman, Skaaning, Staton, Sundström, Tzelgov, Wang, Wig, Wilson and Ziblatt2024) have excellent temporal and geographical reach, but only contain information on top leaders, typically heads of state or government (or both). Hence, except for Gerring, Jerzak and Öncel (Reference Gerring, Jerzak and Öncel2023) and Nyrup and Bramwell (2020), even the most extensive dataset on cabinet members or leaders (Coppedge et al. Reference Coppedge, Gerring, Knutsen, Linberg, Teorell, Altman, Angiolillo, Bernhard, Borella, Cornell, Fish, Fox, Gastaldi, Gjerløw, Glynn, God, Grahn, Hicken, Kinzelbach, Krusell, Marquardt, McMann, Mechkova, Medzihorsky, Natsika, Neundorf, Paxton, Permstein, Pernes, Rydén, von Römer, Seim, Sigman, Skaaning, Staton, Sundström, Tzelgov, Wang, Wig, Wilson and Ziblatt2024) incorporates ‘only’ around 7,500 individuals, around one-sixth of the (close to 45,000) ministers included in PtP.
Regarding content, several datasets have very good variable coverage, but these datasets tend to cover specific ministerial positions, legislators, or top leaders. For instance, Bäck et al. (Reference Bäck, Teorell, Von Hagen-Jamar and Quiroz Flores2021) include impressively detailed information on the educational, social, and occupational background of foreign ministers, Baturo (Reference Baturo2016) includes information on the educational, class, and occupational background of political leaders, whereas Ellis, Horowitz and Stam (Reference Ellis, Horowitz and Stam2015) cover leaders’ education and occupational background. Archigos (Goemans, Gleditsch and Chiozza Reference Goemans, Gleditsch and Chiozza2009) and V-Dem (Coppedge et al. Reference Coppedge, Gerring, Knutsen, Linberg, Teorell, Altman, Angiolillo, Bernhard, Borella, Cornell, Fish, Fox, Gastaldi, Gjerløw, Glynn, God, Grahn, Hicken, Kinzelbach, Krusell, Marquardt, McMann, Mechkova, Medzihorsky, Natsika, Neundorf, Paxton, Permstein, Pernes, Rydén, von Römer, Seim, Sigman, Skaaning, Staton, Sundström, Tzelgov, Wang, Wig, Wilson and Ziblatt2024) contain information on several leader characteristics, albeit not education, occupation, or class origins. The Global Legislator database also contains extensive information on legislators in 97 electoral democracies between 2015-2017 (Carnes et al. Reference Carnes, Ferrer, Golden, Lillywhite, Lupu and NazrullaevaForthcoming), but has no information on cabinet members. Finally, WhoGov (Nyrup and Bramwell Reference Nyrup and Bramwell2020) codes all cabinet members and has extensive geographical and temporal coverage, but only covers more basic types of information, such as party affiliations and ministerial posts.
The above-cited data collection efforts have been essential in allowing for the systematized study of political elites across countries.Footnote 1 Yet, they all have certain limitations, whether geographic or temporal, types of elites included, or individual characteristics coded. Indeed, in a recent survey of the literature on politicians’ social background, Carnes and Lupu highlighted the lack of such data over time as well as in the global South, noting that ‘[w]ithout global data on a given measure, researchers cannot clearly see differences across place, and without historical data, they cannot see changes over time’ (Carnes and Lupu Reference Carnes and Lupu2023, 264). With PtP, we include 141 countries from all regions of the world. Moreover, PtP codes a 55-year period, from 1966-2021, and it includes the entire government, not only heads of state or specific ministers. Finally, it includes varied and detailed information – whenever available – on these (close to 45,000) individuals’ backgrounds. We intend to update the dataset to include more recent years, as well as more countries, going forward.
Creating the Dataset
In this section, we describe how the dataset was gathered and the variables made available in the dataset. Given its coverage, and with the permission of their team, we used an updated version of WhoGov (Nyrup and Bramwell Reference Nyrup and Bramwell2020), containing basic information on cabinet members across 1966-2021, for identifying PtP’s coding units. We subsequently collected information on eleven variables pertaining to these cabinet members’ educational, occupational, and social background.
Data collection lasted about three years, involving fourteen research assistants (RAs) with diverse language skills and twenty external researchers with in-depth knowledge of specific countries that would otherwise have been hard to code. The RAs and researchers conducted systematic internet searches for each minister, mainly relying on openly available sources such as Wikipedia, LinkedIn, biographies, obituaries, and news articles. Coders were trained to obtain the same understanding and interpretation of the codebook. Internal communication between RAs was facilitated via a Slack channel and joint coding sessions. External researchers were assigned contacts within the RA team to ease the coding process and ensure consistency. Further, the larger team regularly held meetings to discuss challenges and settle joint interpretations of recurring and difficult coding decisions.
