Health Care reform has been a hot topic where statistics have been used to: Abuse anecdotal math to falsify the truth and truthify falsehood.(This is the sixth of the series on misuse of statistics. For the earlier postings, click: 1-Going to St. Ives, 2-Playing Percentages, 3-Correlation and Causation, 4-Fun with the Normal Curve, 5-Global Warming.)
In a December 2009 posting, I pointed out that the map of per capita Medicare spending by county in the US looked a lot like the political division between the "Blue Counties" (Democrats, L-Minds) and the "Red Counties" (Republicans, C-Minds).
Since counties are so numerous and therefore confusing, I used Congressional Budget Office 2004 statistics of per capita Medicare spending on a statewide basis to show that the top five Highest Spending States tended to be Blue States and the top five Lowest Spending States tended to be Red States.
(My stated purpose -agenda if you like- was to indicate that Liberals consume an outsized share of the common pot of health care resources, as compared to Conservatives who take a smaller piece of the pie per capita.)
The figure above shows the CBO bar graph that orders the states by raw Medicare Spending per capita, from lowest (Utah) at the top to highest (Massachusetts) at the bottom. The box on the right is a magnification of the top and bottom of the bar graph. As indicated, the lowest spending states have consumed $4,000 to $4,800 per beneficiary. The highest spending states consumed $6,300 to $6,700 per capita. [Click on figure for a larger version.] I consider the above to be a good, clear use of statistics. The bar graph, however, is a bit misleading. Notice that the horzontal axis does not start at $0, but rather at $3,500, which is visually misleading because it seems to show that the highest spending states consume over twice as much as the lowest spending. Actually, they consume only about 150% as much.
Lesson Learned: When viewing graphs, be very suspicious when the axes do not start at zero.
In a comment, Howard* provided a link to Hopson and Rettenmaier (NCPA) that, it seemed to him, "... finds high cost states are Louisiana, Maryland, Oklahoma, Texas, Kansas, and low cost states are New York, Vermont, New Mexico, Virginia, Iowa. That’s closer to the opposite correlation."
Indeed, on page 37 of NPCA, as shown in the figure below, they do identify the ten states listed by Howard as the five "Low-Cost States" and five "High-Cost States". The NCPA study is based on the exact same Medicare data from 2004/2005. What is going on??? [Click on figure for a larger version.]
The figure above uses the CBO bar graph and shows that the NCPA list does not match at all. New York for example, is listed as a "Low-Cost State" by NCPA but, acording to the CBO bar graph, it is the third highest cost! Texas, on the other hand, that is sixth on the CBO list, indicating low cost, is listed as a "High-Cost State" by NCPA! Yikes!
Part of the answer is in the caption of the NCPA Table VII: "Summary of High- and Low-Cost States After Adjusting for Observable County Differences" Oh, this is not raw data but it has been ADJUSTED.
OK, how was it adjusted? That is not obvious without reading the whole report. They have taken the following into account: "Percent Black", Percent Hispanic", "Percent Female". Each of these factors tend to lean heavily Democratic in their voting patterns, indicating they tend to be L-Minds.
So, NCPA has adjusted the data to reduce the apparent costs in states with Democratic constituencies and effectively increase the apparent costs in states with Republican consitituencies. (Of course NCPA would say they are adjusting for clear patterns of medical need. Blacks, Hispanics and Females seem to need - or at least consume- more medical care than Males and non-Minorities on a per capita basis.)
The above adjustment could be described as reasonable and scientific. Some identifiable groups do need more health care for good reasons that are not their fault. However, NCPA has done some other adjusting that is totally absurd, tricky, and non-scientific. Read on!
Notice the header on the second column: "% of Counties in the Top Quintile". They have treated counties equally even though they vary greatly in population!
For example, I was born in Kings County (Brooklyn, NY) which has a population of 2,556,598. My professional life was spent in Tioga County (Newark Valley and Apalachin, NY) which has a population of 50,171. So, in computing that New York is a "Low-Cost State" they count Kings County and Tioga County equally, even though the ratio of their populations is over 50 to 1 !!!
Comparisons of spending for government programs should always be done based on the per capita statistics. CBO compares states and counties on a per capita basis. NCPA clearly has an agenda because they figure "Low-Spending States" on the basis of the percentage of counties in the highest and lowest quintiles. What a crock!
