!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!

CHANCE News 3.05
(21 March to 12 April 1994)

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!

Prepared by J. Laurie Snell, with help from Jeanne
Albert and William Peterson, as part of the CHANCE
Course Project supported by the National Science
Foundation and the New England Consortium for

Previous issues of CHANCE News can be found on our
chance gopher.  Just point your gopher to:
chance.dartmouth.edu

==========================================
An appropriate answer to the right problem is worth a
good deal more than an exact answer to an approximate
problem.
John Tukey
=========================================

COMING ATTRACTIONS: Should you quite taking your
vitamin pills?

Jim Hilton sent the following answer to my query about the probability of rain. (For a related story, see CHANCE News 3.04: 6 to 20 March, 1994,  "Ask Marylin" .)

Here in San Diego we have a local weekly paper (magazine) called The Reader. There is a section called "Straight From The Hip" by Matthew Alice. This section attempts to answer questions submitted by readers. The Thursday March 24, 1994 edition had the following question and reply: Question: If a weather forecaster says there is a 10% chance of rain today, 20% tonight, and 30% tomorrow, what happens if it rains early today? Does it stay 20% chance for tonight or is that gone? If it doesn't rain earlier today, do you add the 10% to the 20%, making it 30% chance of rain tonight? Or what? --Jeffrey Michael Austin Foxmore, La Jolla Answer: Put away the calculator and clear your mind of all extraneous thoughts. Weather predicting is not simple. Though you may be sorry you asked, cause you'll know less about the likelihood of rain in your particular neighborhood after I explain this than you did before. But if you're a faithful reader, maybe you're used to that situation. Anyway, if the National Weather Service says there's a 30% chance of rain in the San Diego area tomorrow, what they're saying is the odds are three in ten that someplace (or places) in the area will get wet. In some large geographical areas, the probability is expressed slightly differently. If the meteorologist sees a 60% chance of rain over half the area, he'll express the pre-iction as a 30% chance of rain. Doubly misleading, it seems to me. But weather predictors know the general public doesn't understand the system anyway so maybe it doesn't matter The National Weather Service's system of calculation is called the Precipitation Probability Program. It's based on computer models of weather patterns, and they've been refining the models since early 1970s. They now can predict weather for the next 72 hours with about 85% accuracy, though weather prediction is still a bit of an art as well as a science. -- Mathew Alice DISCUSSION QUESTION: What is the answer to the question that the reader asked? <<<========<< >>>>>==========>>

• ARTICLES ABSTRACTED

<<<========<<

The Boston Globe, 21 March 1994, p25. Scott Allen

This article discusses the extent to which the mechanisms of the greenhouse effect are still poorly understood, leading to a great deal of uncertainty associated with climate predictions.

Many climatologists still predict a 3 to 7 degree F increase in global temperatures over the next 100 years if nothing is done about current trends in greenhouse gas emissions, although these estimates are down a degree or so from those after the hot summer of 1988.

The article states, however, that "they [climatologists] concede that the forecast is so uncertain that it comes down to a value judgment." In other words, people must decide how much to do to avoid the risk. One surprising quote, from federal researcher Thomas Karl: "It's like if the forecast says there's a 60 percent chance of rain, do you take an umbrella?"

DISCUSSION QUESTION: What do you think of the umbrella analogy? <<<========<<

>>>>>==========>> "Probability Experts May Decide Pennsylvania Vote. "

New York Times, 11 April 1994, A15 Peter Passell

In a special election to fill a Senate vacancy in Pennsylvania's Second State Senatorial District in Philadelphia, Bruce Marks (Republican) received 19,691 votes on the voting machine and William Stinson (Democrat) received 19,127 votes. In absentee ballots Stinson received 1,391 votes and Marks 366, making Stinson the winner.

Charges of fraud in the collection of absentee ballots led Judge Clarence Newcomber of the Federal District Court in Philadelphia to decide whether to give the election to Marks or to call for a new election in this district. To help him decide, he has hired as consultant Professor Orley Ashenfelter, an econometrician from Princeton, and the Republicans and the Democrats have hired their own statisticians.

