ABSTRACT
Previous studies have reached mixed conclusions regarding the relationship between inequality and per capita income. These studies, however, fail to consider gender differences in income inequality and how these may impact on the relationship between income inequality and per capita income. Using Australian taxation statistics, we derive three sets of Gini coefficients (i.e. female, male and total) for the period 1950–2013. We then examine the relationship between inequality and real per capita income and find that a gender-specific threshold panel regression outperforms three other conventional models. Our findings suggest that ‘one set of coefficients does not fit all’ in that the use of aggregate and constant coefficients may mask variations within, and between, gender inequality over time.
KEYWORDS:
Disclosure statement
No potential conflict of interest was reported by the authors.
Notes
1 While OECD (Citation2012) suggests the gender wage gap in Australia is smaller than the OECD average, other research suggests that the gender wage gap in Australia has barely shifted since the 1990s, stuck between 15% and 17%, which is almost identical to the OECD average (NATSEM Citation2009; Young Citation2013).
2 We separately compute a series incorporating non-lodgers based on the ATO’s Review into the Non-Lodgment of Individual Income Tax Returns (Inspector General of Taxation Citation2009). Individuals with income below the tax-free threshold account for 80% of nonlodgement and are assumed to earn zero income, as this is the lowest reportable income. This provides an upper bound on calculated Gini coefficients. Late submitters and tax evaders account for the remaining 20% of nonlodgers, who are thought to have incomes matching the reported distribution and are assimilated accordingly. We find that incorporating nonlodgers in this way does not materially alter the path of inequality over time.
3 To address the possible simultaneity problem and check the robustness of the results, we also estimated Models 1 and 2 by two-stage least square (2SLS) in , utilizing Ln(Git–1) and Ln(Yit–1) as internal instruments. and show that the OLS and 2SLS estimators are very similar in terms of sign and magnitude, and hence, our inferences remain the same. The results also changed very little when we used Ln(Git–1), Ln(Git–2), Ln(Yit–1) and Ln(Yit–2) as internal instruments. As such, we proceed with the results in .