Development status as a measure of development

2.1 Classification schemes and their implications for data quality

A classification scheme can be understood as the descriptive information about the way observation units are arranged into groups, based on common characteristics [12]. This arrangement into groups can reduce complexity and thereby facilitate the interpretation and processing of larger bundles of data. A DSCS can be understood as a classification scheme for the arrangement of countries into groups defined by development status. For the purpose of this study, this definition is slightly broadened. A DSCS will be referred to as the descriptive information about the way countries are arranged into development status groups, with or without the application of common characteristics. To facilitate comparison across schemes, only two different development status groups will be distinguished: developing countries and all others.

Looking into the question what characterizes a good classification scheme, the Organisation for Economic Co-operation and Development (OECD), based on recommendations of the Expert Group on International Statistical Classifications [13], establishes the following criteria [14]:

the categories should be exhaustive and mutually exclusive;

Further criteria are identified by Shorrock, an Ergonomist, including: (8) “face validity”, by which he means that a classification system should “look valid” to people who use it. He recommends to “stick with what is well understood and accepted” [15].

OPEN IN VIEWER

Table 1

Relation between quality criteria of classification schemes and the statistical output

A final criterion arises from the fact that the arrangement into groups based on common characteristics necessarily leads to a certain loss of information. The various attributes characterizing the classified objects are transformed into only one (categorical) variable, so that the differences between objects assigned to the same group are suppressed, and the differences between objects assigned to different groups are treated the same way regardless of their size. The information loss caused by a classification scheme can be considered to be the smaller the more similar the objects of the same categories and the more dissimilar the objects of different categories. On these grounds, Farias argues that (9) low intra-class heterogeneity and high inter-class heterogeneity are also useful quality criteria of classification schemes [11].

The design of a classification scheme is important for data quality, as the criteria above are causally linked with widely recognized quality aspects of statistical output, in particular with comparability and coherence, clarity and interpretability, and with relevance [16, 17, 18]. The criteria 1, 2 and 3 above are linked with comparability and coherence. If the categories defined by a classification scheme are not mutually exclusive (criterion 1) then an observation unit may be assigned to one category in one dataset and to a different category in another. If the classification is not fully comparable to other related classifications (criterion 2) then opportunities to compare groupings and aggregates across datasets and to combine them for further analyses are limited. If a classification is changed too frequently (criterion 3) then the risk increases that data published in one release cannot be compared or reasonably combined with data from an earlier release, thus comparability over time is reduced.

Criteria 1, 4, 5, 7 and 9 are linked with the clarity and interpretability of data. If the categories are not mutually exclusive or too narrowly defined (criterion 1) then the actual significance of the categories becomes blurred. If too few categories are chosen, the classification scheme risks being simplistic; if there are too many, it may be difficult for users to keep overview (criterion 5). If dissimilar objects are united in same categories or categories are too similar to each other (criterion 9) then data aggregated at category level cannot be interpreted as being representative of and specific to the objects covered by these categories. If the classification scheme or its resulting categories are not well described and documented (criteria 4 and 7) then users may be unclear about their actual meanings and falsely interpret them according to their subjective ideas about the objects described by the category labels.

Criteria 6 and 8 above are linked with the relevance of statistical output. If categories are defined in a way that they have not much to do with reality (criterion 6) or if they do not “look valid” to users (criterion 8) then the aggregates and groupings produced will be – or be perceived as being – of limited utility for users. Table 1 sketches the just described links between quality criteria of classification schemes and quality dimensions of statistical output.

For a proper understanding of a classification scheme and for users’ evaluation of its relevance, concepts play an important role. A concept can be understood as a “unit of knowledge created by a unique combination of characteristics” [12]. In Cognitive Psychology, concepts occurring as mental representations have been found to be crucial for people’s ability to understand their environment, as they are used for categorization. If we can assign new cases to categories then we do not need to explore them in detail and we save time and energy [19]. In principle, everyone develops their own concept of a developing country. The more this concept matches the concept used in a DSCS, the clearer the meaning of the grouped or aggregated data will be, and the easier it will be for users to interpret those data, and the more relevant the data will appear to them. In turn, classification schemes also shape people’s understanding of categories. This is the reason why high incongruence between DSCSs can hamper productive discourse and scientific progress, as pointed out by Nielsen [3] above. To summarize, the match between classifications applied in statistics and concepts formed in people’s minds constitutes an important determinant of the clarity, interpretability and relevance of aggregated or grouped data. Analysing this match, for the case of development status classifications, is the primary objective of the present study.

The history of DSCSs begins in the 1960s. In the first global cross-national datasets of economic indicators, the Statistical Yearbook of the League of Nations [20] published from 1919 to 1944, and the International Financial Statistics (IFS) [21] published since 1948 by the International Monetary Fund (IMF), the data were, if at all, grouped and aggregated by continents, not by development status. After World War II, a need for other types of groupings and aggregations emerged, as the divide between the richer industrialized countries in the ‘North’ and the poorer countries in the ‘South’, that had recently achieved independence from colonialism, became a focus of public discourse. The poverty in the regions of the South increasingly caused concern all over the world. The affected countries became aware of their common problems and began jointly defending their interests within the United Nations system. A group of 75 countries which considered themselves as “developing countries” successfully struggled for the organisation of the first United Nations Conference on Trade and Development (UNCTAD) in 1964, at which specific problems of the developing world were addressed and changes in the “international economic order” discussed. They signed in 1963 a document, titled “Joint Declaration of the Developing Countries” in which their common position was articulated [22]. After the conference, most of these countries, reinforced by a few others, established the Group of 77 (G77) [23, 24, 25]. Today, the G77 comprises 134 member states and considers itself as “the largest intergovernmental organisation of developing countries in the United Nations” providing “the means for the countries of the South to articulate and promote their collective economic interests” [26].

A statistical background document prepared for the first UNCTAD presented data tables in which countries were grouped into the following three classes:

“Economic Class I” comprising whole Northern America and Western Europe, as well as South Africa, Japan, Australia and New Zealand;

In 1967, an updated and revised version of that document was published as the first edition of UNCTAD’s Handbook of International Trade and Development Statistics [27]. In the second edition of the Handbook, it was established that the classes I, II and III can be interpreted as “developed countries”, “developing countries” and “socialist countries”, respectively [28]. UNCTAD’s DSCS was born. Over time, almost all countries in class II became members of G77. The few exceptions comprise Aruba, Hong Kong and Israel, where Israel was later re-classified by UNCTAD as developed [26].

