The investment development path and human development: Is there a nexus?

Djokoto, Justice Gameli Article The investment development path and human development: Is there a nexus? Research in Globalization Provided in Cooperation with: Elsevier Suggested Citation: Djokoto, Justice Gameli (2022) : The investment development path and human development: Is there a nexus?, Research in Globalization, ISSN 2590-051X, Elsevier, Amsterdam, Vol. 4, pp. 1-10, https://doi.org/10.1016/j.resglo.2021.100079 This Version is available at: https://hdl.handle.net/10419/331008 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. https://creativecommons.org/licenses/by-nc-nd/4.0/ Research in Globalization 4 (2022) 100079 Available online 21 December 2021 2590-051X/© 2021 The Author(s). Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/bync-nd/4.0/). The investment development path and human development: Is there a nexus? Justice Gameli Djokoto Central University, Ghana ARTICLE INFO Keywords: Human capital Human development HDI Generalized estimation equations Investment development path Panel data ABSTRACT To investigate the nexus between the investment development path (IDP) and human development, we modelled the human development (human development index, HDI) using generalized estimation equations and accounted for the unit interval property using non-linear link functions. The gross domestic product per capita was used as the proxy for the IDP. Using panel data of 2,568 observations from 135 countries for 1990–2019, we found a strong and positive effect of IDP on HDI. Thus, there is a nexus between IDP and human development. The 4% elasticity of the IDP, confirms the broader scope of the HDI than the IDP in capturing development. Coupled with the 60% correlation between the IDP variable and HDI, aside from serving as a proxy for each other, these could also be instruments for each other provided they are uncorrelated with the error term in the model under consideration. Trade, infrastructure, and human capital were found to enhance human development. These offer policy options for enhancing human development. Introduction Measuring development and categorizing countries based on the measures of development have gained the attention of economists. Djokoto et al., (2017); Djokoto et al., (2016); Dunning (1982), Dunning (1986), Dunning (1993), Dunning and Narula (1996); Narula (1996) and Ramalho et al., (2011) have used the plot of net outward foreign direct investment (NOFDI) and gross domestic product per capita (GDPPC) or gross national product (GNP) to identify five stages of development, namely, stages I, II, III, IV and V. Described as the investment development path (IDP), stages I and II were assigned to developing countries, stage III was theorized for economies in transition and stages IV and V were linked to developed countries. Countries could move forward and backwards within these five stages of development over time (Djokoto, 2021a; Gorynia, Nowak, & Wolniak, 2010a; Gorynia, Nowak, & Wolniak, 2010b; Gorynia, Nowak, Tarka, & Wolniak, 2012; Iacovoiu & Panait, 2014a). The IDP has found application in assessing competitive advantages of countries (Djokoto, 2021a; Dunning & Narula, 1994; Duran & Ubeda, 2005; Iacovoiu & Panait, 2014b); pace and pattern of economic development (Dunning, Kim, & Lin, 2001; Tchorek, 2016; Trąpczy´ nski, Gorynia, Nowak, & Wolniak, 2019; Voica, Panait, & Radulescu, 2020) and others. The IDP has, however, been criticized. Not only that it has two indicators, but these are also financial indicators, and it is less humanistic (Djokoto, 2021b; Sen, 1975; Stewart, Ranis, & Samman, 2018). A more humanistic measure of development is the human development index (HDI). Published by the UNDP, the HDI is composed of a wealth indicator, gross national income (GNI) per capita, knowledge indicators (mean years of schooling and expected years of schooling) and a long and healthy life indicator (life expectancy at birth) (Djokoto, 2021b; UNDP, 2020). The HDI is thus, the geometric mean of the normalized indices of each of the three dimensions (UNDP, 2020). The HDI, which is bounded between 0 and 1, has been classified into four categories of human development: low, medium, high, and very high. The HDI has informed industrial policy (Ferrannini, Barbieri, Biggeri, & Di Tommaso, 2021); energy policy (Acheampong, Erdiaw-Kwasie, & Abunyewah, 2021; Yumashev, ´ Slusarczyk, Kondrashev, & Mikhaylov, 2020); tourism policy and competitiveness (Croes, Ridderstaat, & Shapoval, 2020; Croes, Ridderstaat, Bąk, & Zientara, 2021); knowledge diffusion (Asongu & Nwachukwu, 2018); and climate justice (Alves & Mariano, 2018) among others. As the HDI measures the more intrinsic aspects of human development than the IDP, what is the direction and extent of the relationship between the IDP and the HDI? Some studies have investigated the determinants of HDI (Afoakwa, 2016; Arisman, 2018; Butar & Rahmanta, 2021; Ganiyu, 2016; Handalani, 2018; Kolster, 2015; Sofilda, Hermiyanti, & Hamzah, 2015). Whilst the analyses in these studies did not take account of the unit interval properties of the HDI, they did not investigate the role of IDP in HDI. E-mail addresses: [email protected], [email protected]. Contents lists available at ScienceDirect Research in Globalization journal homepage: www.sciencedirect.com/journal/research-in-globalization https://doi.org/10.1016/j.resglo.2021.100079 Received 14 October 2021; Received in revised