We are quite confident that we, by following the outlined strategy, are gathering close to as much information as is possible for this type of cross-country data collection exercise (under reasonable budget constraints). We also believe that our coding strategy ensures that the integrity and quality of the PtP data are high. In this regard, we note that source validity and transparency are evaluated on a case-by-case basis, and when necessary, a double-verification process was used to ensure that the correct information is recorded. Still, when coding close to 45,000 ministers, coding errors are unavoidable, and we address specific issues in a separate limitations section below.
Variables
PtP registers characteristics of individuals before entering government on 11 variables from three main variable categories: 1) educational, 2) social, and 3) occupational background. Table 2 lists the categories for selected categorical variables, while additional variables, including several derived variables and aggregated country-year variables, are listed in Appendices B and C. Appendix A contains the complete codebook. We limited our data collection to these variables due to resource constraints (for example, we do not gather information on ministers’ careers after leaving government), because the data was too hard to obtain (for example, salary information), or for ethical and privacy reasons (for example, sexual orientation).
Table 2. Selected (categorical) variables in PtP
The variables in PtP come in addition to the original variables already coded in WhoGov, which can easily be linked to PtP given the identical coding units. These are the minister’s name, gender, year of birth, year of death, party affiliation, position (such as a minister of finance), and various ways of classifying the position. We refer to Nyrup and Bramwell (Reference Nyrup and Bramwell2020) for more information on the WhoGov variables.
Educational Background
PtP contains five educational variables that register information on cabinet members’ highest level of education before entering the cabinet. These include the highest degree obtained (ten categories) and type of degree (sixteen categories; for example, medicine, economics, law). Moreover, we coded the (English) name of the university for the minister’s last degree. The university names have been cleaned, so they follow the World Higher Education Database. Finally, we code whether the minister has studied abroad and, if yes, in which country or countries.
Social Background
PtP contains four social background variables, which register information on geographic and socioeconomic conditions during early childhood. First, we register birthplace at the city/village level. Second, we code whether ministers belong to their country’s nobility or royal family (if relevant). Third, we code whether they come from political families, operationalized as families where at least one parent or grandparent held political office.Footnote 2 Finally, whenever we can find relevant sources, we code family background. This coding is primarily based on class origins as reported in sources or occupation of the minister’s caretakers (often mother and/or father; see codebook for detailed coding rules). The resulting multi-category variable captures two relevant dimensions, namely a conventional trichotomous class dimension (lower, middle, upper) and sector (primary vs. other).
Occupational Background
Finally, two variables concern occupational background. The first is a 26-category measure of main occupation before entering politics. We code the work-related activity on which the minister spent most of their time during the decade before becoming a full-time politician. The categories were originally based on the ISCO-08 classifications, but were amended through an iterative process where we tried to limit the number of categories to ease coding, and simultaneously cover the diverse occupations of cabinet members before entering politics. The second (trichotomous) variable concerns political experience prior to their first ministerial appointment. Table 2 displays the categories for both variables.
Building More Measures
The many multicategory variables and different types of information contained in PtP give flexibility, as dataset users may construct several measures of theoretical interest beyond the original variables.
Consider occupational categorization. Although a nuanced 26-category variable is useful for many purposes, more aggregated measures are better suited for others. We can, for example, raise the level of abstraction by creating a simple occupational class variable. Although lacking the information to accurately reconstruct established class schemes from political sociology (e.g. Erikson & Goldthorpe, Reference Erikson and Goldthorpe1992; Oesch Reference Oesch2006), we create a simplified, occupationally based class index that captures the main hierarchical divisions. We distinguish between the 1) upper and upper-middle classes, 2) middle and lower-middle classes, and 3) working class. Note that this derivative measure is based on the minister’s own occupation before entering politics, unlike the original ‘class origins’ variable in PtP, which is based on the class of the minister’s parents (or other primary caretakers).
For most occupational categories, class category placement is clear-cut. ‘Blue-collar worker in industrial sector’ is a working-class background, whereas CEOs and judges undoubtedly belong in the upper or upper-middle class bin. Other categories are more complicated, such as ‘medical doctor or other health care workers’. Here, we combine occupational categorization with information on education to distinguish occupational class positions. For instance, ministers are classified as working class if they belong to this occupational group and do not have any post-secondary education (typical of, for example, unskilled care workers); middle or lower-middle class if they have some post-secondary education or lower tertiary education (for example, nurses); and upper or upper-middle class if they hold higher tertiary education degrees (for example, doctors).
