Fractionalization and long-run economic growth: webs and direction of association between the economic and the social -- South Africa as a time series case study

Abstract

Recent cross-sectional growth studies have found that ethnolinguistic fractionalization is an important explanatory variable of long-run growth performance. In the present article, we follow the call of earlier studies to conduct a more detailed clinical analysis of the growth experience of a specific country. South Africa constitutes an interesting case in which to explore these questions. The results of this study provide important nuance to the existing body of evidence. We find that fractionalization is subject to strong change over time. In addition, we find strong evidence of webs of association between the various social, political and institutional dimensions. Thus various forms of social cleavage tend to go hand in hand, which presents the danger of spurious inference of association. Further, the direction of association in the preponderance of cases runs from economic to social, political and institutional variables, rather than the other way around. However, there remain significant impacts from some, but only some fractionalization indexes on economic growth. Which social cleavage, when, how and for what period of time will depend on the historical path of specific societies.

Notes

¹ Also see Alesina et al . (2002), Collier (2001), Knack and Keefer (1997) and Posner (2004). For a broader conception of institutions and their impact see Assane and Grammy (2003), Bhattacharyya (2004) and Hall and Jones (1999).

² South Africa constitutes an interesting case in which to explore these questions. In terms of the Ethnolinguistic Fractionalization Index cited by Easterly and Levine, South Africa in 1960 ranked 6th out of 66 countries covered by this index with a value on the measure of 88 (that is a probability of 88% that two randomly chosen individuals are of different ethno-linguistic origin). Important, too, in the South African case is that from 1948 through to the demise of Apartheid, a major attempt was made to socially and politically ‘engineer’ South Africa along the lines of supposedly ‘essentialist’ ethnic identities. At least with regard to the ‘African’ community these ethnic divisions were meant to coincide with linguistic identities. That South Africa continues to be linguistically highly fractionalized is registered by the fact that the present (post-1994). Constitution acknowledges linguistic diversity through the legal recognition of eleven official languages. If fractionalization is an important explanatory variable in the long-run growth process of a country, South Africa is ideal to test this hypothesis.

³ Questions surrounding the direction of association between variables in growth contexts are both pervasive and crucial. For discussions surrounding methodological and estimation issues in these contexts see Awokuse (2005), Chang and Cuadill (2005) Shan (2005) and Wahab (2004).

⁴ The possibility of associations between and amongst SPI variables themselves, raises the additional concern that the introduction of any one of the SPI-dimensions into any growth equation for estimation, may generate spurious results. Analysis of the webs of association between the SPI variables should improve our understanding of this danger. The availability of time series data is of particular relevance here. Muscatelli and Tirelli (2001) provide an example for the link between unemployment and economic growth.

⁵ For a South African perspective, see Fedderke et al . (2001b).

⁶ In traditional South African growth debates, this view underpins the interpretation of the South African growth path that is present in the contributions of O’Dowd (1974, 1978) and Bromberger (1974, 1978), for instance. Presentation of international evidence is myriad with both affirmative (for instance Muller, 1997) and more sceptical views present in the literature (see for instance, Huntington, 1984, though the scepticism is revised somewhat in 1991).

⁷ There exists a separate but related literature that explicitly examines the link between income inequality and long-run growth. See for instance the discussion in Bengoa and Sanchez-Robles (2005) and in Alesina and Perotti (1993).

⁸ A more comprehensive discussion of the fractionalization indexes and methodological questions surrounding their construction is presented in Fedderke et al. (2005).

⁹ We follow the standard measure given by , where n_i denotes the number of members of households that cite the i’th language as the principal medium of communication within the household, N denotes the number of members in the population. F thus computes the probability that two randomly chosen individuals speak different first languages.

¹⁰ A general point that emerges for studies employing measures of linguistic fractionalization, is that given the data problems even for a country such as South Africa, which amongst developing countries has relatively extensive and sound data collection procedures, this raises significant questions concerning the quality of data for developing countries in general.

¹¹ On a prior analysis of religious fractionalization and its economic impacts, see Barro and McCleary (2003).

¹² Again the time frame was dictated by data availability.

¹³ The measure was first introduced in Fedderke et al . (2001a).

¹⁴ From 1910 to 1983 South Africa had only the House of Assembly as its formal legislative chamber. It essentially represented Whites, though even prior to 1983 representation of Africans, Coloureds and Asians existed under the Representation of Natives Act (1936), and the Separate Representation of Voters Act (1955). Such representation was more notional than real. From 1983 South Africa introduced a three-chamber parliament, with a White House of Assembly, the Asian House of Delegates, and the Coloured House of Representatives. Africans had no representation (outside of the notionally independent ethnically defined homelands), until 1994 and the introduction and the first democratic South African parliament.

¹⁵ See Fedderke et al . (2005) for a more detailed exposition.

¹⁶ At this point we do not develop this point further – but note that in cross-sectional contexts this issue surely compounds – see Fedderke et al . (2005) for a more detailed exposition.