form 15 December 2021; Accepted 16 December 2021 Research in Globalization 4 (2022) 100079 2 Djokoto (2021b) however, accounted for the fractional property of the HDI as well as the effect of IDP on HDI. IDP was found to have a weak relationship with HDI. But the countries used in the study were 23 developing countries. As the IDP is established for a country over time, this limited the number of countries resulting in cross-sectional data. Further, the stages of the IDP were assigned numeric values with the highest not exceeding 5. Whilst these served the purpose of the study of Djokoto (2021b), not only did the number of countries reduce the efficiency of the estimates, but the range of the numeric values also reduced the variability in the IDP variable. The link between IDP and HDI is important because whilst the IDP and HDI measure the level of development (Djokoto, 2021b; Dunning, 1981a; Dunning, 1981b; Dunning, 1986), the GDP used in IDP is also used in the construction of the HDI through the GNI (Djokoto, 2021b; UNDP, 2020; UNDP, 2021). In this paper, whilst accounting for the fractional properties of the HDI and including IDP to explain HDI, we used a proxy of the IDP that allowed for each country to have data for each year. This made it possible to include more countries in the data and analyze the data as a panel to reap spatial and temporal benefits. This would certainly produce more efficient parameter estimates. The outcome of this paper would provide evidence on the link between the more financial measure of development and the humanistic measure of development based on a more efficient estimate. It would also add to the indicators of measurement in the development space (Djokoto, 2021b). Further, either of the development indicators could serve as proxies and instruments for each other. In what follows, the theoretical and empirical literature presented is followed by the data and methods of analysis. The presentation and discussion of the results are in Section “Results and discussions”. The final section is the conclusions and recommendations. Literature review Theoretical review In line with the subject of the study, a review of the IDP and development theories are presented before the empirical review. The IDP relates a country’s NOFDI to gross domestic product to show the country’s level of development (Djokoto, 2021b; Dunning, 1981a; Dunning and Narula, 1996). The graphical relation, a chart or curve is marked out to show the stage of development (Fig. 1). The basis of each of the five stages; I, II, III, IV and V lies in the levels and interaction of three factors: ownership, localization, and internationalization (Djokoto, 2021b; Dunning, 1981b). In stage I, the inward foreign direct investment (IFDI) exceeds outward foreign direct investment (OFDI). Ownership advantages such as indigenous technology and limited created assets explain this. The location advantages are insufficient to entice IFDI. Internationalization from native firms is almost nonexistent. In stage II, there is an additional rise in IFDI compared to OFDI. Ownership and location advantages move beyond the levels for stage I. The development of supportive businesses for the ownership advantage and the changes in the business setting such as comparative cost of production for the location advantage explains stage II. Developing countries are hypothesized to be in stages I and II. The gradual decrease in the IFDI and increase in OFDI precipitate a fall in the excess of IFDI over OFDI marking the beginning of stage III. This is attributable to the ownership and location advantages of indigenous firms. As a result, native firms start to invest outside their borders. This stage is proposed for emerging economies. Stages IV and V are reserved for developed countries. In the former, OFDI exceeds IFDI, thus, NOFDI becomes greater than zero. This is because the ownership and location merits would have increased substantially. Foreign investments would, therefore, increase in volume culminating in internationalization by native firms. In the case of the latter, there is the backward and forward movement of OFDI and IFDI leading to fleeting peaks and troughs of NOFDI. This scenario is accentuated by the maturity of the ownership, location, and internationalization merits. Thus, no one economy has the overarching ownership, location, and internationalization merits. Multinationals tend to value their business interest above that of their country of origin owing to the high intensity of internationalization. Stewart et al. (2018) traced the evolution of development thinking in their book, Advancing Human Development: Theory and Practice. In the 1950s and 1960s, within the context of low incomes and postcolonialism and to fulfil the development desire, the main strand of developed country thinking was Keynesianism in which development was considered as economic growth. This was to be achieved through planning and industrialization (Rostow, 1960; Lewis, 1954). By the 1970s, the context changed to an observation of growth but high poverty and unemployment. Two other dominant strands joined Keynesianism, Marxism, and the Neoclassical revival in the 1970s. In response to the existing context, the thinking of Keynesianism was the need to increase employment (Seer, 1972), and the redistribution of the observed growth (Chenery, Ahluwalia, Bell, Duloy, & Jolly, 1979). Streeten (1981) and Stewart (1985) championed the provision of basic