Another difficult issue is how to deal with white-collar categories that may include both working-class and middle-class occupations depending on the temporal or geographical context. We discuss this further in Appendix B, where we suggest a class schema (with two alternative ways of coding these white collar categories), detail the coding rules, and provide descriptive statistics. However, this is only one possible class index that researchers could build from our original data; more complex indices that combine occupational and educational information with family background information can be easily constructed. More generally, researchers may construct numerous interesting variables from our original multi-category ones. As another example, we have constructed a derivative variable that registers whether ministers have a ‘military background’ by using information from several variables, namely study field, occupational background, and whether the person uses a military title. Appendix B lists all variables, including derived ones that we have constructed.
The individual-level dataset (that is, with ministers as units of analysis) can also be used to create a country-year dataset with information for the whole cabinet for every country on variables such as the share of ministers with a higher university degree or the share who worked as lawyers or CEOs prior to entering politics. We have created a set of country-year variables (listed in Appendix C) to be made publicly available, but users can easily create additional such variables.
Assessing the Data Quality
We assess data quality through three types of tests. First, we conducted an intercoder reliability test of 400 ministers, randomly selecting and recoding ten ministers each from forty different countries. Countries were chosen through a block-randomization procedure to ensure variation in GDP per capita, population, and democracy levels. We detail the procedure and results in Appendix G. Generally, the reliability between coders is very high (Average: >80 per cent), albeit lower for the class origins variable (71.1 per cent) and occupation variable (60 per cent). The former variable is more open to subjective interpretations, and the latter is harder to code consistently due to the 26 fine-grained answer categories. Education degree also scores lower (67.4 per cent), although its reliability performance is substantially increased when weight-adjusted by the extent of disagreement. The latter finding reflects that when coders disagree on degree, they tend not to disagree ‘by much’, often selecting neighbouring ordinal categories (for example, one coder says a minister has a PhD, the other that the minister has a master’s degree). Other variables have near-perfect agreement, such as whether the minister belongs to a royal family (99 per cent) or studied abroad (92 per cent). Moreover, inter-coder reliability is typically somewhat higher for richer, populous, and more democratic countries.
Second, we test the convergent validity by comparing the PtP measures with relevant measures from other datasets, namely Alexiadou, Spaniel and Gunaydin (Reference Alexiadou, Spaniel and Gunaydin2022), LEAD (Ellis, Horowitz and Stam Reference Ellis, Horowitz and Stam2015), and Vittori et al. (Reference Vittori, Pilet, Rojon and Paulis2023). We calculate country-level means for these variables across all ministers, for each dataset.Footnote 3 Figure 1 shows, generally, high agreement, but several discrepancies. Still, closer scrutiny indicates that divergences are often (largely) explained by different coding-rule decisions and definitions, rather than by measurement error in any dataset. For example, PtP generally codes educational attainment as lower than the comparison datasets. This is, in part, because some educations, such as teacher education, are university degrees in some countries and post-secondary education in others. PtP has a more nuanced scheme than comparable datasets and generally codes these educations according to country-specific norms. Moreover, PtP tends to register more members of the government with military careers. This is because our military career variable is constructed using occupational, educational, and title data, providing additional paths to register a military career compared to the relevant LEAD variable.Footnote 4
Figure 1. Cross-Validating PtP against other datasets.
Third, we assess missingness patterns. In Appendix D, we show the share of missingness for all variables and for each variable for each country. Missingness is, on average, around 20 per cent for most variables. However, it is higher for political family (34 per cent) and class origins (58 per cent). There are large variations from country to country; in many countries, the data is complete, while relevant information has been harder to come by in others. We conduct analyses in Appendix E to identify systematic missingness patterns. We find that missingness is disproportionately concentrated in poorer and less democratic countries, cabinet posts regarded as less important (for example, ministers of culture or transportation), and for male ministers.Footnote 5 Regarding time trends in missingness, we observe some differences between variables. We find a decreasing trend in missingness for degree, occupation, and political experience, while we see no such trend for our class origins variable.
Limitations and Considerations
Although validation tests generally provide confidence in the data, challenges with data gathering and resulting measurement errors and missingness patterns deserve attention. In the following, we discuss challenges and limitations that users should be aware of when using the data or specific variables. We consider the three main variable categories in turn, before discussing general issues.