¹⁷ See also the discussion in Pesaran (1997) and Pesaran and Shin (1995a, 1995b) and Pesaran et al. (2001). Suppose that the question is whether there exists a long-run relationship between the set of variables y_t , x _{1, t} , … x_{n , t} . Univariate time series characteristics of the data are not known for certain. The PSS approach to testing for the presence of a long-run relationship proceeds by estimating the error correction specification given by: . The test proceeds by computing the standard F-statistic for the joint significance of δ₁ = δ₂ = ··· = δ _{n + 1} = 0 , under all feasible alternative LHS variables. While the distribution of the test statistic is nonstandard, with x_{i, t} ∼ I(0) ∀ i providing a lower bound value, x_{i, t} ∼ I(1) ∀ i an upper bound value to the test statistic. The test is analogous to a Granger causality test, but in the presence of nonstationary data. This renders the PSS F-test suitable in the current context.

¹⁸ We also examined correlation coefficients. Appendix B reports the results. In particular, we find that: high religious fractionalization is associated with lower political fractionalization, and lower political rights; high religious fractionalization is associated with higher output; high linguistic fractionalization is associated with lower property rights; high linguistic fractionalization is associated with lower output and capital stock, but with higher investment in capital; high political fractionalization is associated with higher political rights; high political fractionalization is associated with lower output; there is a positive association between political and property rights; property rights are negatively associated with racial fractionalization; property rights are positively associated with output, capital and investment; instability does not appear to be strongly associated with any other dimension; racial fractionalization is negatively associated with output and capital stock, and positively associated with investment.

¹⁹ The VECM to cointegration is now standard, and requires little elaboration. See Johansen (1988) and Johansen and Juselius (1990, 1992).

²⁰ The PSS ARDL approach to cointegration does allow for some flexibility in the use of both I(0) and I(1) data, and we proceed with the analysis on this basis. Consideration of the stationarity properties of the data implied by augmented Dickey--Fuller test statistics, precludes the possibility of systematic association between at least some variable groups. Some variables are ∼I(0) (LANGFRAC, GROWTH), some ∼I(1) (POLFRAC, INSTAB, POLRGHT, PROPRGHT, LNRGDP, dK/dt), and in two cases the variables are ∼I(2) (RELFRAC, RACE). The implication is immediately that at least some of the associations suggested by the F-tests for direction of association of Table 1 are likely to be spurious. Instead, the univariate time series structure of the data suggests that associations between the variables included in this study may have to be found between transformed variables. We note at the outset that where the univariate time series characteristics of the data are rigorously adhered to, the implied associations between variables becomes between levels, changes in variables, as well as rates of change. Problems of interpretation compound.

²¹ A countervailing view on the conflict in Rwanda argues that it was less about ethnicity and rather an outcome of the unbearable stress on the land as a result of a Malthusian population trap (Andre and Platteau, 1998).

²² This emerges from the fact that in the loading matrix of the VECM structure, both elements of the loading matrix associated with political instability reject the presence of an equilibrium relationship for political instability. By contrast, the loading matrix confirms the presence of an equilibrium relationship for property rights.

²³ Also see Fedderke and Klitgaard (1998).

²⁴ Though note that both Fielding (2000) and Fedderke (2004) find political instability to be a significant and powerful driver of investment in the South African manufacturing sector.

²⁵ See Johansen and Juselius (1990).

²⁶ See Wickens (1996), Johansen and Juselius (1990, 1992), Pesaran and Shin (1995a, b), Pesaran, Shin and Smith (1996).

²⁷ The first difference specification is driven by the ∼I(2) structure of racial fractionalization in South Africa. Inclusion into the Johansen VECM framework thus requires the first difference transformation.

²⁸ The exogeneity restrictions were extensively tested. Results that justify the restrictions under which estimation proceeds here have been presented more extensively elsewhere (Fedderke and Luiz, 2005). Exogeneity of the racial composition of a population has immediate intuitive purchase. While property rights exogeneity may be more controversial, note that the restriction is merely on contemporaneous feedback effects. Moreover, that property rights might lead political rights has additional support in the literature (Schultz, 1992; Sened, 1997; Weimer, 1997), and is justifiable on the grounds that rights over property might be granted in an attempt to lower the danger of political change that might widen access to rights over setting the fundamental rules of the game.

²⁹ One interpretation of the sign of the association between racial fractionalization and instability is that it is a reflection of a minority elite that responded to a challenge of its authority through two mechanisms. First, as the (White) ruling elite came to constitute an ever smaller proportion of the total population, its response to the rising challenge to its monopoly on power (see the political instability index reported in Fedderke et al ., 2001a) was first to increase repression, and secondly to increase attempts at buying off portions of the majority of the population (for instance through the creation of homelands). Or conversely: higher racial fractionalization meant less need to buy off the largest section of the population, hence higher growth. The sign on the relationship between fractionalization and growth is not stable across alternative specifications. For instance, Fedderke and Luiz (2005) in an alternative specification find contrary evidence on the sign of the association. A possible reason for the lack of stability may be the presence of nonlinearities in the association – necessitating the use of threshold autoregressive estimation methodologies. Unfortunately limited degrees of freedom preclude the use in the present context. Our discussion suggests the possibility of non-linearity -- as the ruling elite came to constitute an ever smaller proportion of the total population, its response to the rising challenge was first to increase repression, and then, once it became too small to resist further, it resorted to accommodation through a negotiated settlement.

³⁰ Impulse response analysis confirms equilibriating behaviour in the system, though stability emerges over a relatively long period of time.

³¹ See Rodrik et al . (2004) and Glaeser et al . (2004) for evidence of the association between institutions (social and political) and the productive capacities of societies.