needs. Morris (1978) and Morris (1980) proposed the Physical Quality of Life Index (PQLI). Morris argued that for humans to flourish, money incomes - even for the poor - was a deficient indicator (Stewart et al. (2018)). Later, UNDP (United Nations Development Programme) (1990), noted that income was a means to an end. Marxism pointed out issues of dependency (dependencia) (Frank, 1966; Furtado, 1980). As expected of the neoclassical, they revived the role of prices/markets (Scitovsky, 1976). By the 1980s, there was an acute balance of payments and debt problems. Not to be outdone by the neoclassical, the Monetarists launched a ‘counter-revolution’ for pro-market and anti-state monetarism in macroeconomic policy (Balassa, 1969). From the late 1980s to the middle of the 1990s and 2000, the world found itself in negative or weak growth, high poverty and debt. This called for new theories of growth and trade, informational asymmetries, alternative motivations, and the strong role of institutions. The thought marked the move towards human-oriented objectives of development such as poverty reduction, capabilities, human development, and the key role of the millennium development goals (MDGs) (Sen, 1975; Ranis, 1977; Stewart, 1985; Streeten, 1981; Streeten, 1994). In the subsequent decades, there was a global agreement on the Sustainable Development Goals (SDGs) in 2015. In 1990, the UNDP published its first Human Development Report. The report combined the capabilities approach of Sen with the basic needs focused on the needs of the most deprived. The main dimensions of human development comprise ‘a long and healthy life, being knowledgeable and having a decent standard of living. The HDI is the geometric mean of the normalized indices for each of the three dimensions Fig. 1. Investment development path (Dunning and Narula, 1996). Note: Not drawn to scale – For illustrative purposes only. J.G. Djokoto Research in Globalization 4 (2022) 100079 3 (UNDP, 2021). Whilst the health is measured by life expectancy at birth, that of education dimension is assessed by mean of years of schooling for adults aged 25 years as well as the expected years of schooling for children of school entering age. The standard of living dimension is captured as the gross national income per capita (UNDP, 2020) (Fig. 2). The resulting index of human development, HDI, which ranges between 0 and 1, is categorized into four; low human development, medium human development, high human development, and very high human development (Fig. 3). In that order, countries in the very high human development group (very high HDI) have the highest human development, higher than those lower; high, medium, and low HDI. Aside from the HDI, other forms of development measurement have been published (UNDP, 2020): Inequality-adjusted HDI, Gender development index (GDI), Gender inequality index (GII) and the Multidimensional poverty index (MPI) (Fig. 4). Beyond the three categories and four indicators, Ranis, Stewart, and Samman (2006) identified eleven categories and 39 indicators. Interestingly, they found eight indicators to be highly correlated with the HDI, hence may therefore be represented by it. Under-five mortality rates performed equally as well as the HDI, and income per capita was less representative of other dimensions of human development. Using data on OECD and developing countries, Ranis et al. (2006) found that the HDI (and the other two broad indicators) were worse indicators of the extended categories of human development for OECD countries than for developing countries. Some similarities can be drawn from IDP and HDI. First, they are both indicators of development. Second, GDPPC contributes to both IDP and HDI. Thirdly, IDP and HDI are categorized into stages, the former into five stages and the latter into four stages. The two development indicators differ in the sense that whilst IDP is established based on the interaction of GDPPC and NOFDI, the HDI is an index. Whilst the IDP relies on financial variables only, the HDI employs financial plus nonfinancial variables. IDP is less humanistic than the HDI. Whilst the measure of HDI is computable annually, it is more difficult to establish the level of development using the IDP for a country on an annual basis. Whilst the similarities form the basis of a possible nexus between IDP and HDI that informed this study, the differences make the use of the HDI on the rise in recent times. Empirical review Recently, Djokoto (2021b) used data on 23 developing countries to assess the variability in HDI explained by the stages of the IDP. Modelling the HDI as a fraction, the results showed that the IDP had a weak relationship with human development. A 1% increase in IDP induced a 2% increase in human development. Djokoto (2021b) attributed the statistical significance to the presence of GDP in the construction of the HDI through the GNI. Further, just as IDP measured the level of development, HDI did the same. The weak significance was associated with the fact that GDP was embedded in GNI, but the GNI formed a portion of the HDI. Other results showed that human capital development was the strongest driver of human development and inflation worsened human development. Infrastructure, measured by the sum of fixed and mobile telephony subscriptions per 100 persons, also enhanced HDI. Some studies provided empirical evidence on the effect of the GDPPC and the control variables on HDI. These focused on