That is like a "50/50 Horse and Rabbit Stew, one Horse and one Rabbit".
Lesson Learned: Be very suspicious of "adjusted" data, especially when the person or group doing the "adjusting" has a particular agenda.
*I have the highest respect for Howard's grasp of scientific matters in general, and statistics in particular. I do not fault him for being led astray by the NCPA's statistical manipulations. They were far from obvious and NCPA seems to have taken special care to hide what they did (while remaining on the right side of "truth" by revealing their machinations in the "fine print").
7 comments:
Ira, as I said in my later post, the NCPA data are too course to support your Democrat/Republican correlation with Medicare costs. In that post I said, “I agree with JohnS such coarse statistics are meaningless. I gave two links, one of which was The Dartmouth Atlas of Health Care.
From a few checks, in counties in CA, Texas, Florida, NY, and Mass. I found the opposite correlation of politics and cost. I found my theory of correlation with Hospital regional costs more plausible, as well as better supported by a few spot checks.
I agree, my spot checks are not statistically significant, but there is just too much data here for me to waste time on. Anyway, Medicare costs are only one part of health care.
Perhaps your anti-L mind agenda will motivate you spend more time on the issue.
Howard, your linked Dartmouth map is, as you said, based on "Hospital Referral Region" which seem to be 3,436 geographic regions in the US served by one or more major hospitals. I agree this is a good way to group the data.
The NCPA Report was based on Counties and made no mention of Hospital Referral Regions.
You have not commented on what I consider faulty use of statistics by NCPA in ranking states on the basis of the number of Counties in high- and low-spending quintiles. How can anyone justify tallying tiny Tioga County NY equally with Kings County NY that has 50 times the population? Most other states also have counties with widely-divergent populations. NCPA has cooked up a "50/50 horse and rabbit stew - one horse and one rabbit!"
Note that your Dartmouth link also uses adjusted data based on sex and race which partially cancels out the higher-than-normal medical resources consumed by females, Blacks, and Hispanics.
Fortunately, the Dartmouth map has a button you can click to get their data by state. So I compared Dartmouth vs NCPA vs CBO for highest- and lowest-spending states.
Dartmouth shows 8 states in the highest-spending category (after adjustment): MA, NY, NJ, CA, NV, TX, LA and FL.
The highest-spending 5 for NCPA are LA, MD, OK, TX, and KS. (Only two of the NCPA match Dartmouth!)
According to the raw, unadjusted CBO data, the highest-spending 5 are MA, ME, NY, AK, CT. (This time a different two match Dartmouth.)
Note that, even with the sex/race adjustment, UT and ID are in Dartmouth's lowest-spending category and MA and NY are in the highest-spending category. This conflicts with the NCPA data that lists NY as one of the five lowest-spending states!
Dartmouth thus confirms what I wrote in the original posting that started this statistical discussion: "On a statewise basis, the people of Massachusetts and New York, for example, employ more than twice as many physicians per capita as Idaho and Utah and spend accordingly."
The above conflicts show what happens when data are "adjusted" differently by various groups that have particular agendas.
Ira Glickstein
The NCPA statistics is not the issue. I would probably agree with all your analysis if I had the time to follow it.
Let's stick to the issue. The issue you presented for discussion was "Do Liberals consume more health care than Conservatives?"
Because of (or in spite of) the enormous complexity and detail of health cost variables and the massive amounts of statistics, I still see no meaningful evidence in support of your hypothesis (which is already stated as a biased question).
OK, Howard, let us stick to the issue: "Do Liberals consume more health care than Conservatives?"
Yes, this is what you call "a biased question".
The maps in the main Topic seem to me to correlate Blue counties and states with greater consumption of Medicare resources per capita than Red jurisdictions. Assuming Blue tends towards L-minds and vice-versa for Red, the answer to my question is yes.
You linked to a number of sources that you thought indicate the reverse - that C-minds consume more.
I used statistics-based arguments to show your sources either supported my original answer -or- used misleading methods, or both.