Let us start with the educational variables. One issue is that we only code the last obtained degree: if a minister obtains a bachelor’s degree in law, a bachelor’s degree in theology, and a master’s degree in theology, we only capture the master’s in theology. Thus, we do not record the full information on the minister’s educational achievement. Another issue is that the level at which a person graduated is often indeterminate, as the available sources simply state that the person graduated from a given university. In such situations, we do not know whether the minister obtained a bachelor’s or master’s degree. Here, we record that they have attended university, but the degree is unclear. Furthermore, we uncovered several cases where it was difficult to know whether a minister has received an honorary degree or whether the person has actually finished their education. An additional issue is that some cabinet members may lie about their educational attainment or have cheated during their studies, and it is practically impossible for us to verify the authenticity of their educational qualifications.
There are limitations also for the variables covering social background. First, when coding the place of birth, information is not always available at the city/village level. Instead, and especially in rural areas, we can only find information on the district or region. There may also be multiple places with identical names. To address this, we systematically double-checked entries where Google Maps says there are multiple locations with the same name. Second, when considering whether the person comes from a royal family, we also code whether the person comes from local nobility. For example, in Malaysia, there are several sultanates, and we code whether cabinet members come from any of these. Hence, the binary coding for this variable masks heterogeneous characteristics. Third, it is hard to find information on whether a minister belongs to a political family, especially in developing countries with more limited sources. Fourth, in some parts of the world, we seem to find more information (on other variables) for ministers of upper-class origins than for other ministers. This difference in available information might generate systematic missingness patterns. Moreover, concerning class origins, we consider a minister’s upbringing in relation to the norms of their country. This choice also partly reflects that we, for this variable, often must rely on secondary sources’ fairly crude description of a minister coming from, say, ‘an upper-class family’ or having ‘a working-class background’ (instead of more specific occupational descriptions). Thus, what is considered middle-class origins in one country may occasionally register as lower-class origins in another country, especially when there are cross-country discrepancies in the status of particular occupations. More generally, we rely on the regional, cultural, or language expertise of our research assistants to make such classifications (often after reading or speaking with others), implying that there inevitably is an element of subjective evaluation in the coding of this variable.
Concerning our occupational variables, the list of occupational categories is comprehensive and covers many common occupations. However, it is far from exhaustive; providing a full list of (less common) jobs is simply not feasible, and it would make the coding process more complex and time-consuming. This entails that we, in practice, often rely on the residual categories pertaining to ‘other’ jobs, but which we then separate into white- or blue-collar jobs as well as private-, public- or unknown sector jobs for the (heterogeneous) white-collar category. Further, there is a fair amount of subjective judgment when the coder transfers a job title into a concrete score, since many real-world jobs do not fit squarely within a category. In addition, we only code cabinet members’ main occupation based on the decade before they entered politics. Thus, ministers could have had other occupations that are not recorded. Finally, to reduce missingness and coding costs, we rely on a trichotomy when coding cabinet members’ political experience before entering government. Unfortunately, this simple categorization leaves out information on the level at which ministers have been involved in politics.
Turning to more general limitations with the PtP data, we first want to highlight that three overarching goals behind this project have been to create a dataset that is 1) truly global, 2) comparable across countries and time, and 3) to minimize missingness. To achieve these goals, we have often opted for fewer or simpler categories than what some users may wish for. This is, for example, the case with occupation and political experience. In addition, we limit the richness of information by only coding one occupation or degree. Further, there is a risk of false negatives on several variables, such as having studied abroad and whether they are from a political family, as coding the absence of such characteristics must often be based on the absence of information; for example, it is very rarely stated that a person did not study abroad. As already noted, there is also substantial missingness on some variables for some countries. This is despite having hired country experts to code the most challenging countries (for example, smaller ones with country-specific languages). This issue is, however, smaller for more important cabinet positions and in recent years. Nonetheless, researchers may want to run robustness tests to account for potential biases stemming from certain systematic missingness patterns, for example, by dropping countries with high degrees of missingness.
Applications
Despite the above-described limitations, we are hopeful that the measures and data described in the previous sections will contribute to new descriptive knowledge and ultimately a better understanding of members of government worldwide. Figure 2 illustrates this point by mapping four measures calculated from the PtP data. These are the shares of cabinet ministers who have attended university, studied abroad, have a political background, and come from a political family. The colour shading indicates the average for each country across all cabinet members serving from 2000-2021.
Figure 2. Country averages on core variables across 2000-2021.