provinces of Indonesia (Butar & Rahmanta, 2021; Sofilda et al., 2015), the Association of South-East Asian Nations (ASEAN) (Arisman, 2018; Handalani, 2018), North Africa (Kolster, 2015), sub-Saharan Africa (Aloui, 2019; Ganiyu, 2016; Ranjkeshan, 2021), Africa (Gohou & Soumar´ e, 2012), developing countries (De Groot, 2014) and the world (Oladapo & Abdul, 2016). The data structure used in all the studies was a panel, which ranged from 1990 (Kolster, 2015; Ranjkeshan, 2021) to 2019 (Ranjkeshan, 2021). The estimation methods have largely been fixed effects (FE) (Arisman, 2018; Butar & Rahmanta, 2021; Ganiyu, 2016; Handalani, 2018; Sofilda et al., 2015). Three studies employed the general method of moments (GMM) (De Groot, 2014; Kolster, 2015; Ranjkeshan, 2021). Without specifically addressing IDP, Arisman (2018) assessed the effect of GDP growth rate on HDI whilst De Groot (2014), Handalani (2018) and Ranjkeshan (2021) used GDPPC as a determinant (control variable) of HDI. All four studies found a strong positive effect of GDP growth rate and GDPPC on HDI. These results encompass ASEAN nations, sub-Saharan Africa and developing countries. Recognizing these countries are all developing states, the results appear to be consistent for developing countries. Our results will provide evidence on all this plus transition and developed countries. De Groot (2014), Kolster (2015) and Ranjkeshan (2021) reported a neutral effect of trade openness on HDI. Inflation measured as the growth rate of the consumer price index (CPI) had a mixed effect on HDI. Aloui (2019) reported positive effects whilst Arisman (2018), Ganiyu (2016) and Handalani (2018) found negative effects. De Groot (2014), Gohou and Soumar´ e (2012), Kolster (2015) and Ranjkeshan (2021) reported a neutral effect of inflation on HDI. Regarding the effect of population on HDI, the population had been measured as the number of males and females (quantum) (Arisman, 2018) and the annual growth rate of the number of males and females (De Groot, 2014; Sofilda et al., 2015). Arisman (2018) and De Groot (2014) found an inverse relationship between the population indicators and HDI. A neutral effect was found by Sofilda et al. (2015). Human capital had also been measured in two ways in the literature; education expenditure (Butar & Rahmanta, 2021; Oladapo & Abdul, 2016; Sofilda et al., 2015) and secondary school enrolment as a percentage of gross enrolment (Ganiyu, 2016; De Groot, 2014). Ganiyu Fig. 2. Graphical representation – Calculating the human development index Source: UNDP (2020). Fig. 3. Stages of human development. J.G. Djokoto Research in Globalization 4 (2022) 100079 4 (2016), Oladapo and Abdul (2016) and Sofilda et al. (2015) found a positive effect on HDI whilst a neutral relationship was reported for the effect of education on HDI (Butar & Rahmanta, 2021; De Groot, 2014). The effect of final government expenditure on goods and services as a percentage of GDP had a mixed effect. These were positive effects (Kolster, 2015), negative effects (Aloui, 2019) and neutral effects (Ganiyu, 2016; De Groot, 2014; Gohou & Soumar´ e, 2012; Sofilda et al., 2015). Measuring infrastructure as internet subscription per 100 persons, Ganiyu (2016) found no effect of infrastructure on HDI for sub-Saharan African countries. In the case of Kolster (2015), infrastructure was found to have a positive effect on HDI. In the case of the latter, the proxy for infrastructure was paved roads. In the literature reviewed and summarized in Fig. 5, except income and trade, all other determinants of HDI showed mixed effects. Whilst this observation calls for further study on the effect of these factors on HDI, review studies, either qualitative or quantitative would also be useful. From the foregoing, the role of IDP in HDI was unexplored by many studies that also failed to appropriately account for the unit interval properties of the HDI. Only Djokoto (2021b) considered the role of IDP in HDI. Whilst accounting for the unit interval properties of the HDI by non-linear modelling; fractional regression with a probit link function, the data was limited in size and the IDP measure was constrained to range from 1 to 5. To fill these gaps, we modelled the HDI using generalized estimation equations and accounting for the unit interval property using non-linear link functions. Also, a more flexible proxy for Fig. 4. Other human developmebt indices (UNDP, 2020). J.G. Djokoto Research in Globalization 4 (2022) 100079 5 IDP and a larger data set with many countries over many years was employed in this study. Data and methods Data All the data was drawn from the World Development Indicators (WDI) of the World Bank except the HDI which was obtained from the United Nations Development Programme (UNDP) website. The data consisted of an unbalanced panel of 135 countries (Appendix) from 1990 to 2019 making up 2,573 observations. The countries are modelled together without distinctions into levels of development because although these countries could be classified into stages of IDP, in practice, the countries within the development groups do move across the IDP stages over time (Djokoto, 2021a; Gorynia et al., 2010a; Gorynia et al., 2010b; Gorynia et al., 2012; Iacovoiu & Panait, 2014b). Moreover, the goal of the paper is to assess the relationship between IDP and HDI irrespective of the stage of IDP or HDI. Owing to the unbalanced nature of the panel, many countries had few years, with some as low as 