**************************
OK, forget about the original maps I presented. Just look at two sources you linked. Both of them adjust the data on the basis of what Dartmouth calls "age-sex-race adjusted" (see the fine print under the map) and Hopson and Rettenmaier (NCPA) call "Hispanic, Black, and Other, with White serving as the excluded category. ... [and] percent of the population 65 and above who are female and the percentage who are 85 and above.." (see page 14).
It is quite obvious that Americans tend to consume more Medicare resources as they age. Until I read your links I did not know that non-Whites and females are similarly above-average consumers of Medicare.
Given the above (and I am sure you will correct me if I am wrong), and the exit polls that confirm Hispanics, Blacks, and females tend to vote more Blue (L-minded) than Whites and males, I think my answer to the original question is confirmed.
I do not necessarily put any blame on the L-minded constituency for consuming more Medicare resources. Perhaps their relatively higher consumption is due to other factors such as racial discrimination that leads to poverty and poor pre-natal nutrition. Perhaps the female anatomy is more complicated and therefore more subject to breakdown. Perhaps the causation is the other way around and their tendency to L-mindedness has its origin in their extra need for welfare and medical care.
Ira Glickstein
As I said in my original objection, “The regional differences between voting patterns is so great, I doubt if any coarse generalizations are meaningful.” I was not trying to show that the opposite of your speculation was true. Neither makes any statistical sense. The whole idea of statewide comparison is not meaningful because both politics and cost vary enormously within states. Anyway, it is the county level where the data are gathered. You know that by gerrymandering the data you can get any result you want. State divisions are essentially meaningless in this case.
Look at the Hospital Referral Region Dartmouth map. Populated states like California, Texas, New York, Massachussetts, Florida go from the least expensive to the most expensive within each state. It appears simply by eyeballing (agreed this is not good statistics) that the most expensive regions tend to be near cities with famous medical schools and hospitals. This confirms my rather dull hypothesis that the most expensive Medicare is where there is the most expensive medical care!
According to your hypothesis, you would have to show that the low-cost counties voted Republican and vice versa. For example, the cost gradation increasing gradually from western MA, central MA to the Boston area would have to correspond to gradual political increase of Republicans from west to east. Maybe that is the case, but I have no evidence of it.
I also think the cost problem is too complex for any meaningful simplistic theory. For example, see today’s NY Times article.
Ira, I just thought of how my “dull hypothesis” could be used in favor of your hypothesis:
The locations of leading medical centers are correlated with the location of medical schools. Medical schools are often located near leading academic institutions. Academic institutions tend to be liberal. Therefore liberals are largely responsible for the high cost of Medicare.
Howard wrote: "...liberals are largely responsible for the high cost of Medicare." I could go along with that! (Of course I realize Howard's conclusion was tongue-in-cheek :^)
I agree the data is questionable and the situation too complex for any simplistic theory.
Thanks for the link to the NYTimes piece that says UCLA spends, on average, $50,000 in the last six months of a patient's life while Mayo spends only half as much.
The UCLA doctor defended their higher costs with anecdotal evidence of a 71 year old heart patient who was given intensive treatment and recovered and went on to win the Olympic high jump. (OK, I made up the last part :^)
That illustrates the difference in L/C-Mind analysis. The C-Mind looks at hard numbers and concludes (as the UK nationalized health system has) that an extra year of life, "quality-adjusted", is worth only $30-$50K. Mayo meets that number while UCLA exceeds it by a factor of two or more. The L-Mind looks at a specific patient who might have died at Mayo but lived on at UCLA and concludes it was worth it for that particular patient. Yes, perhaps it was, but can we, the larger society afford it?
If we spend twice as much, on average, for each and every end-of-life patient, it will cost trillions. The benefit is questionable given the generally low-quality of life in our last years. It seems to me UCLA is getting rich by, on average, lengthening the period of pain and misery for the patients and their loved ones, at mostly public expense. Could limited resources be better spent on pre-natal care and young people and improve their lives? Unlimited end-of-life spending will destroy our economy, put our hard-working workers out of jobs, and ultimately strangle the tax stream that supports medical care.
The UK and every nationalized health system has come to the same conclusion about end-of-life care. (I am not in favor of nationalized health care, but I have said that I favor the NIH setting standards for end-of-life care, see End-of-life: Honest brokers (not Death Panels)).
Ira Glickstein
Post a Comment