You're assuming direct causality. Think of the practically limitless number of major confounding variables here that are practically impossible to control for in a reasonable way. At the most fundamental think of the distinction between nations that legalize prostitution versus those that do not. There tends to be a very different system of ethos between these nations and this is going to trickle down to every single aspect of society.
So for instance I think it's extremely likely that you can show that legalizing certain drugs (such as marijuana) can be shown to 'cause' increases in sex trafficking. Of course it almost certainly does not but I'm making the educated assumption that nations that have legalized prostitution are going to have legalized drugs of various sorts more liberally than nations that still have criminalized prostitution. And so, lo and behold, we can show a correlation between legalizing drugs and sex trafficking if we assume that paper is correct that a link exists between sex trafficking and legalized prostitution.
Though indeed even the link itself should be called into question. For instance could the cause not be that legalization enables greater scrutiny over workers and thus the legalization itself helps to increase the detection efficiency of unlawful trafficking? Differences in what you're measuring mean nothing unless you are measuring things in the exact same way, with the exact same 'resolution'. Again you can try to control for things like this, but the controls are subject to just as many problems.
---
Maybe the most important issue here is that this is an issue that is incredibly difficult to do right even when you are 100% impartial and objective analysis is your one and only purpose. In reality people are biased, and a negative result means you basically just wasted however many months researching this topic. And this is a topic where you could, even without malice or intent, show anything to be true if you wanted by just seeking out slightly different data, or controlling for confounding issues in slightly different ways.