2. As the number of countries far exceeded the number of years, we did not anticipate problems with the time-series properties of the panel (Baltagi, 2021; Greene, 2003; M´ aty´ as & Sevestre, 2013; Pesaran, 2015; Tsionas, 2019). Models and modelling The base model is HDI =f(IDP)(1) Based on the empirical review (Aloui, 2019; Arisman, 2018; Butar & Rahmanta, 2021; De Groot, 2014; Djokoto, 2021b; Ganiyu, 2016; Gohou & Soumar´ e, 2012; Handalani, 2018; Kolster, 2015; Oladapo & Abdul, 2016; Ranjkeshan, 2021; Sofilda et al., 2015), other variables explain HDI. Incorporating these into Eq. (1) yields Eq. (2). HDI =f(IDP,INFLA,TO,INFRAS,HC,GE,POPG)(2) The variables are described in Table 1. Eq. (2) can be specified as HDIi,t= α 0+ α 1IDPi,t+ α 2INFLAi,t+ α 3TOi,t+ α 4INFRASi,t+ α 5HCi,t + α 6GEi,t+ α 7POPGi,t+ ε i,t(3) where α k are parameters to be estimated whilst ε i,t is the idiosyncratic error term. There are i cross-sections and t periods (years). Based on the literature (Djokoto, 2021a; Dunning, 1982; Dunning, 1986; Dunning, 1993; Dunning and Narula, 1996; Narula, 1996) two variables represented IDP, IDP_GD and IDP_NOFDI. The choice of the appropriate one was accomplished by the information criteria; Akaike information criterion (AIC) (Akaike, 1974) and Bayesian information criterion (BIC) (Schwarz, 1978). Estimation procedure Earlier studies have often modelled the HDI as a linear variable (Afoakwa, 2016; Arisman, 2018; Butar & Rahmanta, 2021; Ganiyu, 2016; Handalani, 2018; Kilimova, 2016; Kolster, 2015; Nakouwo, 2019; ˇ Simanov´ a, 2013; Sofilda et al., 2015; Tamer, 2013), E(y|x) = xθ (4) With the marginal effect as ∂ E(y|x) ∂ xj =θj(5) which is constant over the range of y. However, by construction, HDI is bounded between 0 and 1, hence the plausible link functions are logit, probit, loglog, cloglog and cauchit (Djokoto, 2021b; Papke & Wooldridge, 1996; Ramalho, Ramalho, & Henriques, 2010). With this characteristic, three options can be considered to model HDI. 1. Transform the HDI by the family of link functions to linear data (Ramalho, Ramalho, & Coelho, 2018). 2. Model the HDI directly as a fraction using fractional regression (Djokoto, 2015; Djokoto & Gidiglo, 2016; Djokoto, 2021b; Papke & Wooldridge, 1996; Ramalho et al., 2010). 3. Apply the generalized estimation equations (GEE) (Gyimah, Kwansa, Kyiu, & Sikochi, 2021; Papke & Wooldridge, 2008; Xu, Solanki, & Fink, 2021). The generalised estimation equations (GEE) is a powerful tool for analyzing a variety of non-normally distributed data with correlations (Hardin & Hilbe, 2013; Liang, 1987; Pek´ ar & Brabec, 2018; Zeger & Liang, 1986; Zeger, Liang, & Albert, 1988). Whilst GEE provides correct estimates and inferences, it is often based on fewer assumptions and its Fig. 5. Summary of empirical review. Table 1 Variables, definitions, and measurement. Variable Definitions Measurement HDI Human development index Ranges between 0 and 1 IDP Either IDP_GD or IDP_NOFDI International US dollars per person or International US dollars IDP_NOFDI Net outward foreign direct investment as Investment development path International US dollars IDP_GD Gross domestic product per capita as Investment development path International US dollars per person INFLA Inflation Annual growth rate TO Trade openness Sum of exports and imports divided by GDP INFRAS Infrastructure Sum of fixed and mobile phone lines per 100 persons HC Human capital Secondary school enrolment as percentage of gross enrolment GE Government expenditure Final government expenditure on goods and services to GDP ratio POPG Population growth Annual growth rate J.G. Djokoto Research in Globalization 4 (2022) 100079 6 parameters have a more straightforward interpretation when population-level inferences are desired (Wang, Chen, & Yan, 2011; Zarei, Shabani, & Mahmoudi, 2020; Yang, Tang, & Deng, 2009). This property is particularly useful for our study as all the countries are modelled jointly. Theoretically, GEE could be used to analyze data with any number of random effects. GEE can be limited by the number of preprogrammed working correlation structures (Pek´ ar & Brabec, 2018; Wang et al., 2011). The use of GEE is very efficient when the design of a study can be modelled using any of the predefined structures (Pek´ ar & Brabec, 2018; Wang et al., 2011; Yang et al., 2009; Zarei et al., 2020). GEE accounts for correlations within the data and can handle heteroscedasticity. Three of the papers reviewed used GMM a priori (De Groot, 2014; Kolster, 2015; Ranjkeshan, 2021). However, not accounting for endogeneity is preponderated in the pertinent literature (Arisman, 2018; Butar & Rahmanta, 2021; Ganiyu, 2016; De Groot, 2014; Handalani, 2018; Sofilda et al., 2015). Moreover, the results of the effect of GDPPC on HDI has been consistent for models estimated with the GMM estimator (De Groot, 2014; Ranjkeshan, 2021) and those estimated without the GMM estimator (Arisman, 2018; Djokoto, 2021b; Handalani, 2018). Thus, we did not suspect endogeneity hence, we did not specifically account for it. Eq. (3) was estimated by Stata’s xtgee code. The family of distribution used in the GEE model was binomial, informed by the distribution of HDI (Fig. 6). The link functions included logit, probit and cloglog (Djokoto, 2015; Djokoto & Gidiglo, 2016; Papke & Wooldridge, 1996; Ramalho et al., 2010). Results and discussions The mean HDI ranged from 0.0122 to 0.9956 (Table 2). Whilst this confirms the unit interval property of the