Countries that are grey are not included in the dataset
The main takeaway from Figure 2 is the substantial cross-national variation in government member characteristics, both within and across geographical regions. The upper left map, for instance, shows that close to all cabinet members in the coded American countries have at least some university education, whereas only around half of Scandinavian ministers attended university. Likewise, the map on the bottom left displays considerable cross-country variation in whether government members have studied abroad. Only a few ministers have done so in the USA and Russia, while studying abroad is the norm in many non-Western developing or middle-income countries, such as Kenya and Madagascar, but also richer ones, such as Saudi Arabia.
Other background characteristics also vary widely across countries. The upper-right map in Figure 2 shows that in some countries, such as Australia or Germany, almost all cabinet members are involved in politics before becoming ministers, suggesting a professionalized political class and the presence of institutionalized political parties. In other countries – notably autocracies, presidential systems, and countries experiencing political instability – ministers with prior political experience are rarer. Egypt and Libya stand out, but also in Colombia, Chile, and South Korea, less than half of the ministers have a prior political background before becoming a member of the government. Finally, the lower right map shows variation in the share of ministers from a political family. Royal dictatorships, such as Oman or Saudi Arabia, have high shares, but so do some more democratic countries, such as the Philippines and India.
To further illustrate the broad applicability of PtP and its relevance for answering very different types of research questions, we move on to show how the data can enable empirical analysis in six different subfields. We do so through simple descriptive analysis on different measures included in the dataset.
Comparative Politics
PtP measures may enable researchers to shed new light on several prominent questions and debates in comparative politics. To illustrate this point, we consider the question of regime differences in education and competence of political elites. More specifically, Besley and Reynal-Querol (Reference Besley and Reynal-Querol2011) theorize and show that democracies tend to select more educated leaders than autocracies. But, do democracies also select better-educated cabinet members? We consider this to be an open empirical question. On the one hand, autocratic leaders with limited educational attainment may try to compensate for whatever missing skills (or credibility) their lack of education entails by appointing highly educated cabinet ministers (Ketchley and Wenig Reference Ketchley and Wenig2023). Yet, autocratic leaders might also be wary of appointing highly skilled individuals to powerful positions, since different arguments indicate that they could more effectively threaten the autocrat’s own position and hold on power (Egorov and Sonin Reference Egorov and Sonin2011; Zakharov Reference Zakharov2016; Aaskoven and Nyrup Reference Aaskoven and Nyrup2021).
Shares of leaders and cabinet members with a graduate degree, by regime type. Following Besley and Reynal-Querol (Reference Besley and Reynal-Querol2011), we differentiate regimes by using 0 on Polity2 as a cut-off for categorizing democracies and autocracies. Operationally, we calculate the average of the cabinet/country averages for each year. We use data on 7,132 leader years and 173,388 cabinet member years.
Figure 3 compares autocracies and democracies by showing the proportions of, respectively, top leaders and ministers, globally, with graduate degrees. Despite differences in the periods and countries included, we corroborate the substantial difference in leader education found by Besley and Reynal-Querol. Differences are marked throughout the period after 1966, with less than one-fifth of autocratic leaders having graduate degrees. In contrast, the majority of democratic leaders have typically had such degrees, especially post-1980. Findings differ for cabinet ministers. While ministers are, on average, more highly educated in democracies than in autocracies throughout the period, the proportions do not deviate much from each other. Especially in later years, shares have been approximately similar, close to 60 per cent. In Appendix F, we assess robustness by using similar econometric specifications as in Besley and Reynal-Querol’s paper (Table 1), and show that there is no significant difference in educational attainment between cabinet members in autocracies and democracies. This suggests that autocrats may be hiring more competent (or highly educated) subordinates than previously assumed in previous research.
Figure 3. Do democracies select more educated cabinet members?
Political Sociology and Representation
The lack of global and over-time data on cabinet ministers’ socio-economic background was recently highlighted by Carnes and Lupu (Reference Carnes and Lupu2023). Key descriptive questions in the literature on political representation that remain unanswered are thus, for example, to what extent have different social classes been represented (descriptively) in governments, in different parts of the world and in different periods of time? The PtP data allow us to address this question, and, in extension, study related questions of interest to political sociologists, political scientists, and others (Barnes, Kerevel and Saxton Reference Barnes, Kerevel and Saxton2023). For instance, concerning party politics and electoral behaviour, one may use PtP data to address how and when parties choose to represent the working class, and the implications that this may have for parties’ electoral support in different voter groups (Heath Reference Heath2015). Alternatively, researchers may combine PtP data on working-class representation with available survey data to assess how workers and other groups react in terms of political trust or satisfaction with democracy, or to study if changes in such descriptive representation lead to corresponding changes in substantive representation of working-class voters’ interests.
