Human Trafficking is a tremendously vexing crime. While abuses are often hidden in plain sight, the psychological effects of this crime against humanity are all too apparent. As a form of modern slavery where traffickers use force, fraud, or coercion to compel people to work or engage in sex acts, the signs are often difficult to identify and the victims are hard to find. This is especially true for child sex trafficking, where, according to international law, abusers do not need to use coercion to be considered traffickers. Yet, even though trafficking affects both sexes, it affects them differently.
Human trafficking disproportionately impacts women. In 2014, over 90% of sex trafficking cases reported to the the National Human Trafficking Resource Center in 2014 involved women, while men were only involved about 5% of the time. Yet although it is evident that human trafficking affects men and women differently, the reason for this is not as clear. Some anti-trafficking groups focus on ending the demand for sex, and imply that men's demand for sexual services is stronger than women's. Perhaps traffickers more often perceive women to be vulnerable and therefore target them for forced labor. While these factors are likely at play within individual trafficking cases or trafficking networks, I wondered how structural inequities between men and women might impact the prevalence of human trafficking and modern slavery.
 |
| National Human Trafficking Resource Center |
The key question that I initially sought out to answer was "What is the effect of differing levels of opportunity for women on the prevalence of human trafficking?" I hoped to shed some light on the complexity of this seemingly simple question. I theorized that states where women had more economic and political rights would have less slavery. In theory, women with the ability to chose a job without the consent of a male relative, obtain equal pay for equal work, and relocate freely will be at a lower risk for trafficking than women who do not have these rights because traffickers would be less able to keep victims in their situations [Note 1]. Similarly, women with more political rights such as the right to vote, run for office, and petition the government might be able construct policy environments where fewer women are enslaved. This question is vitally important because answering it could help anti-trafficking advocates develop interventions to increase communities' resilience trafficking.
I constructed a statistical model using data provided by the Walk Free Foundation's 2014 Global Slavery Index (GSI). The GSI attempts to estimate the global prevalence modern slavery by extrapolating country-level estimates of the amount of trafficking to similar countries. The GSI is based on a number of sources, including random surveys that coded qualitative data about the subject's family's slavery experiences in the past five years [Note 2]. Although this extrapolation approach is problematic, and has been criticized, the GSI provides the only relatively complete estimate of the prevalence of trafficking around the world, even it it is only a very rough estimate, at best. The researchers who developed the GSI helpfully explain their methodological choices and also provide recent data for dozens of other potential predictors of trafficking. The data on women's rights come from the CIRI Human Rights Data Project and is included in the GSI data set [Note 3].
A box plot with some of the most extreme values removed (see also Figure 1b) provides a good first visualization of the relationship between women's economic rights and the prevalence of slavery. States with higher economic rights have less slavery than states where women have low economic rights or no economic rights (Level 1).
This second visualization also shows the relationship between slavery and women's economic rights. It attempts to fit a regression line to the data which have been "jittered" to differentiate the points from each other. This particular regression model (the dashed line) matches the non-parametric lowess fit line reasonably well, but still has trouble describing everything all the points in the graph.
The full statistical model shown below, estimates the impact that women's economic rights (Column 1) and women's political rights (Column 2) have on the prevalence of human trafficking holding a number of other variables (economic equality, income, development, corruption, and governance) constant [Note 4, 5].
The model explains that while increases in economic rights for women are associated with a lower percentage of enslaved people, the same effect was not borne out for increases in women's political rights [Note 6]. The predicted effect of Women's Economic Rights was relatively small: a one-unit increase in the 1-4 scale of Women's Economic Rights was correlated with a .201 unit decrease in the percent of trafficking in a country in the model that only included Women's Economic Rights.
While there is a correlation between Women's economic rights and the prevalence of trafficking, that alone does not constitute a causal relationship. There still might be confounding variables that affect both the prevalence of slavery and Women's Economic Rights (such as human development, see appendix) that this model cannot account for. Reverse causality is also a concern here - respondents to the Women's Economic Rights survey might have felt like women have more economic rights in part because of their freedom from slavery.
More investigation would help to clarify these relationships. Although the statistical model above could not conclude that a relationship between human trafficking and women's rights existed, controlling for other relevant variables, it was able to suggest that Women's Economic Rights may be correlated with the prevalence of trafficking. This may be an important piece that can help anti-trafficking advocates develop approaches that seek to prevent trafficking, perhaps as part of a holistic development effort..
Appendix:
[Note 1]
Increased economic rights could also have the reverse effect - women would have more freedom to seek employment far away, where they lave less social support from their families and communities, potentially making them more vulnerable to traffickers.
[Note 2]
For example, if a respondent in Kenya said that his sister had been forced into domestic servitude in Canada last year, that victim would count towards the estimate for Canada.
[Note 3]
The Women's Political Rights and Women's Economic Rights measurement originally ran from 0 to 3 and were each resealed to a 1-4 scale with 1 representing no rights for women.
[Note 4]
The variable "Gini Coefficient (Flipped)" is a measure of income equality. It is the Gini coefficient, a measure of income in flipped into a measurement of economic equality. This measure ranges from a value of 0, which indicates perfect income inequality to 100 which represents perfect equality.
[Note 5]
The figures employ a quadratically transformed linear regression prediction instead of a standard OLS prediction because the outlier values (~4.0%) skew the relationships into a parabolic ones. This is done by regressing the Y variable (the prevalence of slavery) on the X variable (women's economic rights or HDI) and the X variable squared. The final model however employs standard OLS regression because the quadratic transformation did not yield a statistically significant model (see Table 3 below).