HDI, it also shows that some countries fall within the low human development category and others fall within the very high human development category. On average, however, the mean HDI is 0.7161 which falls within the high human development stage. With the standard deviation less than 1, the square, which is the variance would be closer to 0 than 1. As the variance is less than the mean, the HDI is under dispersed. Similarly, TO, INFRAS, HC and GE also show under dispersion around their respective means. The exceptions are IDP_NOFDI, IDP_GD, INFLA and POPG. Except for INFLA and GE, the correlation coefficients exceed 0.30 (Table 3). The size of the linear correlation coefficients is in order of HC, INFRAS, IDP_GD, POPG, TO, GE and INFLA. It is left to be seen if these would reflect in the size of the elasticities. As the basic IDP model relates the GDPPC to the NOFDI, either of these could be the indicator for IDP in equation (4). To choose between the two, equation (4) was estimated with an OLS estimator, one with NOFDI (IDP_NOFDI) and the other with GDPPC (IDP_GD). The AIC and BIC were computed (Table 4). The model with the GDPPC showed the lowest AIC and BIC. Hence, the IDP_GD was selected to proxy the IDP. The estimates of the GEE are presented in Table 5. As these may not be useful, the elasticities have been computed as shown in Table 6. The elasticities for the total sample are models 1, 5 and 9 respectively, for the logit, probit and complementary loglog link functions. Five layers of robustness were attained. The first was to obtain heteroscedasticity robust estimates of the coefficients from the GEE. The second was to obtain semi-robust elasticities from the coefficients of the GEE estimations. Third, was to compare the robust elasticities obtained for the total sample without observations from sub-sample without low human development (HDI below 0.550), medium human development (0.550 ≤HDI ≤0.699) and high human development (0.700 ≤HDI ≤0.799), respectively. The sub-sample had respectively, 2,139, 2,025 and 1,893 observations. For the logit link function, the elasticities are in models 2–4, models 6–8 for the probit link function and models 10–12 for the cloglog link function. Our elasticity of interest is that for the IDP_GD. For models with each link function, the elasticities of the sub-samples are generally like those of the total sample in sign, size, and statistical Fig. 6. Distribution of HDI. Table 2 Descriptive statistics. Variable N Mean Standard Deviation Minimum Maximum HDI 2,569 0.7161 0.1658 0.0122 0.9956 IDP_NOFDI 2,570 −2,143 17,584 −310,229 144,111 IDP_GD 2,573 14,696 18,873 86.1872 11,9211 INFLA 2,573 22.2657 501.2533 −9.7976 23,773 TO 2,573 86.7743 52.8581 18.9665 442.62 INFRAS 2,573 87.8800 61.5983 0.0000 342.722 HC 2,572 82.9586 28.9053 5.2208 163.9347 GE 2,573 16.7076 6.2850 0.9517 103.1732 POPG 2,573 1.2102 1.3358 −3.8477 16.7002 Table 3 Correlation matrix. HDI IDP_GD INFLA TO INFRAS HC GE POPG HDI 1.0000 IDP_GD 0.5641 1.0000 INFLA −0.0445 −0.0282 1.0000 TO 0.3072 0.3054 −0.0285 1.0000 INFRAS 0.6646 0.5537 −0.0494 0.3738 1.0000 HC 0.8576 0.5263 −0.0424 0.2353 0.6137 1.0000 GE 0.2271 0.2252 −0.0493 0.1368 0.1698 0.3038 1.0000 POPG −0.4089 −0.0735 0.0306 −0.0800 −0.2544 −0.4594 −0.1421 1.0000 Table 4 Information criteria assessment for choice between IDP_NOFDI and ID_GD. IDP variable AIC BIC IDP_NOFDI −5,711 −5,665 IDP_GD −5,792 −5,745 J.G. Djokoto Research in Globalization 4 (2022) 100079 7 significance. The fourth layer of robustness was to compare the models of the link functions of the total sample (models 1, 5 and 9). The fifth, is the estimation of the two-step system GMM. For all three GEE models, not only are the sign, size, and statistical significance of the elasticities of IDP_GD consistent but the elasticities of control variables also look-alike regarding the size and statistical significance. Thus, the elasticity of the IDP_GD from the GEE models is robust to sample size (stages of human development), and link function. Thus, much confidence can be reposed in the estimates of the relationship between IDP and HDI. For the total sample models, the elasticities of IDP_GD, TO, INFRAS and HC are statistically significant. The order of the magnitude of the statistically significant elasticities is HC, IDP_GD, TO, INFRAS. Regarding the order of the correlation coefficients, only INFRAS is out of order. By way of robustness, the general method of moments (GMM) estimator was applied to the data (Table 7). Lags 1, 2 and 3 of the dependent variables did pnot possess the acceptable statistics of the overidentifying restriction as well the second order serial correlation statistics. It was the fourth lag that has acceptable test statistics. The estimates of the control variables are like those of models 1, 5 and 9. The key variables are also statistically significant. Thus, not only are the results of the IDP_GD robust to sample size and link function, but it is also robust to estimation procedure. Thus, whether endogeneity is accounted for or not, the results are consistent. This has already been identified in the estimation procedure section. Indeed, accounting for endogeneity did not change the results. Thus, attention is turned to the GEE estimates for discussion. Rounded to two decimal places, the elasticities of IDP_GD are 0.04. This suggests that a 1% rise in IDP_GD (GDPPC) would induce a 4% rise in human development. Although the 4% elasticity is far from 