Other international organisations also introduced DSCSs into the global cross-country datasets they published. In 1964, the IMF set up a classification scheme for the IFS which distinguished between “industrial countries”, “other high income countries” and “less developed countries” [3]. In 1970, UNSD published a standard for the naming and groupings of countries, known as “M49” [29], which defined development status groups similar to the classes of the UNCTAD scheme. Over time, the M49 standard has become a key reference for the definition of country codes, names and groupings within the United Nations system and beyond. In the late 1970s, the World Bank (1978) published the first edition of its WDI, as a statistical annex to the World Development Report [30], in which countries were divided up into “developing countries”, “capital surplus oil exporters”, “industrialized countries” and “centrally planned economies”, thereby taking account of the particularities of oil-exporting countries that had come to the fore during the oil crisis. In 1981, the United Nations Industrial Development Organisation (UNIDO) began compiling cross-country data and in that context introduced a classification scheme which split countries up into “developing countries”, “centrally planned economies”, “developed market economies”, and China representing a class of its own [31].

UNSD, the IMF and UNIDO have not provided any explanation regarding the criteria applied to classify countries as developing [3]. UNSD has added a disclaimer to its M49 standard (quoted in the introduction), pointing out that the designations of developed and developing countries are intended for statistical convenience only and that they are not meant to express any judgement about a country’s stage in the development process. The IMF states, in the Statistical Annex to the World Economic Outlook, that their classification “is not based on strict criteria, economic or otherwise” and “has evolved over time”. Its objective would be “to facilitate analysis by providing a reasonably meaningful method of organizing data” [32]. UNIDO’s classification, according to Updahyaya, “evolved historically with no particular statistical measure being used”, and was “occasionally based on a country’s preference for one designation over another” [31].

OPEN IN VIEWER

UNCTAD provides an explanatory note on the UNCTADstat website [33], pointing out that the applied development status classification “has its origin in the coalitions formed during the preparation of the first United Nations Conference on Trade and Development” and “primarily reflects historically formed common interests and identities of economies”. The World Bank explained they used the criteria gross national income per capita, OECD membership, net exports of oil and the capital account balance as input for establishing their development status classes [30]. However, as Nielsen points out, the application of these criteria was not fully consistent [3].

All in all, it appears that until the 1980s the formation of development status groups was rather an outcome of countries’ self-identification, political considerations and expert judgement than of an objective application of specified criteria. Strictly speaking, these DSCSs of the first generation comply with only one of the two conditions established in the definition of a classification scheme given above (Section 2.1). They do provide descriptive information about the way observation units are arranged into groups, but they do not specify any common characteristics applied in that arrangement.

During this early phase, international organisation revised their DSCSs to different extents. The UNSD scheme remained unchanged; UNCTAD and the World Bank re-classified a few countries in the 1980s;1 and the IMF entirely reorganized its scheme for three times by redefining, merging and splitting categories [3].

The Venn diagram in Fig. 1 shows the commonalities and differences in the group of developing countries defined by the DSCSs above at the beginning of the 1980s. The five schemes had in common that most parts of Africa, Latin America and the Caribbean, and Asia were considered as developing. The World Bank and the IMF, unlike UNCTAD and UNSD, also included several Southern European countries in their definitions. The World Bank was the only organisation which excluded Libya and several countries on the Arabian Peninsula from the developing countries group, classifying them as “capital surplus oil exporting countries”. The World Bank also deviated from the UNCTAD, UNSD, and UNIDO classifications by not treating Cuba as developing, but as “centrally planned”. Similarly, Mongolia was considered as developing by UNIDO, but as “centrally planned” or “socialist” by the other organisations. The IMF was the only organisation at that time which classified South Africa as developing, thereby applying the same development status to all countries in Africa. Israel was classified as developing by UNCTAD, the IMF and the World Bank, but not by UNSD and UNIDO.

Despite these differences, the large majority of countries were classified in the same way by all five organisations, in the early 1980s. This high concordance is confirmed by the rank correlation coefficients depicted in Table 2. The DSCSs of UNCTAD and UNSD yield the most similar groupings, in line with the high overlap between the corresponding circles in Fig. 1. The scheme of UNIDO appears to be generally the most closely correlated with other schemes.

In the late 1980s, the breakdown of socialism created a need for a fundamental rethinking of the development status classification practice, as the categories of the “socialist countries” (used by UNCTAD) and the “centrally planned economies” (used by UNSD, World Bank and UNIDO) had become obsolete. In the absence of objective classification criteria, reallocating the members of these categories to others was not straightforward.

UNCTAD, in a revision of 1994, reclassified the Asian countries previously classified as “socialist”, comprising China, Mongolia, the Democratic People’s Republic of Korea, and Viet Nam, as well as the former Asian Republics of the Soviet Union, to developing economies, while the former socialist countries in Europe, except Yugoslavia, and the former European Republics of the Soviet Union, were kept in a group called “countries in Eastern Europe” [37]. By a revision in 2004, countries that had joined the European Union were removed from “countries in Eastern Europe” and added to developed economies. The remainder of this group was merged with Asian former Republics of the Soviet Union and with the successor states of Yugoslavia, previously a “developing economy”. The new group was given the name “South East Europe and the Commonwealth of Independent States” [38]. In 2007, this group was renamed into “economies in transition” [39]. It has kept its composition until today, except that Bulgaria, Romania and Croatia were reassigned to the developed economies after their accession to the European Union. Other revisions since the 1990s comprised a reclassification of South Africa from developed to developing, after it joined the G77 in 1994, and of Cyprus from developing to developed, following its accession to the European Union in 2004 [26, 39].

UNSD revised the M49 standard in 1996. The changes made reflected an aim to maintain congruence with broad geographic regions. Unlike UNCTAD, UNSD allocated the former centrally planned economies to either developing or developed regions – those in Asia to developing and those in Europe to developed. Similar to the UNCTAD scheme, South Africa was reclassified from developed to developing [1]. In a recent 2018 revision, the statuses of Cyprus and Israel were changed from “developing” to “developed” [40].

The IMF created a specific group for the former socialist countries which in the past, as non-members of the Fund, had not been assigned any development status. From 1993 on, this group was named “countries in transition”. After the first eastward enlargement of the European Union, in 2004, it was dissolved by reassigning the new European Union member states to developed and the rest of the group to the developing countries. The latter group was given the new name “emerging and developing economies” [3].

The World Bank followed an entirely different approach from UNCTAD and UNSD, by introducing average income per capita as an objective classification criterion. Income is measured as gross national income multiplied by an adjustment factor to smooth out the impact of exchange rate fluctuations. The cut-off between developing and developed countries was set at a gross national income per capita of US$6,000 in 1987 prices, the income threshold previously set up to differentiate high-income from middle- and low-income countries [36]. This threshold is revised each year to adjust for movements in prices and currencies, and the compositions of the groups are updated accordingly [41, 42].

