The impact of trade liberalization on child labor in Pakistan

ABSTRACT

This study investigates the impact of trade liberalization on child labor using micro-level datasets of urban Pakistan. It is anticipated that when economies are open to external competition, this will benefit the poor in those economies, and hence lead to a contraction of child labor. We analyze the geographical differences in the effects of trade policy directed toward the reduction of import tariffs. Trade liberalization and child labor are associated in the case of Pakistan. Trade liberalization reduces child labor in urban Pakistan. Our findings also reveal a statistically significant association between lagged tariffs and child labor. Our findings are robust to various control variables.

KEYWORDS:

• Trade Liberalization
• child Labor
• Pakistan

JEL CLASSIFICATION:

• F14
• F16
• O53

Acknowledgements

We would like to thank Hubert Visas, Huaping Sun, Farwa Wajid, Noureen Khanum and seminar participants at the Department of Economics, University of Sargodha for helpful comments and suggestions.

Disclosure statement

The authors declare that they have no conflict of interest.

Notes

1. For theoretical links detail see Edmonds and Pavcnik (2006).

2. See Zaidi (2015) for details about trade reforms and their timing in Pakistan.

3. For trade ratio see Edmonds and Pavcnik (2006), Jintong (2017), Kis-Katos (2007).

4. First, we check pooled vs fixed effects models using testparm command after LSDV and found that time & province fixed effects are significant. We also test simple OLS vs random effects models and Breusch-Pagan Lagrange multiplier (LM) guided us for random-effects models. Hausman test further suggested us random-effects model. Modified Wald test pointed out heteroskedasticity across provinces. We have N = 4 and T = 13, FGLS is a better choice (because T > N) for estimation of Equation (1)(1) $\mathrm{C}{\mathrm{L}}_{\mathrm{p}\mathrm{t}}={\beta }_0+{\beta }_1\mathrm{P}\mathrm{L}\mathrm{I}{\mathrm{B}}_{\mathrm{p}\mathrm{t}}+{\beta }_2{\mathrm{X}}_{\mathrm{p}\mathrm{t}+}{\epsilon }_{\mathrm{p}\mathrm{t}}$(1) for the said analysis than the random-effects model with robust standard errors.

5. H =$\frac{q}{N}$, where Headcount index = H, number of poor = q, population size = N. it is sensitive to number of poor, and most widely used poverty measure but it does not satisfy the transferability and monotonicity axioms.

6. $PG=\frac{1}{n}\sum _{i=1}^q\left[\frac{z-y_i}{z}\right]$, $y_i$ shows individual’ income & sum is taken just on those people whom incomes lower than poverty line.it shows the poverty depth. This measure fulfill the HCR defect (i.e satisfy monotonicity axioms) but does not fulfill transfer axiom.

7. $SPG=\frac{1}{n}\sum _{i=1}^q{\left[\frac{z-y_i}{z}\right]}^2$, it satisfies both the axioms of monotonicity and transferability it shows the ‘severity of poverty’.

8. $Gini=\frac{1}{2n^2\bar{y}}\sum _{i=1}^n\sum _{j=1}^n|y_i-y_j|$, the Gini coefficient value of zero indicates the perfect equality and the value of one indicates the perfect inequality.

9. Atkinson index ${\mathrm{A}}_{\epsilon }=1-[\frac{1}{\mathrm{n}}\sum _{\mathrm{i}=1}^{\mathrm{n}}{\left(\frac{{\mathrm{y}}_{\mathrm{i}}}{\bar{y}}\right)}^{1-\epsilon }{]}^{\frac{1}{\left(1-\epsilon \right)}}$ whereas ε is inequality aversion parameter, 0 < ε<∞: higher value of ε indicates that people are more concerned to inequality and lower value indicates that people are less concerned to inequality. $A_{\epsilon }$ can takes the values from zero to one. Zero value of ε means that people are indifferent about inequality whereas one shows the perfect inequality (Litchfield 1999).

10. $GE\left(\alpha \right)=\frac{1}{{\alpha }^2-\alpha }\left[\frac{1}{n}\sum _{i=1}^n{\left(\frac{y_i}{\bar{y}}\right)}^{\alpha }-1\right]$ Where n denotes individuals, I for income $\bar{y}=\frac{\mathrm{\Sigma }{y}_i}{n}$ is mean income. This GE can take values from 0 to ∞. Zero shows equality while higher values indicate higher levels of inequality. The weight given to distances between incomes are indicated by α. 0,1 & 2 are the commonly used values for it. GE is more sensitive to change in lower tail for lower values of α while higher values affect the upper tail. α = 0 stress more weight in lower tail, α = 1 gives equal weights across distribution while α = 2 stress more weights in upper tail, when in GE α is given 0 it becomes Theil’s L index or mean log deviation. While (1) parameter makes it Theil’s T.Index. Such as $GE\left(0\right)=\frac{1}{n}\sum _i^{n}log\frac{\bar{y}}{y_i}$ and $GE\left(1\right)=\frac{1}{n}\sum _{i=1}^n\frac{y_i}{\bar{y}}log\frac{y_i}{\bar{y}}$.

Additional information

Funding

This research is not supported by any research agency or institution.