Figure 4 exemplifies the wide variation in working-class representation in cabinets, using the above-described derivative class measure that we constructed from the original PtP measures on occupation (and education). Concretely, the figure visualizes the share of ministers with working-class occupations before entering politics in two countries, Norway and the USA, across 1966-2021.Footnote 6
Figure 4. Do cabinet members have working-class occupations before entering politics?
The figure shows the share of cabinet members in Norway and the USA, with working-class occupations prior to entering politics. The classification of occupations follows the definition provided in Appendix B
The USA – following our operationalization of occupational class – has never had working-class representation in government throughout the period under study, mirroring findings from legislator data (Carnes, Lupu and Pontusson Reference Carnes, Lupu and Pontusson2021). This results in considerable (descriptive) under-representation for a group comprising a substantial portion of the American population (Carnes, Lupu and Pontusson Reference Carnes, Lupu and Pontusson2021, 185). Indeed, the PtP data indicates the USA is not exceptional in this regard; the majority of countries, globally, have never had a single working-class minister since 1966.
However, some countries have had considerable working-class representation. In Norway, the working class has, historically, been rather well-represented compared to in most other countries for which we have data. Yet, closer inspection suggests differential political representation strategies for different political parties. More specifically, the working class is much better represented in the cabinet whenever the Labour Party governs (indicated by shaded areas) than when right-wing parties or centre-right coalitions govern. Yet, we also observe decreasing working-class representation in Labour Party governments over time. Finally, one important exception to the first pattern appears towards the end of the period, when a right-wing government had higher-than-usual working-class representation. Interestingly, this reflects that a populist right party (the Progress Party) entered government for the first time in 2013 (in coalition with the Conservative Party). This party, like many other European populist right parties, aims to represent and attract working-class voters (for example, Evans & Langsæther (Reference Evans and Langsæther2021); Langsæther (Reference Langsæther2023), but see Greilinger and Mudde (Reference Greilinger and Mudde2024) for a critical perspective on the extent to which they actually represent the working class’ interests).
Gender Studies
An ongoing debate in research on gender and politics concerns the degree to which women are at a disadvantage in politics. PtP can be used to address this broader issue, inter alia, by comparing male and female cabinet members on relevant background characteristics. Here, we focus on only one such characteristic, namely, whether a cabinet minister is from a political family. Figure 5 shows that female ministers were much more likely to come from political families than male ministers in the 1960s and 70s, whereas today, no such gender differences remain. This corroborates country-specific trends found in studies of legislators in Sweden and Ireland, and indicates a decreasing level of screening discrimination against women over time (Folke, Rickne and Smith Reference Folke, Rickne and Smith2021).
Figure 5. Are male or female cabinet members most likely to come from a family of politicians?
The plot shows the share of cabinet members by gender who are from a family of politicians. A person is defined as coming from a family of politicians if at least one of the person’s parents or grandparents held political office during their lifetime. The plot includes data on 117,087 minister years.
Figure 5 also illustrates how new information in PtP may fruitfully be combined with existing ascriptive information from WhoGov – in this case, gender – to gain new knowledge. More generally, combining such information may allow us to study how female and male political elites differ from each other, both across countries and over time, and contribute to a rapidly growing field focusing on the role of women in government (Armstrong et al. Reference Armstrong, Barnes, Chiba and O’Brien2024; Armstrong et al. Reference Armstrong, Barnes, O’Brien and Taylor-Robinson2022; Nyrup, Yamagishi and Bramwell Reference Nyrup, Yamagishi and Bramwell2024; Kroeger and Kang Reference Kroeger and Kang2022).
Public Administration
PtP provides rich and nuanced information on who oversees and manages government policies at the highest level in countries across the world. This makes it relevant to scholars in public administration. Figure 6 shows the pre-political occupations of government officials. Leaders, that is, heads of government or state, have varied occupational backgrounds, but we find much more specialization for some key cabinet portfolios. For example, 35 per cent of defence ministers were military employees and 26 per cent of foreign ministers were foreign service employees before entering politics.
Figure 6. What did different types of cabinet members do before entering politics?
The plot shows the top five occupations (and the share with this occupation) for different types of cabinet members globally in the period 1966–2021. There are 1,165 leaders (heads of government or state), 2,139 defence ministers, 2,133 foreign ministers, and 2,325 finance ministers in the sample.
