An Unconditional Quantile Regression Based Decomposition of Spatial Welfare Inequalities in Ghana

Abstract

Ghana has witnessed tremendous economic growth since the 1990s and attained the Millennium Development Goals target of halving poverty. This notwithstanding, inequality in Ghana increased over the same period, suggesting growth benefits were not equitably distributed. This study provides evidence on the determinants of household consumption expenditure and factors that explain rural-urban welfare gaps between 1998 and 2013. The study employs an unconditional quantile regression and recently proposed decomposition technique based on re-centred influence functions. We find significant spatial differences in consumption expenditure across selected quantiles, with rural-urban inequalities driven largely by differences in returns to households’ endowments.

Disclosure statement

No potential conflict of interest was reported by the authors.

Supplemental Material

A Supplementary Material is available for this article which can be accessed via the online version of this journal available at http://dx.doi.org/10.1080/00220388.2017.1296571.

Notes

1. The Ghana Statistics Service (GSS) constructs an equivalent adult measure of household size which allows for variation in household composition. The measurement takes into consideration the fact that adults’ consumption requirements are higher than those of babies and children. Thus, the measure is based on calorie needs of different household members, with the calorie requirements distinguished by age and gender (GSS, 2014b).

2. A discussion on rural-urban inequalities in Ghana is provided in Supplementary Material.

3. We follow the standard poverty theory by including age squared as an explanatory variable. The life-cycle hypothesis indicates a possible quadratic relationship between welfare and age – comparatively poverty is higher at younger ages, decreases in one’s middle-age, and rises again beyond a certain age.

4. This encompasses all other household income earned excluding remittances and wages.

5. We provide advantages of the unconditional quantile regression over conditional quantile regression in Section 2 of the Supplementary Materials.

6. For further understanding of the differences between estimated coefficients from unconditional quantile regression and conditional quantile estimations see Borah and Basu (2013).

7. The law states that the expectation of the conditional expectation is the unconditional expectation. That is the average of the conditional averages is the unconditional average.

8. Yun (2005) provides a solution to the problem of using categorical variables by limiting the sum of the coefficients for a set of categorical variables by imposing normalisations on coefficient to purge the intercept from the effect of the omitted category (Fortin et al., 2011). It then expresses the coefficient of the transformed equation to reflect a deviation from the estimated parameters instead of deviations from the base category. However, Fortin et al. (2011) show that some degree of arbitrariness is used to derive the normalised equation and implementing this solution is at the cost of interpretational challenges. In this regard, this study avoids the use of Yun’s (2005) solution to normalise the categorical variables in order to maintain their economic significance.

9. The assumption of ignorability is widely used in the programme evaluation literature and allows ruling out the selection into a particular group based on unobservables, making them identical across groups once we condition on a vector of observed component. The overlapping support assumption requires overlap of observable characteristics of groups, that is to say no single value of X = x or ε = e can serve to identify membership into one of the groups (Fortin et al., 2011).

10. The data for the study can be sourced from GSS and available on request at a fee. The STATA codes used to generate the results in this article are available on request from the corresponding authors.

11. In 2007, the Bank of Ghana redenominated the national currency from the Cedi to Ghana Cedi at a conversion rate of 10,000 Cedis equal to 1 Ghana Cedi. Expenditures were measured in Cedis in 1998/1999 and Ghana Cedis in 2012/2013. To ensure consistency and ease of comparison between estimated coefficients, the conversion rate is applied to real household expenditures in 1998/1999.

12. There is the possibility of endogeneity of some covariates in the model, however, due to restriction of the dataset the study could not identify suitable instruments for resolving the problem. In this regard, interpretation of the result must be done with care. In spite of this challenge the importance of this study cannot be underestimated.