In 2013, also UNIDO began grounding its DSCS on objective quantitative criteria, based on an analysis of the size of the manufacturing sector at different stages of development (see Section 3.3 for the underlying theory). Countries with a manufacturing value added greater than 2,500 United States dollars per capita, measured at purchasing power parity (PPP), or a gross domestic product greater than 20,000 United States dollars per capita, measured at PPP, are defined as “industrialized”. The others are classified as either “emerging industrialized economies” or “other developing economies”. The former category applies to countries with a manufacturing value added greater than 1,000 United States dollars per capita, measured at PPP, and for countries accounting for at least 0.5 percent of world manufacturing value added [31].

In 1990 a new DSCS entered the scene, introduced by the United Nations Development Programme (UNDP) [43]. Like the revised schemes of the World Bank and UNIDO, the UNDP scheme is strictly based on a quantitative measure, the Human Development Index (HDI), a composite index aimed at taking account of the multi-dimensional nature of development, in line with Sen’s capability approach (see Section 3.2). The index is a weighted average of four indicators: life-expectancy at birth, as an indicator of the ability to lead a long and healthy life; mean and expected years of schooling, as indicators of the ability to acquire knowledge; and GDP per capita, as an indicator of the ability to maintain a decent standard of living. Depending on the index values, countries are divided into classes with “low”, “medium”, “high” and “very high” human development [44].

Since 2000, the World Trade Organisation (WTO) has been publishing a statistical compendium called “International Trade Statistics”, in 2016 renamed into “World Trade Statistical Review”. It contains, among other data, aggregated figures for “developing” and “developed economies”. Since the 2014 edition, the countries included in these groups have been defined in the methodological annex. A justification for that grouping cannot be found. However, in the 2019 edition, the following disclaimer is made: “The references to developing and developed economies, as well as any other sub-categories of members used in this report, is for statistical purposes only, and does not imply an expression of opinion by the Secretariat concerning the status of any country or territory, the delimitation of its frontiers, nor the rights and obligations of any WTO member in respect of WTO agreements.” It is also explained that “there are no WTO definitions of ’developed’ and ’developing’ economies” [45].

In summary, the new schemes of the World Bank, UNIDO and UNDP can be seen as representatives of a new generation of DSCSs. By applying specified common characteristics for the definition of development status groups, these schemes substantially differ from the first generation of schemes, still applied by UNCTAD, UNSD and the IMF, and more recently introduced by WTO, which ‘nominate’ countries for different groups rather than identifying them by objective means.

As a result of the revisions above, the groups of developing countries defined by international organisations, sometimes with slightly different labels,2 have become increasingly heterogeneous. This is revealed by Fig. 2. For example, many former socialist countries in Europe (on the right side of the Venn diagram) are now classified in different ways. They are treated as developing by the IMF, the World Bank, UNIDO and UNDP, but not by UNCTAD and UNSD. UNCTAD does not consider the Asian former Republics of the Soviet Union as developing either, classifying them as “transition economies”. In turn, several countries in Asia and Latin America and the Caribbean (on the left side of the diagram) are still classified as developing by UNCTAD and UNSD, but not (anymore) by the other four organisations. This applies for example to the Asian ‘tiger states’ Hong Kong, Macao, the Republic of Korea and Singapore, classified as developing by UNCTAD and UNSD only. Brunei Darussalam and several Western Asian oil-producing countries are considered as developing also by the IMF, Brunei Darussalam and Saudi Arabia also by UNIDO, and Oman also by UNDP. The World Bank scheme has a high propensity to exclude countries in the Latin America and the Caribbean (in the left centre of the diagram) from the developing countries. Chile and Argentina are not given developing status by UNDP either, while UNIDO does not classify Trinidad and Tobago as developing. Further particularities include that UNIDO is the only organisation that considers Cyprus, Spain and Latvia as developing countries; WTO is the only organisation which classifies Israel as developing; and the IMF represents an exception by giving Hungary the status of a developing country.

OPEN IN VIEWER

The increase in heterogeneity of development status classification is confirmed by the rank correlation coefficients presented in Table 3. In 2018, unlike the early 1980s, for several pairs of DSCSs the correlation coefficient is below 0.6. In only a few cases it exceeds 0.8 (compare with Table 2). Correlation is in general lower between schemes of the first and second generations than between schemes of the same generation.

OPEN IN VIEWER

Figure 3 displays the stability of the groups of developing countries defined by the different schemes, between the 1980s and today. The bars measure the concordance between the former and the present versions of each scheme, based on Kendall’s tau, a rank correlation coefficient. The UNCTAD scheme turns out to be one which has changed least among the five DSCSs, followed by the schemes of UNSD and the IMF, which were subjected to far-reaching revisions after the dissolution of the socialist countries group. The schemes of UNIDO and the World Bank have changed most over the last 30 years. This is not surprising, given that in those cases not only the composition of the groups but also the criteria for their formation have been revised, and considering that the groups of the World Bank scheme are updated each year.

3.1 Difficult starting points

An early school of thought that influenced discussions on development in the 1960s was the dependency theory. It sees developing countries mainly as countries which have inherited from their colonial past an ineluctable dependence from the developed world. Exploited in the past as colonies by European empires for the extraction of raw materials and for slave trade, on achieving independence, the theory suggests, the old imperialistic structures were transformed into new dependencies, now driven by liberal markets, monopoly power, foreign direct investment and conditions linked to development aid. These dependencies would be facilitated by a growing penetration of the developing world with so-called “satellites” of decision centres established in the developed world [46, 47, 48].

Later, a number of economic studies focused on analysing long-term effects of a colonial past on economic growth. These studies often distinguish between colonies that provided favourable climatic and geographic conditions as destinations for settlers from Europe, namely Northern America, Australia and New Zealand, and the others which were primarily used for extraction of raw materials and supply of slaves and colonies. Negative long-term effects of colonialism on economic growth are primarily identified in the latter group of colonies, whereas in the former group it is claimed that the settlers’ descendants successfully struggled for the development of democratic institutions and free market exchange following the examples of their home countries [49, 50, 51, 52].

Some studies point out location in the tropics as an obstacle for economic growth characteristic for developing countries. Kamarck describes in a report of the 1970s the challenges which tropical conditions impose on agricultural activity, the treatment of raw materials and transport, as well as on people’s health and physiology [53]. Later the effect of location in the tropics on economic growth has been analysed based on econometric models [54, 55].

Difficult starting points, as they arise from the colonial past and geographic conditions, have been found not only to represent a burden of their own, but also to have a tendency to reinforce themselves. Neo-institutionalism deals with the strong persistence of sub-optimal hierarchic institutional structures inherited from colonial times, as these prevent formation of social trust among citizens [56]. The New Economic Geography provides evidence for diverging paths of economic growth between central and peripheral regions, caused by self-reinforcing effects of access to skilled labour, availability of physical factors of production and proximity to markets [57].