A growing line of research considers the explanatory power of previous occupations of political leaders for their behaviour while in office (Krcmaric, Nelson and Roberts Reference Krcmaric, Nelson and Roberts2020). However, existing studies are limited to a few countries or focus solely on the top leader. PtP enables scholars in public administration to assess such arguments – and consider their general applicability or identify core heterogeneities across contexts or leadership positions – by using more extensive data.
International Relations
PtP allows us to study how governing elites can affect foreign policy behaviour and other behaviour with implications for inter-state relationships. For example, leaders’ personal attributes have been shown to influence whether a state goes to war or not (Horowitz and Stam Reference Horowitz and Stam2014). Scholars can now extend these types of studies by looking at the backgrounds of individuals in the entire cabinet (or selected portfolios) instead of just the top leader.
We might also study soft power in international relations by examining where cabinet members have studied (Nye Reference Nye1990). Figure 7 depicts the share of cabinet members worldwide who studied in five major countries over time. We observe that the USA overtook the UK as the biggest educator of cabinet ministers in the early 1980s. The registered increase in Russia/Soviet Union comes after the Soviet Union’s dissolution, and the Soviet/Russian education is registered for many ministers in newly independent post-Soviet republics (outside Russia). Very few non-Chinese cabinet ministers have historically studied in China.
Figure 7. Where have the world’s cabinet members studied?
The plot shows the share of the world’s government members who have studied in a particular country over time. We exclude ministers studying in their own country. Russia includes both the Soviet Union and Russia. The plot includes data on 148,611 minister-years.
More generally, the relevance of political elites’ place of study often receives considerable attention – consider, for example, the presumably important role played by Chicago-educated economists (The Chicago boys) in Pinochet’s Chile – as learning and socialization at (foreign) universities constitute key formative experiences (Krcmaric, Nelson and Roberts Reference Krcmaric, Nelson and Roberts2020). PtP data, with university-specific coding of ministers’ educational background may be used to assess such notions in a systematic manner. Researchers in international relations may also use PtP data to investigate whether having many foreign-educated ministers enhances alliance-building or open trade policies.
American Politics and Single Country Studies
In the applications so far, we have used data on cabinet members across countries. However, PtP can also be used to study single countries to better understand micro-level processes and give us a better understanding of politics by providing a more detailed mapping of relevant actors and their characteristics, something single-country studies are especially suited for (Pepinsky Reference Pepinsky2019).
Figure 8 maps US cabinet members’ birthplace, showing that relatively more cabinet members registered as Democrats are born on the West or East Coasts, whereas relatively more Republicans are born further inland. While this might not surprise current observers of US politics, the PtP data allow us to substantiate and document this pattern in a systematic manner, and also, potentially, identify and analyze more nuanced geographical patterns.
Figure 8. Where are democratic and republican cabinet members born?
The map shows where cabinet members appointed to office in the USA in the period 1992-2021 were born. 182 cabinet members are included in the sample. Independent cabinet members are those who do not belong to either party.
And, although we here plot the geographical origins of US cabinet members, patterns of geographic representation can be explored for all countries included in PtP. Related, this information can also be used to analyze questions about policies and their redistributive effects. Previous work has shown that leaders, when in office, pursue policies that favour their native regions (see, for example, Hodler and Raschky Reference Hodler and Raschky2014). By using PtP data, we may assess whether similar patterns exist when we consider the geographical origins of the wider government (or more specific ministers, such as ministers of transportation, industry, or finance).
Lastly, Figure 9 shows where British cabinet members attended university, revealing a considerable concentration in university backgrounds; unsurprisingly, Oxford and Cambridge graduates dominate. This pattern is also found in other studies (Berlinski, Dewan and Dowding Reference Berlinski, Dewan and Dowding2007) and may indicate that the political elite in the UK is highly unrepresentative of the population (also in other regards that correlate with the place of education, such as social background). Future research can use PtP to assess such patterns of university concentration and representativeness also in other countries and between cabinets (with different ideological backgrounds or under different regimes) within countries.
Figure 9. Where did British cabinet ministers attend university?
The plot includes data on 249 British cabinet members who have been in office in the period 1966-2021.
The applications, in different fields, that we have presented provide just a few illustrations of the potential ways in which researchers may use PtP variables to map the educational, occupational, and social backgrounds of ministers across countries and over time. We note that the extensive time series, with ample cabinet- and minister-level variation, may also facilitate studies of both the causes and consequences of such variation. For example, by combining our data with global (for example, World Values Study), regional (for example, Afrobarometer or European Social Survey), or national survey data, scholars may now study how various forms of descriptive representation shape political trust, protest behavior, or electoral participation of various over- or underrepresented social groups.
