Posts
There is good evidence, foremost by Eric (corrected: Frank) Lichtenberg, that the introduction of new drugs improves health outcomes. But at the same time, there is a perception that new drugs increase health care costs, and maybe so more than it improves lives.
Rexford Santerre dismisses this idea showing that new drugs even reduce medical costs, primarily by making medical procedures less necessary. The empirical analysis is performed using aggregate data by health category, by regressing the change in expenditures on the previous year's change, the change in income and the number of new drugs. I find it heroic to make claims on causation based on such a "model". All you have is a correlation after controlling for income. If you want to say anything about causation, write down a proper model with testable hypotheses, and test those, not some random equation.
Another point where I think this paper misses the mark is in the presumption that drug price controls would be bad. The pharmaceutical industry enjoys some of the highest returns thanks to the protection given by patents. These firms need to be regulated for that very reason. In fact, it would be better to drop the patenting system in the first place: this would increase the competition to be ahead of competitors, instead of hampering progress with strategic patenting. Also, drug prices would be much lower.
Monday, March 2, 2009
Subscribe to:
Post Comments (Atom)
12 comments:
Studies that just do a linear regression on some variables with little thought do indeed discredit the profession. If at least they would be using nifty instruments like Levitt...
Interesting topic, very poor execution. Track it to see whether a journal publishes it. Medical journals might fall for it.
Give the author a break. This is pretty standard practice in health economics. Sad but true.
The Economic Logician should get his facts straight. The name is Frank and not Eric Lichtenberg.
This post is not just an indictment of this paper, but of health economics in general. In fact one could include much of international trade and development economics, as well as real estate economics in the same boat. There are entire journals that publish almost exclusively such research.
Research in health economics is much more econometrically careful than research in nearly every other field except perhaps labor economics. Take this paper for instance. At least the author replicated the same experiment: once with time series data from the U.S. and then again with a fixed effects model using a sample of OECD countries. How often is that done in economics? The variables were first differenced because of nonstationarity in the data. Amazingly both tests resulted in the same point estimate for the marginal effect of a new drug on overall costs.
I stand corrected on Frank Lichtenberg. However, I maintain that doing a linear regression does not make a model. There is nothing you can claim in terms of causality from it. Think in the terms of the Lucas critique. You need a proper structural model to claim anything, in particular that price controls would be bad. With or without price controls, the behavior of agents is different is ways an elasticity from a simple regression cannot recover. And adding controls or differencing does not resolve that.
I still find interesting that there is a negative correlation between health expenditures and the introduction of new drugs. I would have assumed the opposite.
This post seems to have hit a few nerves...
Great blog, by the way. I discovered it on Tim Worstall's blog, which discusses this blog. He concurs that one should think before making regressions.
Why does the author not answer to the post? Has he been alerted?
Kansan,the author has already responded to the post as anonymous (the second and third ones). I think research is all about extending what we know about a particular issue and not necessarily arriving at truth in one quick and giant step. This paper contributes to the literature by alerting many to the new drug offset theory and what time series data for the U.S. and a fixed effects model of OECD data has to offer in terms of supporting or refuting the theory. Most health care expenditure studies argue that income and new technologies, like new drugs, explain a large portion of any variation in health care spending. The test may not be perfect but no empirical test will be perfect unless a social experiment is conducted with both control and treatment groups. I would hope that others with a comparative advantage in other areas of research (structural modeling, econometrics) would take the analysis to the next stage. If these results are incorrect, which they may not be, the new research will reveal that. Perfections never can't anything done and like most things in life, we optimize using the resources we have. I find it amazing that both sets of observations arrive at the same point estimate on the new drug coefficient. I do not have any type of agenda with respect to supporting or bashing the pharmaceutical industry. I also think they make a lot of profits but wonder if static inefficiency may be the price we have to pay for dynamic efficiency. I just want to share my findings with others. I certainly hope others replicate what I have done with different data and approaches as they have done on other research projects I have worked on. Usually my results are supported by more ambitious studies. I invite economic logician and others to examine this same hypothesis using alternative approagches.
It is about time the charlatans of reduced form econometrics get called for their abuse of statistics.
Post a Comment