Furthermore, in the post-World-War-II period, Prebish and Singer showed that a continuous devaluation of primary relative to manufactured goods on the world market widened the productivity gap between the developed and the developing world and thereby further reduced the chances for developing countries to catch up [58, 59]. These findings were further developed by Prebish in a preparatory report for the UNCTAD 1 conference [60] which provided the basis for the joint position of the G77 countries at that conference [25].

3.2 Low welfare

Another strand of the development literature treats low levels of welfare as the main distinguishing characteristic between developing and developed countries. Development economists of the post-war era, such as Albert Hirschman, W. Arthur Lewis, Harvey Liebenstein, Gunnar Myrdal, Ragnar Nurkse and Paul Rosenstein-Rodan, focused mainly on average GDP and national income per capita as indicators of welfare, seeing GDP as a key factor for the achievement of the various other determinants of people’s wellbeing [61]. Over time, other aspects of welfare were increasingly considered, such as poverty, undernutrition, health, education, inequality, and access to public services [62]. Easterly emphasizes the protection of individual rights, democratic participation, political stability and peace as important aspects of welfare [63]. Sen developed an overarching approach, defining development as a process of expanding freedoms, where freedom represents the totality of people’s capabilities, with capabilities meaning the various things which people want and are able to achieve within the constraints of their economic, legal, cultural, social and political environment [64].

The perception of developing countries as countries with low welfare is prominently reflected in the Millennium Development Goals [65, 66]. The World Bank uses income as the underlying criterion for the classification of countries by development status (see above); and Sen’s capability approach serves as the conceptual basis for the HDI and hence for the DSCS of UNDP presented above.

3.3 Early system stage

A third strand of literature views developing countries as countries that lag behind in a fundamental transition from a pre-modern, agrarian, autocratic society towards a modern, industrial, services oriented and democratic one. A first example of this transition was observed in the United Kingdom, which took off in the 18 th century. Other examples followed in the rest of Europe, in Northern America, in Australia, New Zealand, and later in Japan [62, 67].

Rostow describes in detail the substantial changes taking place in the course of this transition in the demographic, social and economic spheres. In the early stages, secularism leads to a growing importance of science in people’s world view and accelerated scientific progress enabling technological innovations and thereby paving the way for a rapid increase of productivity. Entrepreneurial activity increases and people accumulate savings that become available for large-scale investments in infrastructure and machinery. Consumption patterns change, and production shifts from agriculture to manufacture, accompanied by a growing differentiation of products. In the demographic sphere, life expectancy rises, people increasingly live in cities and increasingly participate in the public affairs [68]. In a similar vein, Kuznets describes the drivers and effects of what he refers to as “modern economic growth”, a type of persistent high growth which emerged as a result of “epochal changes” in Europe, leading to a resolute application of science to problems of economic production. He traces in detail the effects of that growth on the population structure, the structure of production and the distribution of income [69].

Different theories deal with the different transformation processes observed in different domains. Classical Growth Theory, particularly the models of Solow and Swan and the Ramsey Model, show how a rising propensity to save and technological innovations foster long-term economic growth [70]. The New Growth Theories point out the self-reinforcing effects of increases in output, fuelled by spill overs in investment into knowledge and know-how [71, 72, 73].

The theory of structural transformation demonstrates how economic growth is accompanied and reinforced by changes in the structure of consumption and production. These changes consist in a shift from the primary sector first towards manufacturing and later towards services, and in a rising differentiation within economic sectors [69, 74, 75, 76, 77]. The theory of structural transformation has provided the basis for the current DSCS of UNIDO (see Section 2.3).

In the demographic sphere, development reflects a demographic transition characterized, in an early phase, by rising life expectancy, mainly due to improved medical and nutritional conditions, and, in a later phase, by a declining birth rate, presumably as a result of changing cultural factors. Between these phases, population growth surges [78, 79, 80].

Parsons deals with changes in social organisation and culture resulting, according to him, from an evolutionary process of copying and improving. In Parsons’ view, the social systems of countries face pressure to adapt to changes in their environment, like organisms in nature. Certain modes of social organisation, proven efficient in some countries, were copied by others and developed further, so that standard modes of social organisation have emerged which Parsons refers to as “evolutionary universals”. These comprise, in the earlier phases of development, social stratification, i.e. a growing differentiation of status across socio-demographic groups, and cultural legitimization, i.e. an institutionalized cultural self-definition of a country’s society as a we-group. In the later phases, countries also developed bureaucratic organisations, money and markets, generalized universalistic norms, for example in the form, of laws and formal rules equally applicable to all community members, and democratic association, ensuring that governance is carried out by elected leaders and that policies are supported by the large majority of society [81].

Barder proposes a less deterministic view, seeing development as an open-ended process of continuous systemic adaptation and coevolution of the agents in a country. He suggests dealing with countries as cases of “complex adaptive system” explored recently in Physics and Biology. Those systems are characterized by a multitude of interactions taking place among a high number of agents, where each agent continuously adapts their behaviour to that of the others. With each round of adaptation, the system produces a new outcome. The form of this outcome is difficult to predict, due to the complexity and high number of the interactions. For Barder, this process of iterative optimization is “development”. He thus sees development as an “emergent property” of a country’s society. He claims, “the countries we call ‘developed’ have experienced a largely spontaneous rapid change to a more complex, self-organized system which does a better job of supporting the capabilities of their citizens” than the systems of less developed countries [82].

4.1 Indicators

Above, three broad concepts have been found to be transmitted by social science literature ascribing different attributes to the category ‘developing country’. One focuses on difficult starting points in the post-World-War II era, one on low welfare, and one on an early stage in systemic transition. These three concepts are certainly not independent from one another. For example, starting points refer to the origins, early system stage to the means, and welfare to the outcome of development. One might take a broad perspective and combine different concepts of a developing country, while others may consider only selected attributes.

Out of the numerous attributes associated with the three concepts above, for the purpose of this study the focus has been restricted to those measurable for a respective number of countries in the late 1970s and today, based on available data. The selection of indicators to measure these attributes has been guided by the aims to reflect a broad spectrum of each concept and at the same time to dispose of data for a high number of countries in the 1970s and today. The latter criterion considerably constrained the scope of indicators, especially due to low availability of data for the 1970s. It has thereby also reduced the influence of subjective choice in the selection process.

As shown above, some DSCSs use specific development indicators as classification criteria. Measured correspondence can be expected to be especially high in those cases. It is important for the interpretability of the empirical results below that such coincidence between classification criteria and indicators used for the evaluation of correspondence is solely attributed to the fact that the selected indicator actually represents a meaningful and sensual measure of the given concept of development as reflected in the reviewed literature, and not to an outcome of a subjective decision. To ensure maximum objectivity in the selection of indicators, the methods by which the analysed classification schemes have been built were tried to be ignored as much as possible, and the selection was made before the in-depth study of the classification methods. However, as the selection was made by a human, subjectivity cannot be perfectly ruled out.