[Note 6]
This effect can also be seen in Figure 1c, where it is apparent that the states where women have some of the most political rights are the same states where the prevalence of slavery is the highest, a trend that run contrary to the trend seen with women's economic rights.
[Note 7]
Although logging GDP Per Capita did produce a statistically significant result in a potential model (shown below,) the coefficient was still 0.000 it did not make a strong contribution to the explanatory power of the rest of the model, judging by the relative differences in the R squared statistics between models 5 and 6 in Table 3.
Table 2. Summary Statistics
An interesting model, shown above, that might prove to be an effective control variable for another model is the influence of Human Development on Trafficking. In the final model (Column 5 of Table 1) Human Development was statically significant and when regressed against the prevalence of slavery by itself, a one unit increase in Human Development was associated with a 1.522 unit drop in the prevalence of human trafficking.
Big Picture
ReplyDelete1. Clear research?
Yes, it is very clear that you are trying to identify the causes of the human trafficking and although you mentioned the importance of incorporating a gender variable since most trafficking victims are women, it is not very clear how you ended up using the women’s right variable as the independent variable.
IV: Women's Economic Rights (CIRI)
DV: Estimated % of the Population Enslaved
2. Who cares? It is very clear that it is an important question because of the urgency of the problem. You did a good job in explaining the importance of the problem.
3. Answered the question? Yes, you did answer the question. Women’s economic rights does not have a significant effect on the dependable variable while controlling for the other variables. Alone however, the variable did have a significant positive effect. You could have been a little more clear with the explanation. I understand what you are trying to communicate with the interpretation of results because I’m in the class and it has been hard to interpret my own results too. But if I weren’t in the class, I imagine it would have been a bit hard to understand.
Nuts and Bolts
4. Bivariate graph: there are bivariate graphs but they only show the lowess and the confidence interval, not the lfit
5. Yes
6. Yes, they look really good. My only comment would be to align both graphs. Also, I need to do this myself, try to make the tables maybe smaller and easier to read.
Modelling and inference
7. Evidence of transformation as needed?
8. Cofounds? Yes, it was a very good idea to include gini coefficients as a control variable as well as to include corruption perception index and women political rights. I think you covered most of your cofounds with the Governance Worldwide variable.
9. Reverse causality? Yes, you used a large and well chosen selection of control variables.
10. Correct interpretation free of jargon? The interpretation is correct. I would probably also comment in other variables that were not significant that could contribute to the understanding of the model. I think overall the general audience would have understood, however it could have a bit more of a simple language.
Overall comments:
I think you did a great job with identifying cofounds and selecting your control variables. The intro was great because it explained really well why the problem of human trafficking is so important. I would recommend to work a bit more in the interpretation of the results and make the tables a bit more easy to read. Great work!
I think this is a super interesting topic idea. Obviously human trafficking is a major issue in all countries and it is difficult to understand why the problem persists. I feel that your DV is clear: % of the population enslaved. It was less clear what was your IV(s), why you chose them, and how they were measured. You mentioned CIRI (economic rights) but also looked at political rights. The bivariate regression figure showed Political Rights instead of Economic Rights, so I was confused as to which was your primary IV. I would also include the parametric line to show that the linear regression was appropriate and there was no significant skew. I would suggest bringing up the tscores to be right under the beta coefficient so it is more clear. Other than that the tables and figures were clear. The linear assumption was clear and the analysis of the regression table was easy to understand. I would have liked to have more information on why you chose those particular controls and how they were measured. You had many controls, so it would be good to tell the reader why they are all important to include in the model. You can also include the outcomes of these variables in your final analysis, what do they mean within the model? I really like the idea of the topic of looking into if increasing women's economic rights could reduce the prevalence of human trafficking. This could be a great way to suggest policies to improve women's livelihoods, with a little more clarification about the variables and expansion of your analysis, this will be a great post!
ReplyDeleteBig Picture:
ReplyDeleteResearch question is clear as is the DV. You identify the IV as women's rights but then in your analysis, divide women's rights into political and economoic rights (I'm guessing because that's how it's measured in the data you gathered). It may be more useful to find a single measurment for women's rights rather than analyzing them separately (maybe you can create a single measurment by adding the political and economic rights scores together). Also, you identify your variables sevaral paragraphs in. At the very least, I would identify your variables before you start talking about how you measure human trafficking/slavery (also, on measuring your variables, I know very little about the measurement you chose but I would like to know if when you meausre the % of the population that has been trafficked, are people counted based on their country of origin or where they end up. For example, if a Moldovan woman is trafficked from Moldova to the United States, is she counted in Moldova's trafficked numbers or the US's numbers). I'm also not sure if you have a hypothesis (do you predict human trafficking to go up or down when women's rights increase).
You have a thorough and good discussion of why this is an important question and you answer the question well.
Nuts and Bolts
There is a bivariate graph but no non-parametric fit.
For the summary statistics table, I would linclude a short description of your variables, especially saying what high scores on the various variables mean (e.g., is a high flipped Gini coefficient perfect equality or perfect inequality?)
In your graphs, I find it a little confusing that your title mentions "trafficking" while you used "enslaved" in the rest of your graph. I understand that human trafficking=slavery but, it's still a little confusing to use both in the same graph. Otherwise, I like the use of color and they are professionally formatted.
Modelling and Inference
It doesn't appear that you logged any of your variables, but it is unclear if you needed to (perhaps GDP per capita?) You also control for a number of important confounds. I wonder if economic opportunity or unemployment is also relevant (this may be separate from economic rights, that is, not just can women work but are there jobs available?)
You mention the problem of reverse causality but I'm not sure that you fully address it (though you do mention that your findings are correlated not causal).
The interpretation of the results seem correct and free of jargon.