Why was the estimation wrong? And why by so much as 16%? Well, in complex systems, where we have to use statistics, we are going to often be off. In a case like this, better too many than too few. But 16% -- 6.3 million people -- is being off by a significant number. So what happened?
What happened is epidemiology studies a complex system. To completely understand the HIV epidemic, you have to understand the behavior,s especially the sexual behaviors, of people in many different cultures, plus migration patterns, plus politics in each region, plus education, plus logistics, plus various local medical practices, plus effects and effectiveness of medicines, etc. And that's just to understand what will happen assuming nothing changes except new people getting infected. THrow in the fact that cultural and religious practices, governments, and logistical capabilities can and do change, and throw in the development of new drugs that work much better than do older drugs, and that 16% doesn't look too bad. Most likely, the projections were made using linear models, which should never be used when dealing with complex systems. Indeed, if we draw a trend line on the graph from the Science article on this issue that includes the points from 1990-2000 or so, we see a linear growth rate to almost 40 million. What the data actually shows us a a flattening off. This is a more natural curve anyway, and it is what we would expect with the availability of new and better drugs. This shows what happens when you make linear assumptions about a nonlinear reality.
Africa in particular seems to be turning a corner, primarily due to education, which combatted risky behavior and various myths about HIV< including the myths that fat women are HIV negatives and that having sex with a virgin will cure you. The latter practice in particular no doubt contributed greatly to its spread in Africa. The results out of India, though, demonstrates the problems inherent in estimation, as the HIV rate of India as a whole was estimated based on samples from the cities. It turns out that denser population is correlated with higher concentrations of HIV infection, so those estimates could not be used to estimate the infection rate in rural areas. This would be like using the temperature readings from urban centers -- which are known to be heat islands -- to estimate global temperatures. In fact, I would only trust a temperature trend that excluded all cities, since we know that the larger a city becomes, the more heat it traps, meaning any trend that used city temperatures would naturally and necessarily trend upwards, or at least mask any downward trends, should they occur. In the case of HIV infection, cities must, of course, be used (that's where the people are), but care must be made to include rural areas as well. Thus, the estimate of 5.7 million in India dropped to 2.5 million as the number of sites that report HIV test results increased.
The lesson here is that when we are dealing with a complex system, whether it be the study of HIV epidemiology, ethology, an economy, the environment, the climate, or social behaviors, we have to expect that we will never have enough information to make accurate predictions, as we both won't have enough information about the state of the system now, and we can't take into consideration what will be created in the future to affect the system. This isn't to say we can't come to know something about the system, but it is to say that prescriptions to fix any system must be carefully thought through, and the models used must be nonlinear. With something like HIV, the system is a relatively easy one to deal with, since the goal is the complete annihilation of the system. But with any of the other systems mentioned, where the goal is to create a healthier system, the first thing we need to understand is what makes the system healthy. But if we can't even get it right when it come to a system we are trying to kill, what gives us the hubris to think we can know what to do to make a system healthy?
No comments:
Post a Comment