Table 4 presents the indicators selected and the sources from which their observations have been retrieved. “Extractive colonialism” and “location in tropics” have been chosen as indicators of difficult starting points, as these have been mentioned in the literature as major external factors that hinder development. ‘Extractive colonialism’ is a dummy variable that identifies countries colonialized by a Western European empire in their past, as recorded in the GeoDist database of CEPII [83], except the United States of America, Canada, Australia and New Zealand, as in these countries colonialism was more characterized by European settlement and less by resource extraction than in the other colonies (see above, Section 3.1). ‘Location in tropics’ is a dummy variable which identifies countries located between the Tropics of the Cancer and the Tropics of the Capricorn, at least with a part of their territory.

The “primary sector share”, the “fertility rate” and a “representative government index” have been chosen as indicators for being in an early system stage. The “primary sector share” serves as an indicator for the advancement in structural transformation. It measures the proportion in value added not generated by the services or manufacturing sector, according to the International Standard Industrial Classification of All Economic Activities, Revision 3 (divisions 15 to 37 and 50 to 99). The data for that indicator have been taken from UNCTADstat [33]. The “fertility rate”, an indicator of the stage reached in the demographic transition, is defined as the number of children who would be born per women if these live until the end of their child-bearing years. The data have been taken from the WDI database [84]. The “representative government index” indicates to which extent a country’s social system relies on democratic association, one of the evolutionary universals emerging during the modernization process, according to Parsons [81]. It is a composite index constructed by International IDEA based on valuations of the following features: free and fair elections; equal and universal voting rights; existence for free political parties; national representative government offices being filled through elections [85, 86].

“Income per capita”, “life expectancy” and a “fundamental rights index” have been used as indicators for welfare. “Income per capita” is widely recognized as a basic, yet imperfect, indicator of welfare, as it serves as an important means for the satisfaction of people’s needs. It is measured as gross national income per capita in current United States dollars. Cross-country differences in prices cannot be taken into account, as purchasing power parties are not available for the 1970s. The data are taken from UNCTADstat [33]. “Life expectancy” depicts an important dimension of welfare, namely the ability to lead a long and healthy life [44]. It is measured as life expectancy at birth, as recorded in the WDI database [84]. The “fundamental rights index” refers to another dimension of welfare, namely the enjoyment of civil rights and respect of dignity, emphasized for example by Easterly. It is a composite index constructed by International IDEA from information about access to justice, civil liberties and social rights guaranteed in the form of basic welfare provisions [85, 86].

From the five sources above we construct an indicators dataset (provided in Table A.2 in the Appendix) with observations of 153 countries in the eight indicators at two different points in time: 1977 and 2017. To reduce skewness in the distributions, income per capita is measured in logarithms, and the primary sector share, the representative government index and the fundamental rights index are transformed into logits, where the logit of indicator x is given by log ⁡ ( x / ( 1 - x ) ) .

The countries in the dataset make up for 99.6 percent of the world population in 2017. They represent more than 98 percent of the population of each world region, except Oceania the data account for 94 percent (see Table 5).

The unit of analysis is a country as it exists today, i.e. in 2017. Sub-territories are not considered, as the DSCSs analysed in this study have been designed to classify entire countries only. Some countries did not exist in their present form in the 1970s. They formed part of a larger state which later split, such as the former Republics of the Soviet Union, or they were autonomous parts which later unified, such as the former German Democratic Republic and the Federal Republic of Germany. To assign a single observation of development status to each unit of analysis in 1977, in the first case, the development status of the former country has been applied to each single country existing today, and, in the second case, only the largest of the unified countries, in terms of population size, has been considered.

As outlined above, each concept is represented by several indicators (Table 4). To aggregate them into a single measure, a composite index, or “factor”, has been constructed as the linear combination of the associated indicators that drives the maximum of their pairwise correlations, using principle component analysis.3 In 2017, the factor of the starting-point concept accounted for 85 percent of the total indicator variance; the factor of the welfare concept for 79 percent, and the factor of the system stage concept 66 percent (see Table 6). In 1977, the welfare and the system stage factors were equally strong, both accounting for slightly more than 70 percent of the variance of their related indicators. Apparently, over time, the system-stage indicators have become more independent from one another, whereas the correlation among the welfare indicators has increased.

It is noteworthy that in both years the factor constructed from all indicators combined has a smaller eigenvalue than any of the factors constructed from a subset of these indicators representing an individual concept. Correlation is thus higher within concepts than between them. This confirms the validity of the identified concepts.

4.4 Measuring the relation between development status and development levels

Three measures will be used to analyse the relation between development status and levels.

(1) A basic measure of correspondence between development status classification and development levels consists in the proportion of countries classified as developing that are actually found on the higher ranks in the distribution of a given attribute associated with a developing country. When n 1 countries are classified as developing, we divide the number of developing countries found among the n 1 countries with highest indicator values by the total number of developing countries. If we use development status as a measure of development, the so constructed indicator can be referred to as the positive predictive value (PPV), a widely used measure of the precision of predictions of categorical outcomes [90]. The PPV can take values between one and zero. One represents a situation in which no developing country has a lower indicator value than any non-developing country; zero represents a situation in which no developing country has a higher indicator value than any non-developing one.

(2) While the PPV is an illustrative and intuitive measure of the correspondence between development status and levels, it is also a fairly general one, as it does not reveal any information about the relation between the two among the higher and among the lower ranks of the indicator distribution. A measure of correspondence that takes the entire range of indicator values into account can be derived from the goodness of fit of a logistic regression of the classification outcome over a given development indicator:

log ⁡ ( p i 1 - p i ) = α + β ⁢ x i(1)

where p i is the probability for country i to be classified as developing and x i is the development indicator, normalized to zero mean and unit variance. x i may also represent a factor derived from a set of indicators (see Section 4.3 above).4 A common measure for the goodness of fit is the log likelihood ratio (LLR) which depicts the relative gain in (logarithmic) likelihood when classification probabilities are predicted using the maximum likelihood estimator of β rather than assuming β to be zero. It follows a chi-square distribution. With one degree of freedom, in inferential statistics, the model above is considered significant at the 0.1 percent level when the LLR exceeds 10.8.

