We interpret the previous results as evidence that the local population did benefit from the increase in copper production that culminated from the exogenous shock of the international copper prices. This section explores whether alternative stories could explain the same results. We focus on using alternative measures and estimations. We also explore the use of restricting the sample to certain groups of the population.
33 This kind of taxing system is a revenue system which is neither adequate in a dynamic system nor stable over the longer term.
34 A witness to this unanticipated change was the example of Anglo Americas’ decision to withdraw from copper mining. This led to changes in income levels as the new mine owners could not offer the mine workers the same conditions of service as offered by Anglo America. This also led to an exodus of highly skilled workers from the mining constituencies.
2.5.8.1. Additional Controls
Table 2.8 shows the 2SLS FE regressions of the effect of per capita mine level copper output on real incomes with additional control variables. The specific control variables are listed in the table. These include facilities and female-headed households in urban areas. These variables are important as they are likely to indirectly affect living standards. For example, controlling for female headed households in urban areas somehow disentangles the effects of mine level copper output as most mine workers tend to be male and also concentrates the analysis on the urban population where most of the copper production takes place. In addition, the sex of the household head can have a significant influence on the household expenditure behavior as well as the household incomes. Considering that the analysis is concentered on the effects of mining on living standards, this has serious implications for the study as females are less likely to be employed in the direct mining operations which take up a greater proportion of the workforce. Female headship is also typically expected to increase the likelihood of a household being found amongst the poor. While the presence of good facilities such as good sanitation and access to clean water is likely to have adverse effects on the living standards of communities. The findings on the effect of per capita copper output on real incomes clearly hold when there is an inclusion of more control variables even though the included variables are not statistically significant. It can also be noted that government spending maintains the same sign and level of significance. The first stage results are also clearly valid. The Kleibergen-Paap F-statistic is highly significant. In addition, the Angrist Pischke F-statistics for the interactions statistically significant, ruling out concerns regarding weak instrument. As shown in Table 2.8, the coefficient on per capita mine level copper output demonstrates that the positive relationship between natural resource price shocks and living standards still holds in the case of the regressions with additional controls.
Table 2.8- Effect of Ln(Copper Output) on Ln(Real Incomes) with additional control
Average age of household heads 1.457*
(0.620)
F-statistic
Notes: Robust standard errors in parenthesis. Standard errors are clustered at constituency level. *denotes significant at 10%, ** significant at 5% and *** significant at 1%. All regressions include year and constituency fixed effects. The full set of control variables includes: constituency level age, average household size and population density.
Panel 2 of the table provides diagnostic tests for the IV regression model reported in Table 2. The Kleibergen-Paap is a test of identification distributed as chi-square under the null of under-identification. The Anderson Rubin and Stock-Wright LM statistic are weak instrument-robust inference tests, which are distributed as F-test and chi-square respectively, under the null that coefficients of the endogenous regressors in the structural equation are jointly equal to zero, and the over-identifying restrictions are valid.
2.5.8.2. Analyzing the Effect on poverty with additional control variables
In this section, an analysis of the effect of per capita mine level copper output on poverty levels with additional controls is made. We include other variables such as the facilities that a household has and the proportion of female-headed households in urban areas. The former is likely to be an important determinant of living standards as it will be affected by the employment rate in a particular constituency as households with stable incomes are likely to be surrounded by better living facilities. Thus, we try to disentangle the impact of per capita mine level copper output on poverty.
The results are given in Table 2.9. All columns in the table confirm the same result that one can observe in Table 2.5: the effect of copper production is highly significant. This just proves that poverty levels have undergone a positive transformation which from earlier results could also be attributed to the documented increase in the levels of employment by the mines; which could have been one of the major channels through which living standards have been impacted. Finally, we find that these relationships are stronger for the whole sample. The coefficients of interest have also kept the same magnitude throughout the whole analysis; this also supports the robustness of the obtained results.
Table 2.9- Effect of Ln(Copper Output) on Ln(Poverty) with additional control variables
Panel 1
2SLS FE 2SLS FE 2SLS FE
Independent Variables (1) (2) (3)
Extreme Poverty Moderate Poverty Non-poor Ln(Per capita mine level copper output) -0.152*** 0.095*** 0.073***
Population density -0.371 0.196 0.128
(0.243) (0.167) (0.147)
Female headed households -0.041 0.023 0.018
(0.027) (0.022) (0.019)
F-statistic
Notes: Robust standard errors in parenthesis. Standard errors are clustered at constituency level. *denotes significant at 10%, ** significant at 5% and *** significant at 1%. All regressions include year and constituency fixed effects. The full set of control variables includes: constituency level age, average household size and population density.
Panel 2 of the table provides diagnostic tests for the IV regression model reported in Table 2. The Kleibergen-Paap is a test of identification distributed as chi-square under the null of under-identification. The Anderson Rubin and Stock-Wright LM statistic are weak instrument-robust inference tests, which are distributed as F-test and chi-square respectively, under the null that coefficients of the endogenous regressors in the structural equation are jointly equal to zero, and the over-identifying restrictions are valid.