in partition ∏ of size j(n)’.
Data and Results
We apply the idea drawn on Elbers et al. (2008) to the data collected from two rounds of the India Human Development Survey (IHDS) conducted in 2004–2005 and 2011–2012. The National Council of Applied Economic Research (NCAER) collaborated with the University of Maryland to collect this unique data set on India. The survey covers almost all the states and union territories of India (except Andaman and Nicobar Islands and Lakshadweep). Using stratified random sampling technique, 27,010 rural households from 1,503 villages and 13,126 urban households from over 971 urban blocks were surveyed in 2004–2005. In the second round, conducted in 2011–2012, around 83% of the households were reinterviewed, which also include split households located within the same village or town. Altogether, 42,152 households were surveyed in the second round.
Table 1:Changes in income inequality in India by different measures.
Source:
First, we take a look at the changes in overall inequality in household per capita income in India during the period between 2004–2005 and 2011–2012. Strikingly, no matter which measure we use, income inequality has unambiguously worsened in India during this period (Table 1).
The belief that estimates of inequalities in reported consumption expenditure, which are usually made on the basis of different rounds of National Sample Surveys (NSS), grossly underestimate the true extent of economic inequality in India is somewhat vindicated by this finding. This is no more than reconfirmation of Amaresh Dubey’s finding, based on the same IHDS data, that Gini coefficient for income is around 0.55 in India as compared to that of consumption, which is 0.37 (cited in Dev, 2016). Studies based on NSS data have also shown that inequality in consumption expenditure has increased in the post-reform period, but the increase has been more in urban than in rural areas.
Does the higher level of overall interpersonal inequality necessarily imply higher inter-group inequality? Even though it is conceivable that the higher the overall inequality, the greater is the inequality between groups, the decomposition formula is also consistent with the scenario when overall inequality is rising while between-group inequality is declining or not changing. We now examine the contribution of the between-group component to total inequality following the conventional method of decomposition and with four mutually exclusive and exhaustive categories across the religious and caste divides, viz., Muslims, SCs, STs, and Hindu others. The last category is the residual category excluding the SC and ST households from all Hindu households. We find that the general entropy measures for c = 0, 1, 2 — all show declines in the contributions of between-group components to total inequality (Table 2). It can also be observed that the contributions seem rather low because of the reason explained earlier.
As discussed earlier, the conventional method is sensitive to the number of groups considered as well as the relative composition of the groups. We check this point with six mutually exclusive groups, splitting the Hindu others category into Brahmins, high castes, and Other Backward Castes (OBC). As expected, the contributions of between-group inequality to overall inequality increase, and the contribution of between-group inequality declines between the two survey rounds (Table 3).
Table 2:Contributions of between-group component (conventional decomposition) (with four social groups).
Source: Calculated from IHDS data.
Table 3:Contributions of between-group component (conventional decomposition) (with six social groups).
Source: Calculated from IHDS data.
Table 4:Shares of between-group component (Elbers et al.) (with four groups).
Source: Calculated from IHDS data.
On application of the modified decomposition method as suggested by Elbers et al. (2008), we find some interesting differences as far as the GE family of measures is concerned. Clearly, the contributions of between-group inequality at both the time points have increased, as expected. The contribution of between-group inequality by applying GE(2) in 2004–2005 turns out to be as high as 13.6%. Yet again, the between-group contributions are all found to have declined in this period (Table 4).
Interestingly, in this modified approach, the contribution of between-group inequality does not necessarily increase with increase in the number of groups, unlike in the case of the conventional approach. A comparison between Tables 4 and 5 illustrates the point. However, what stands out is that, no matter whether we apply the conventional decomposition method or the method suggested by Elbers et al., the contribution of between-group inequality to total inequality between the two rounds of IHDS shows a decline.
Table 5:Shares of the between-group component (Elbers et al.) (with six groups).
Source: Calculated from IHDS data.
Table 6:Income shares and population shares of four social groups in the two rounds of IHDS.
Source: Calculated from IHDS data.
This takes us to a further limitation of this way of looking at inequality between groups. Even though the decomposition methods give the quantitative contribution of between-group inequality in an overall sense, which in our case has declined, it does not allow us to say anything about the relative attainments of different groups among the four groups considered. In order to examine this further, we look at the respective income shares and population shares of four mutually exclusive and exhaustive social groups, viz., SC, ST, Muslims, and Hindu ‘others’. This is one of the simplest approaches toward assessing inequality between groups — called ‘representational inequality’ by Reddy and Jayadev (2011) — which substantially differs from the commonly used approach that views inter-group inequality as a constituent part of overall interpersonal inequality. Table 6 allows us to compare the changes in the respective shares, which shows that among the four groups, only the SC households as a group have been able to improve their income share vis-à-vis their population share, which is reflected in a higher ratio of the two shares in IHDS round 2. This is at the expense of the declining ratios for other three groups.
The category Hindu others in Table 6 includes a wide range of sub-categories with significant differences. We, therefore, further split this category into Brahmins, high castes, and Other Backward Castes (OBC) in order to see if significant differences exist. Now, with six mutually exclusive categories, we find that SCs and OBCs have experienced improvement in their respective income shares vis-à-vis population shares, but the ratio of the income share to the population share for each of the other four categories, viz., STs, Muslims, Brahmins, and other high castes, has declined (Table