(3) While the PPV and the LLR measure correspondence, another interesting property of a classification scheme is the degree to which the classes it defines are homogeneous (see Section 2.1 above). Therefore, within-class heterogeneity of the developing countries, relative to the heterogeneity observed among all countries of the world, is also reported. Heterogeneity is measured by the mean squared Euclidean distance (MSED), known from cluster analysis, which represents the average distance of a set of countries from their centroid in a coordinate system of indicators, where the centroid marks the point of a virtual country with average values in all indicators. The greater the MSED the more the countries are spread throughout the coordinate system; the smaller the MSED the more they are clustered around the centroid. The MSED of the entire set of countries is given by

D = 1 n ⁢ ∑ i = 1 n ∑ k ( x k , i - μ k ) 2 ⁢ with (2) μ k = 1 n ⁢ ∑ i = 1 n x k , i

and the MSED of a subgroup c by

D c = 1 n c ⁢ ∑ i = 1 n c ∑ k ( x k , i - μ k , c ) 2 ⁢ with (3) μ k , c = 1 n c ⁢ ∑ i = 1 n c x k , i

where k is the identifier of the indicators. Note that with only one indicator the MSED is equivalent to the variance.

Like with the variance, the within-group MSEDs of all groups, weighted by their respective group size, add up with the between-group MSED to the overall MSED, regardless of the number of indicators considered:

D = D b ⁢ w + ∑ c n c n ⁢ D c ⁢ where (4) D b ⁢ w = ∑ c n c n ⁢ ∑ k ( μ k , c - μ k ) 2

Below, the within-group MSED ( D c ) of developing countries is presented as a ratio to the total MSED ( D ) calculated for the world as a whole.

It is worth pointing out for the interpretation of the empirical results that a high correspondence does not necessarily mean low relative intra-class heterogeneity. For example, developing countries may on average have higher indicator values than developed, but these may be spread over a wide range; or the developing countries may form a narrow cluster in the coordinate system, surrounded by developed countries.

5.1 Overall patterns

The Tables 8–9 show different measures characterizing the empirical relation between development level and development status for the seven DSCSs of international organisations reviewed above, in 1977 and 2017. Development levels are measured by different indicators, including composite indices (factors) for entire concepts. To help distinguish the effect of changes in indicators from the effect of revisions in classification schemes, the current version of each classification scheme is applied to data observed in the past and present.

OPEN IN VIEWER

The numbers in Tables 8 and 8 display the degree of correspondence. Figure 4 visualizes the patterns in 2017. All in all, correspondence between development levels and classification outcome turns out to be strong for all indicators considered and under all DSCSs analysed. For the year 2017, the PPV is in all cases above 0.8, indicating that more than 80 percent of the countries classified as developing also rank highest in the analysed attributes associated with a developing country. And the LLR obtained from logistic regressions of development status over development level is consistently above its critical value of 10.8, meaning that measured development levels have a significant impact on the probability that a country is classified as developing. In many cases, the LLR reaches levels even higher than 50. The results obtained with the PPV and the LLR appear fairly robust. When relatively high correspondence is reported by one measure, it is mostly assessed high also with the other.

It is striking that correspondence often turns out stronger when measured for factors representing an entire concept than for individual indicators. This indicates that the assumed latent variable that drives the correlations between indicators jointly representing a concept can explain better the classification than the individual indicators on their own. This finding is in line with a view that development is multidimensional.

We can also note, as a general trend, that changes in the distribution of indicator values have led to a reduction of correspondence for some indicators, such as the fertility rate, but to an increase in correspondence for other indicators, such as access to fundamental rights. Revisions of classification schemes have mostly been reflected in a correspondence increase.

The numbers in Table 9 display the within-class heterogeneity among developing countries, as defined by the different schemes, relative to the world as a whole. They suggest that, in general, classification by development status yields groups of developing countries that are more homogeneous group than the countries of the world on average. Only for the fertility rate and the representative government index, in 2017, heterogeneity among developing countries was higher than throughout entire world, under most analysed classification schemes. This had not been the case in 1977.

It appears that the overall decrease in correspondence of classification outcomes with the fertility rate over time, outlined above, has gone hand in hand with a growing heterogeneity among developing countries with regard to that indicator. By contrast, the increase in correspondence observed for access to fundamental rights is not reflected in an overall decrease in heterogeneity. It is striking that changes in the distribution of income per capita and of the primary sector share have led to growing homogeneity among the developing countries under all analysed DSCSs. Revisions in classification schemes also have mostly had a positive effect on developing countries’ within-group homogeneity.

Let us in the following examine the characteristics of the individual DSCSs in more detail. The DSCS of

UNCTAD stands out as having a relatively close correspondence with the difficult starting points concept. This is indicated by a PPV of 0.92 and an LLR of 117. Location in the tropics appears to be especially strongly reflected that scheme, past extractive colonialism to a slightly lesser extent. None of the other assessed DSCSs of international organisations matches better the difficult-starting-points concept than the UNCTAD DSCS. And the correspondence of UNCTAD’s scheme with difficult starting points is much higher than its correspondence with the early-system-stage and low-welfare concepts. We also find that developing countries as defined by UNCTAD form a relatively homogeneous group with respect to difficult-starting-point indicators, evidenced by an MSED among developing countries only half as high as the MSED observed throughout in the entire world.

Correspondence of the UNCTAD DSCS with the early system stage and the welfare-based concepts is about equally weak, in comparison to the difficult starting points concept. The factors representing these concepts record a PPV of only 0.86 to 0.87 and an LLR of around 70. This higher correspondence goes hand in hand with a greater heterogeneity observed with the system-stage and welfare-based than with the starting-points concept. Especially high heterogeneity can be found for the system-stage concept, apparently due to a high variance of the fertility rate and the representative government index. It is worth noting that the spread of the fertility rate among UNCTAD developing countries is equally large as among the world as a whole and the spread of the representative government index even larger than the world total.

Looking at changes over time, overall, the DSCS of UNCTAD reflected better cross-country differences in development levels in 1977 than today. Over time, the PPV calculated for the factor representing the indicator total has decreased from 0.94 to 0.92 and the LLR reduced from 120 to 113, while overall within-class heterogeneity has risen, from 66 to 74 percent of total world heterogeneity. The strongest drops in correspondence and the strongest increases in heterogeneity can be observed for the fertility rate and life expectancy, independently from the correspondence measure used. By contrast, differences in the protection of fundamental rights and in government representativeness are better reflected in UNCTAD’s classification today than in the past.

The effect of revisions of the UNCTAD DSCS on the correspondence between development status and development levels has been overall positive. Except for extractive colonialism, all indicators measured in 1977 match the UNCTAD classification of 2017 more than the classification of 1964. In the case of the primary sector share, this positive effect on correspondence has offset much of the negative effect of changing development levels.

The UNSD scheme, not surprisingly, initially showed similar characteristics as the UNCTAD scheme: a relatively close correspondence with the difficult-starting-points concept and a relatively loose correspondence with the early-system-stage and the low-welfare concepts. This has changed over time. The revisions of 1996 and 2018 have considerably weakened the correspondence with location in the tropics and extractive colonialism and strengthened the correspondence with most other indicators observed in the 1970s, as the PPV and the LLR unambiguously show. In turn, within-class heterogeneity has increased considerably for the difficult-starting-points concept and reduced slightly for the early-system-stage and the low-welfare concept. These results could be expected, given that the former socialist countries in Asia were added to the developing countries and considering that these are located far in the North, were not colonized by European empires and, for many indicators, showed values relatively close to the majority of developing countries.