100%, nevertheless, IDP explains HDI. The 4% agrees with the outcome of the theoretical review that HDI encompasses more relevant indicators of development than IDP. As Djokoto (2021b) explained, the statistical significance can be due to the presence of GDP in the construction of the HDI through the GNI. Moreover, both IDP and HDI measure development. The stronger significance and larger effect size can be attributed to the increased variability contributed by additional data. The contribution of the influence of the IDP_GD on HDI may have been re-enforced from developed, transition and other developing countries. The 4% statistically significant elasticity coupled with the linear correlation coefficient (0.5641) suggest that IDP (GDPPC) may be useful as a proxy and instrument of HDI. In the case of the latter, the condition of no correlation with the error term of the relevant model must be satisfied. The elasticity of 4% is two times the 2% found by Djokoto (2021b). Also, the statistical significance of the IDP_GD is strong, below 0.01 level of probability. In terms of the positive sign and statistical significance of GDPPC, our finding is consistent with those of Arisman (2018), De Groot (2014), Handalani (2018) and Ranjkeshan (2021). Inflation erodes the purchasing power of consumers which may negatively impact their welfare. However, the elasticity of INFLA is statistically insignificant. This suggests inflation does not have a discernible effect on HDI. The conclusions of Gohou and Soumar´ e (2012), Kolster (2015) and Ranjkeshan (2021) are consistent with our findings. Whilst Aloui (2019) reported a positive effect of inflation on HDI, Arisman (2018) and Handalani (2018) found an expected negative effect of inflation on HDI. A 1% rise in trade openness will instigate a 4% rise in HDI. External trade allows consumers to access markets outside the home economy to purchase commodities. The consumption of these would increase welfare. This explains the positive effect. The magnitude of the effect of trade is like that of IDP. However, the result is inconsistent with the existing literature (De Groot, 2014; Kolster, 2015; Ranjkeshan, 2021). Infrastructure plays an important role in an economy. The positive and statistically significant 0.0360 elasticity suggest that a 1% rise in infrastructure can instigate a 4% rise in HDI. The 4% effect is like that of IDP_GD. The positive effect is desirable and consistent in the sign for the coefficients of the infrastructure variable in the existing literature Table 5 Estimates of the GEE. (A1) (A2) (A3) (A4) (A5) (A6) (A7) (A8) (9A) (A10) (A11) (A12) Logit Probit Cloglog Total sample No. low HD No medium HD No high HD Total sample No. low HD No medium HD No high HD Total sample No low HD No medium HD No high HD IDP_GD 0.0000*** 0.0000*** 0.0000*** 0.0000*** 0.0000*** 0.0000*** 0.0000*** 0.0000*** 0.0000*** 0.0000*** 0.0000*** 0.0000*** (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) INFLA 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 −0.0000 −0.0000 0.0000 −0.0000 (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) TO 0.0016*** 0.0012*** 0.0020*** 0.0023*** 0.0009*** 0.0007*** 0.0011*** 0.0013*** 0.0007** 0.0005*** 0.0009*** 0.0009* (0.0005) (0.0003) (0.0006) (0.0008) (0.0003) (0.0002) (0.0004) (0.0005) (0.0003) (0.0002) (0.0003) (0.0005) INFRAS 0.0018*** 0.0021*** 0.0019*** 0.0018*** 0.0011*** 0.0013*** 0.0012*** 0.0011*** 0.0011*** 0.0013*** 0.0012*** 0.0012*** (0.0003) (0.0002) (0.0005) (0.0005) (0.0002) (0.0001) (0.0003) (0.0003) (0.0001) (0.0001) (0.0002) (0.0002) HC 0.0100*** 0.0060*** 0.0121*** 0.0115*** 0.0061*** 0.0035*** 0.0072*** 0.0070*** 0.0059*** 0.0033*** 0.0063*** 0.0066*** (0.0012) (0.0009) (0.0012) (0.0013) (0.0008) (0.0005) (0.0007) (0.0009) (0.0009) (0.0005) (0.0008) (0.0010) GE −0.0004 0.0017 0.0002 −0.0022 −0.0001 0.0011 0.0004 −0.0011 0.0000 0.0010 0.0011 −0.0006 (0.0026) (0.0013) (0.0031) (0.0040) (0.0016) (0.0007) (0.0017) (0.0024) (0.0014) (0.0007) (0.0014) (0.0023) POPG −0.0168 −0.0108 −0.0209 −0.0040 −0.0109 −0.0059 −0.0121 −0.0050 −0.0110 −0.0051 −0.0094 −0.0078 (0.0153) (0.0140) (0.0172) (0.0228) (0.0088) (0.0079) (0.0099) (0.0129) (0.0077) (0.0067) (0.0085) (0.0112) Constant −0.3761*** 0.0610 −0.5574*** −0.5321*** −0.2120*** 0.0701 −0.3089*** −0.2992*** −0.5081*** −0.2115*** −0.5672*** −0.5790*** (0.1124) (0.0843) (0.1169) (0.1252) (0.0710) (0.0513) (0.0745) (0.0800) (0.0803) (0.0511) (0.0871) (0.0971) Model diagnostics N 2,568 2,139 2,025 1,893 2,568 2,139 2,025 1,893 2,568 2,139 2,025 1,893 Countries 135 112 124 131 135 112 124 131 135 112 124 131 Wald χ 2 836*** 1003*** 598*** 571*** 888*** 1145*** 607*** 663*** 865*** 1188*** 594*** 622*** 1. HDI – Human development index. 2. *** and ** are 1% and 5% levels of statistical significance respectively. 