Conclusion
We have introduced the Paths to Power (PtP) dataset, which adds information on the social, educational, and occupational background on 44,789 cabinet ministers (whenever available). These are all cabinet ministers with available information from 141 countries who held a post during July from 1966 to 2021. In addition to discussing the contents and process behind collecting PtP, we have discussed data quality and limitations. As with all datasets of this scale, and despite our best efforts, mistakes are bound to happen. Readers are encouraged to contact the authors at paths-to-power@stv.uio.no if they have suggestions for corrections. This article presents the first version of PtP, but the authors will continue to work on improving the data and adding more countries and years to the data going forward.
Since PtP supplies global and comparable data over almost six decades, it particularly improves on the availability of data outside of ‘Western democracies’, where similar measures on the backgrounds of cabinet ministers are somewhat better covered in existing sources. PtP may thus be of particular relevance for scholars focusing on the global South and non-democratic countries, countries that tend to be understudied in political science (Wilson and Knutsen Reference Wilson and Knutsen2022). Yet, PtP data should open up research opportunities for scholars studying different regions and countries, or scholars doing cross-country work on global samples. In this article, we illustrated how PtP can be used to address key questions in comparative politics, political sociology, public administration, gender studies, international relations, as well as American politics and other single-country studies. However, the questions and relationships addressed in this article only scratch the surface of the opportunities for empirical research opened up by PtP.
To highlight a few specific such questions and relationships, researchers may use PtP to address questions such as how class origins or the occupational class of ministers shape redistributive policies, whether having many foreign-educated ministers enhances alliance building or open trade policies, or what social and political-institutional conditions spur more heterogeneous governments. Several of these questions have been addressed by previous research, but typically relying on more limited data material. PtP can then be useful for assessing the generality and scope conditions of existing findings and theories. Other important questions have yet to be empirically assessed, and yet other questions will hopefully be thought up by researchers working with our dataset. If so, PtP may help generate new insights into the social composition of political elites and how this composition influences policies and various outcomes of importance to citizens worldwide.
Supplementary material
To view supplementary material for this article, please visit https://doi.org/10.1017/S0007123425100598
Data Availability Statement
Replication data for this article can be found in Harvard Dataverse at https://doi.org/10.7910/DVN/TP42YG.
Acknowledgements
First, we would like to thank the many people who helped us with gathering the data: Anna Nybu, Chandler Williams, Eirik Haugen, Gabriel Lönn, Javier Leal, Johanne Skar, Jonas Berge, Lee Cojocaru, Lisa Løkkebø, Mari Flaathen, Ragnhild Dalen, Stephany Amaya Pérez, Tuda Oussous, and Vera Hansen. Furthermore, we would like to thank Aliyah Siddika, Akash Chopra, Dércio Tsandzana, Ella Prihatini, Grace Evans, Ha Eun Choi, Jalal Fetrati, Janina Beiser-McGrath, John Rovic Catangay, Kenya Amano, Lauren Robertson, Johnathan James Guy, Josef Woldense, Jou-Fei Huang, Ratchakit Khonkam, Saida Hassan, Sanghoon Park, Saber Khani, Saowanee Alexander, William Kakenmaster, Jens Koning, Roman Gabriel-Olar, and Yosef Tadesse, who have helped code their countries of expertise for us. In addition, we would like to thank Clara Neupert-Wentz, Matt Reichert, Hannah Smidt, Stuart Turnbull-Dugarte, Florence So, Matthias Haslberger, Matthias Dilling, Julia Leschke, as well as attendants at the 2023 European Political Science Association conference, 2022 Annual Meeting of the American Political Science Association, 2022 Annual Meeting of the Danish Political Science Association, and participants at presentations given at University of St. Gallen, Lund University, University of Oslo, the University of Tokyo, European University Institute, Swansea University, the Peace Research Institute Oslo, and the Comparative Research on the Executive Triangle in Europe conference for their thoughtful comments and suggestions on earlier versions of this manuscript. We are also grateful to the three anonymous reviewers and editors at BJPS, who provided insightful comments that helped improve the manuscript.
Financial support
This work was supported by the Research Council of Norway Grant, pnr 300777, and a Småforsk Grant by the Department of Political Science, University of Oslo.
Competing Interests
The author(s) declare none.
Open access