For life expectancy and the fertility rate, over time, the gains in correspondence due to revisions have been more than offset by changes in the distribution of indicator values, resulting in a certain assimilation between developing and developed countries. It is striking that this offsetting can be observed with all three correspondence measures applied. By contrast, the changes in the distribution of all other indicators have led to an increase in correspondence over time. In 2017, the UNSD DSCS matches all three identified development concepts to a certain degree. The within-class heterogeneity of developing countries is in general relatively high for all three concepts applied.

5.4 The IMF and WTO classification schemes

The IMF and WTO DSCSs are the ones showing the closest correspondence with the system-stage concept among the analysed schemes. With a PPV of around 0.95 to 0.96 and an LLR of 106, measured for the two classification schemes, the factor representing system stage appears as a strong predictor of classification probability. Correspondence is even higher to the welfare-based concept, especially under the scheme of the IMF, although this cannot compare with the exceptionally high correspondence of the World Bank scheme in the welfare domain.

The IMF DSCS can be considered as the one, out of the seven, that reflects best the entirety of the identified concepts of a developing country. If we base our understanding of a developing country on the common factor that drives the indicators associated with all three concepts, the IMF DSCS reflects that factor best, with 99 percent of the declared developing countries represented on the highest ranks with reference to that factor and an LLR of 143. Nevertheless, the IMF scheme, like the WTO scheme, does not lead to particularly low within-class heterogeneity of developing countries. Looking at changes over time, the relatively far-reaching revisions implemented by the IMF (see Section 2) have led to an increasing correspondence with the starting-points indicators, but to a reduction in correspondence with other indicators, especially the representative government and the fundamental rights indices.

5.5 The World Bank classification scheme

The World Bank DSCS is the one most closely linked with the welfare-based concept, especially with income per capita. Its correspondence with the system-stage concept is slightly weaker than the IMF’s, and difficult starting points are reflected by the World Bank scheme less than by any other scheme. As expected, considering that income per capita is used as the sole classification criterion, the PPV reaches one for this indicator, and also the LLR turns out exceptionally high. The World Bank DSCS is also highly correlated with other welfare indicators. This supports the view, taken by the early development economists, that income is a key determinant of the various other dimensions of welfare (see Section 3.2).

It is noteworthy that the correspondence of the World Bank scheme with measured development levels has not always been that high. The initial classification from 1978, which was not purely based on income, less powerfully reflected the various concepts of development, also in comparison to the other schemes in place at that time.

5.6 The UNIDO classification scheme

Like the schemes of the World Bank and the IMF, the UNIDO DSCS also shows a stronger correspondence with the welfare-based than with the system-stage concept. This finding may appear counterintuitive, considering that reflecting disparities in industrialization, a system-stage attribute, has been the underlying aim of that scheme. The scheme’s limited correspondence with the primary sector share compared to other schemes, indicated by a PPV of 0.90 and an LLR of 51, can be explained by the fact that countries with relatively small primary sectors have not necessarily developed relatively large manufacturing sectors. Their economies may be dominated by services. Examining the data, this is often the case, for example, in Lebanon, Cyprus, Greece, Mauritius and Costa Rica; all countries classified as “emerging industrialized or other developing economies” by UNIDO.

The low correspondence of UNIDO’s DSCS with the other system stage indicators may cast doubt on whether industrialization, in the way measured by UNIDO, effectively forms part of the transition process from developing to developed countries as the literature suggests. In an alternative view, one may question that a direct structural transformation of the primary sector into services, skipping the expansion of manufacturing, which is not in line with the theory of structural transformation, should be considered as development.

It is striking that UNIDO’s initial DSCS from 1988, applied to data from 1977, showed a stronger rank-correlation with and stronger impact of system-stage indicators than UNIDO’s current classification scheme applied to data from 2017.

5.7 The UNDP classification scheme

The UNDP DSCS represents a fourth scheme which is more strongly related with the welfare-based concept than with the other identified concepts of underdevelopment. This is not surprising, given that it is grounded in the HDI, a composite welfare indicator (Section 2.3). Correspondence is particularly strong with income per capita and life expectancy, the two variables considered in this analysis which also form part in the HDI. It is noteworthy, however, that correspondence with other welfare dimensions, such as life expectancy and protection of fundamental rights, is weaker than that of the World Bank scheme. Apparently, years of schooling, the third dimension covered by the HDI, dampens the correlation of the index with those indicators.

6.1 Different classification schemes fitting different concepts

This study has also revealed substantial differences between the DSCSs applied by international organisations in the past and today. Over the decades, two different generations of schemes have evolved. Under the DSCSs of the first generation, development status classes are formed by nomination of countries rather than by application of specified criteria. Political considerations, expert judgement and countries’ self-identification seem to have played a major role in that nomination. The lack of grounding on specified criteria characterizing the first generation has been addressed in the DSCSs of the second generation introduced by the World Bank, UNIDO and UNDP after the late 1980s. Over time, the classification of countries into developing and other countries has become increasingly heterogeneous throughout international organisations.

The empirical results above show that the different DSCSs of international organisations have their individual characteristics, in the sense that each scheme fits better some concepts of a developing country than others. The schemes of the first generation, still applied by UNCTAD, UNSD and the IMF, show a relatively strong correspondence with a concept that sees developing countries as countries faced with difficult starting points, as a consequence of their colonial history and their location in the tropics. Correspondence with difficult starting points is especially strong for the UNCTAD DSCS. By contrast, the schemes of the second generation match best with a concept that sees developing countries as countries with low welfare. A third concept identified in this study, which focuses on the developing countries’ early stage in systemic transition, is best reflected by the DSCSs of IMF and WTO, among the seven schemes compared. The IMF DSCS is also the one which matches best a broad concept encompassing all attributes of a developing country measured in this study. The UNSD classification can be regarded as a common denominator which takes account of all three identified concepts to a certain degree.

6.2 Quality aspects of development status classification

Contrasting the analysed DSCSs with suggested quality criteria for classification schemes (see Section 2.1) reveals some points that deserve attention. Firstly, while classification schemes are recommended to be comparable, the DSCSs used by international organisations today differ considerably in the way the classes have been formed. Some are based on specified criteria, others not. Among the former, the applied criteria are different. As a result, the composition of development status classes differs across schemes, and these differences have increased over time (see Section 2). This heterogeneity in DSCSs negatively impacts on the comparability of the statistics to which they are applied. The figures in a dataset which are based on one DSCS cannot be fully compared with the figures in a dataset which are based on another. Using similar labels for categories defined in different ways may cause misunderstanding and false interpretation of the data. That these differences can hinder productive discourse and scientific progress has been pointed out by Nielson [3]. A harmonization of the DSCSs would enhance users’ possibilities to combine international official statistics from different sources for the purpose of their analyses, and to compare and discuss their findings derived from these data.