3. Robust standard errors. The above table contains the raw estimates of the GEE. The theory and usefulness of GEE is outline in the subsection on ‘Estimation procedure’. As with the class of link functions, the raw estimates are not useful for the final decision. Rather, marginal effects are. Consequently, the marginal effects are reported in Table 6. J.G. Djokoto Research in Globalization 4 (2022) 100079 8 (Ganiyu, 2016; Kolster, 2015). However, the coefficient reported by Ganiyu (2016) was not statistically significant whilst that reported by Kolster (2015) was significant. The coefficient for HC is positive and statistically significant. Thus, a 1% rise in HC (secondary school enrolment as a percentage of gross enrolment) will induce a 21% rise in HDI. This is a relatively high elasticity and indeed, the highest elasticity among the estimated elasticities which shows that human capital is the most potent driver of HDI. The result shows that human resource remains the most potent factor in improving human development. The proxy has also been used in other studies to represent education. It must, however, be noted that the education variable in the HDI is mean years of schooling and expected years of schooling which differs from the gross secondary school enrolment. Our finding is consistent with that of Ganiyu (2016), Oladapo and Abdul (2016) and Sofilda et al. (2015). De Groot (2014) reported a positive but statistically insignificant effect. The elasticity of GE is statistically indistinguishable from zero. Thus, GE has no discernible effect on HDI. This finding is consistent with the previous studies (Ganiyu, 2016; De Groot, 2014; Gohou & Soumar´ e, 2012; Sofilda et al., 2015). Whilst Aloui (2019) reported a significant negative effect, Kolster (2015) found a significant positive effect. An increase in population increases competition for health, education, and financial resources. This tends to be pronounced in resourcepoor environments such as developing countries. Thus, Arisman (2018) and De Groot (2014) reported significant negative effects of population growth on HDI in developing countries. This explains the negative sign of the elasticity of POPG. However, the current study includes data on transition and developed economies where the competition effect may be less pronounced. Thus, the competition effect is not strong enough to induce a discernible effect on HDI. Although Sofilda et al. (2015) reported a neutral effect of POPG on HDI, nevertheless, the sign of the coefficient was negative. Conclusions and recommendations In this study, the nexus between the investment development path and the human development index was examined. HDI was modelled using generalized estimation equations and accounting for the unit interval property using non-linear link functions. Also, a more flexible proxy for IDP, gross domestic product per capita and a larger data set with many countries over many years was employed in this study. A Table 6 Elasticities of the GEE estimates. (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) Link Logit Probit Cloglog Sample Total sample No low HD No medium HD No high HD Total sample No low HD No medium HD No high HD Total sample No low HD No medium HD No high HD IDP_GD 0.0401*** 0.0451*** 0.0407*** 0.0415*** 0.0388*** 0.0415*** 0.0395*** 0.0402*** 0.0356*** 0.0365*** 0.0368*** 0.0368*** (0.0076) (0.0045) (0.0112) (0.0085) (0.0061) (0.0040) (0.0097) (0.0069) (0.0042) (0.0034) (0.0075) (0.0049) INFLA 1.72E-05 1.11E-05 4.56E-05 8.80E-06 1.31E-05 5.78E-06 3.86E-05 3.88E-06 −2.32E-06 −4.21E06 1.73E-05 −1.90E-05 (2.48E-05) (8.26E05) (3.11E-05) (2.01E-05) (2.50E-05) (8.44E05) (3.12E-05) (2.00E-05) (2.81E-05) (0.0001) (3.49E-05) (1.95E-05) TO 0.0377*** 0.0252*** 0.0432*** 0.0506*** 0.0355*** 0.0240*** 0.0410*** 0.0468*** 0.0284** 0.0199*** 0.0344*** 0.0351** (0.0119) (0.0068) (0.0132) (0.0173) (0.0119) (0.0068) (0.0130) (0.0174) (0.0118) (0.0072) (0.0129) (0.0177) INFRAS 0.0360*** 0.0414*** 0.0389*** 0.0335*** 0.0376*** 0.0451*** 0.0413*** 0.0353*** 0.0421*** 0.0506*** 0.0471*** 0.0413*** (0.0070) (0.0049) (0.0111) (0.0100) (0.0060) (0.0048) (0.0102) (0.0090) (0.0053) (0.0048) (0.0091) (0.0084) HC 0.2135*** 0.1235*** 0.2395*** 0.2283*** 0.2194*** 0.1256*** 0.2428*** 0.2334*** 0.2230*** 0.1277*** 0.2344*** 0.2339*** (0.0300) (0.0185) (0.0261) (0.0323) (0.0306) (0.0187) (0.0270) (0.0328) (0.0338) (0.0192) (0.0314) (0.0377) GE −0.0017 0.0068 0.0008 −0.0107 −0.0011 0.0072 0.0032 −0.0087 0.0003 0.0074 0.0084 −0.0046 (0.0124) (0.0049) (0.0135) (0.0191) (0.0122) (0.0050) (0.0130) (0.0191) (0.0116) (0.0055) (0.0111) (0.0185) POPG −0.0070 −0.0026 −0.0082 −0.0020 −0.0074 −0.0024 −0.0077 −0.0041 −0.0072 −0.0022 −0.0056 −0.0060 (0.0065) (0.0034) (0.0068) (0.0116) (0.0060) (0.0032) (0.0064) (0.0106) (0.0051) (0.0029) (0.0052) (0.0086) Model properties N 2,568 2,139 2,025 1,893 2,568 2,139 2,025 1,893 2,568 2,139 2,025 1,893 1. HDI – Human development index. 2. *** and ** are 1% and 5% levels of statistical significance respectively. 3. Robust standard errors. Table 7 Two step System GMM. Variables Estimates HDI Coefficients (Robust standard errors) HDI_1 0.8077 (0.0024) HDI_2 −0.0047 (0.0001) HDI_3 −0.0046 (0.0002) HDI_4 0.0279 (0.0003) GDPPC 2.63e−07*** (7.80e−09) INFLA 2.23e−06*** (1.41e−07) TO 3.05e−05*** (2.15e−06) INFRAS −2.58e−05*** (1.54e−06) HC 0.0006*** (1.97e−05) GE 0.0004*** (1.13e−06) POPG 0.0005** (0.0002) Constant 0.0687*** (0.0006) Model diagnostics Observations 1,049 Countries 96 Wald chi2(11) 2.14e +06*** Probability of Sargan statistics 0.3698 Probability of second order serial correlation 0.6645 1. HDI – Human development index. 2. *** and ** are 1% and 5% levels of statistical significance respectively. 3. Robust standard errors. This estimation accounts for endogeneity by using the general method moments. This inherently accounts for heteroscedasticity and serial correlation. The fourth lag is the the Sargan test and second order serial correlation test statistics are acceptable. The positive and statistical significance of the GDPPC suggests the GMM results are robust to estimation procedure and accounting for endogeneity. J.G. Djokoto