Secondly, while it is recommended that DSCSs are well described, among other materials by explanatory notes, coding indexes, manuals and correspondence tables, the documentation supporting the first generation of DSCSs appears rather sparse. It is usually limited to country lists and coding indexes. Explanations of the motivation, let alone of the applied criteria, for the nomination of countries for categories are often lacking. The DSCSs of the second generation are generally more comprehensively explained than those of the first, by notes on the organisations’ websites alongside the data, by statistical annexes and by dedicated papers. However, information on the correspondence between the categories of different DSCSs is scarce, although this information is essential for interoperability. Apart from a sketch of the linkages between broadly defined country groups developed by the World Bank in the 1980s [36], to the best of the author’s knowledge, the correspondence table in the Appendix, that shows the assignment of the individual countries to categories under the different schemes, is the first of its kind.

Thirdly, it is recommended that classification schemes “reflect the realities of the field” to which they relate and “look valid” to users. In the case of DSCSs, this condition is complicated by the fact that development, and thus development status, is not unambiguously defined, and people have different ideas about its meaning. Different users of data may thus expect from a DSCS to reflect different aspects of reality. A DSCS will look the more valid to users the more its classification criteria match their concept of a developing country. Recalling the discourse in the introduction, it can be noted that writers to whom DSCSs do not anymore look as valid today as in the past, such as Gates [4] and Khokhar and Serajuddin [5], point to developments in different indicators than those who assume a continued validity of DSCSs, such as UNCTAD’s Division on Globalization and Development Strategy [6]. The former put the main focus on income per capita, poverty and fertility rates, the latter on levels of industrialization, infrastructure development, conditions of work and digitization. The findings above suggest that all analysed DSCSs are empirically linked with development levels. They all can therefore be considered to reflect the realities in the field, with each DSCSs emphasising different aspects of that reality. A decrease in correspondence and within-class homogeneity can be observed for some indicators under some DSCSs, while in other cases an increase can be observed.

To cope with cases in which within-class heterogeneity has become high, users of international statistics may sometimes require finer granularity than the distinction between developing and other development status classes. Several more narrowly defined categories than developing countries are already widely used, such as least developed and landlocked developing countries [91], small island developing States [91, 92] and highly indebted poor countries [93]. However, these have not been formally incorporated into existing DSCSs.

Fourthly, it is recommended not to change a classification scheme too frequently. The second generation DSCSs show a high stability in the sense that the underlying classification criteria have not been revised. However, as the characteristics of countries do not remain the same, inevitably, the composition of the classes changes relatively often. In a user survey carried out by the World Bank, frequent changes in the composition of the classes of the Bank’s scheme have in fact been raised as a point of concern [41]. By contrast, the first generation DSCSs have been subjected to only occasional revisions in the composition of the groups. They have thus proven more stable in the composition of classes than the schemes of the second generation (see Fig. 3). However, the imposed constancy in the composition implies that the average characteristics represented by the classes have changed over time, in tandem with countries’ progress in development. Only the characteristics associated with difficult starting points have remained stable, by definition, what can be seen as the main reason for the continued high correspondence of UNCTAD’s DSCS with the starting-points concept.

Fifthly, “statistical convenience” has been mentioned as a desirable criterion of DSCS in disclaimers made in United Nations documents (see the introduction). Statistical convenience may consist, for example, of a high congruence between development status groups and geographic regions or formal associations of states, such as the European Union, the OECD or G77, as this may facilitate the representation of grouped and aggregated data in disseminated tables. Spelling out the specific aspects to make a DSCS “statistically convenient” would likely increase the credibility of that scheme from the viewpoint of users.

6.3 Outlook

Based on the findings of this study, DSCSs do not appear obsolete. Development status classes still appear useful as yardsticks of cross-country differences in attributes widely recognized as characteristics of a developing country. Different schemes emphasize different attributes in that measurement. The study has also revealed deficiencies in development status classification as applied at present. DSCSs would arguably do a better job if they were harmonized, or if at least their interlinkages were better documented and communicated than they currently are. The DSCSs of the first generation lack sufficiently comprehensive metadata explaining the methods applied in the formation of classes. This reduces clarity and interpretability of the statistical output.

By introducing appropriately documented objective classification criteria, the appearance of the second generation of DSCSs can be seen as a favourable development for statistical quality. It is striking that these schemes reflect development mainly from the welfare-based perspective, whereas the system-stage and starting-points concepts are better reflected by the DSCSs of the first generation. A DSCS that is based on objective characteristics and at the same time takes full account of other dimensions of development than welfare appears as a gap which calls for being filled. Welfare is not the most relevant or the only relevant aspect for the various topics of research related to development. For many types of analysis, the countries’ historic and geographic preconditions and their advancement in the transformation of the economy and society play a role for a useful categorization of countries. If development studies could rely on a more objective differentiation of countries by development status than today, this would enhance the clarity and interpretability of their findings.

The United Nations’ 2030 Agenda for Sustainable Development [94] has opened a new chapter in the research and reporting on development. A striking difference to its predecessor framework, the Millennium Development Goals [65], is the fact that the Sustainable Development Goals are meant to apply to all countries of the world, not just to the developing countries – how ever defined. By considering a wide range of goals, encompassing social justice, peace, ecology, global partnership and other domains, the 2030 Agenda takes a broader focus than the Millennium Development Goals most of which were welfare targets. The setup of the 2030 Agenda may be taken as an occasion to consider new classification schemes for development status that would incorporate other concepts associated with development than those identified in this study. Development could then be seen not only as a matter of raising welfare or making progress in the transformation of the production structure, demographic reproduction patterns or social organisation. Development that deserves the label “sustainable” should then also encompass characteristics like the degree to which human activity causes depletion of natural resources. Accordingly, the system-stage concept proposed above could be broadened to cover the transition to a greener economy, and the welfare-based concept be extended to cover attributes such as exposure to environmental degradation and natural disasters. Theoretical foundations for such broader concepts of development have already been laid, for example with the writings about the environmental Kuznets curve [95, 96, 97]. New indicator frameworks, adapted to that broader focus, have also already been proposed, such as the “Happy Planet Index” [98] and different types of “Sustainable Development Indices” [99, 100]. All in all, it appears that fifty years after their introduction, DSCSs remain a valid element of international official statistics, worth being adapted to changing economic realities and changing statistical needs.