Ashenfelter's approach was to do a standard regression analysis on the difference between the voting outcomes by voting machine and absentee ballots using the results, of the 22 state senatorial contests in Philadelphia since 1982. He estimated that there was about a 6% chance of obtaining an outcome as extreme as the one being contested by normal chance fluctuations.

Paul Shannon, the statistician from the Wharton school hired by the Democrats, and Elizabeth Hotzman, who has had a lot of experience in analyzing voting behavior, challenge the appropriateness of projecting voter sentiment from historical relationships.

Statistician Brian Sullivan, working for the Republicans, analyzed the findings of a post-election poll of registered voters and estimated that 84 percent of the absentee ballots had been illegally solicited or cast and that if every legitimate voter had voted Democratic the Republican would still win. Shannon challenges the survey and the argument goes on and on.

DISCUSSION QUESTION

The article concludes with the remark that Lowell Finley, who specializes in election law, thinks the burden of choice should rest on those who would declare a winner in the absence of certainty. "In a democracy voters ought to have the last word not judges." Is Finley expressing an opinion about the current situation? If so, what is this opinion? <<<========<<

This is a report from the two fellows who pointed out, in a UK challenge to a conviction based on DNA evidence, the "prosecutor's fallacy".

This occurs when the prosecutor presents the match probability which is the probability that an individual will match, given that they are innocent, and suggests instead that it is the probability that an individual is innocent given a match. The latter probability is what the jury is interested in and can be much larger than the match probability. See Chance News 3.02. <<<========<<

Chicago Tribune, 12 April, 1994, News p. 4 Reuters

A study by a Yale physicist Robert Adair, published April 12 in the Proceedings of the National Academy of Science, concludes that electromagnetic fields emitted by power lines are far too weak to affect human cells and pose no health risk.

A discovery in 1992 that the human brain cells are rich in magnetite, a magnetic iron compound, led to the speculation that magnetite could react to the magnetic pull from power lines to explain a slightly higher occurrence of cancer and leukemia among those who live near power lines.

Adair remarks that the pull is negligible compared to the Earth's magnetic pull. To those who say the danger is due to the constant change in the magnetic field of alternating household currents, he replies that it would be impossible for human magetosomes, in which magnetite is encased, to move fast enough to respond to such weak alternating fields. <<<=== MORE =====<<

The Gazette (Montreal), 31 March 1994, A1 Graeme Hamilton

A study funded by two major power companies, one in Canada and one in France, has found a link between the magnetic fields generated by electric currents and an increased incidence of leukemia among utility workers. The study will be reported in the next issue of the American Journal of Epidemiology.

The researchers studied more than 223,000 people in Quebec, Ontario and France. They found that workers with above-average exposure to magnetic fields were three times more likely to develop acute myeloid leukemia than less-exposed workers. However, because of the small number of cases, the difference was not statistically significant. No link was found between the other 29 cancers studied.

They did not find that the frequency of leukemia multiplied as exposure increased above the average, leading them to write "Caution must be exercised in interpreting the present results as evidence of a causal association."

The study is considered an improvement over previous studies because readings were taken to measure the actual exposure of workers to the magnetic fields instead of being based on an employee's job description.

Experts remark that these results do not say much about the general population who are exposed to much weaker fields, but show that we should continue to worry about this problem

DISCUSSION QUESTION: How could the number of cases of leukemia affect statistical significance? <<<========<<

"Math Horizons" is a new journal published by the Mathematical Association of American to be read by undergraduate mathematics students - so we have some hope of understanding the articles.

This article shows how to test the hypothesis that a sequence of outcomes H (heads) or T (tails) is consistent with tossing a biased coin with probability p for H and 1-p for T on each toss, using the statistic of length of longest run of H's.

To start a H's run, other than on the first toss, we first need a T. In n tosses of a p-coin we expect about n(1-p) T's. A fraction p^r of these are expected to be followed by r or more H's. Thus we expect about n(1-p)p^r runs of length r or greater. One's intuition suggests that the most likely value for the longest run is when this is equal to 1. (We really can trust our intuition most of the time in probability.) This leads to the estimate for the most likely value for the longest run as the nearest integer to the logarithm of n(1-p) to the base 1/p.

For example, in 128 tosses of a fair coin we expect 64 T's, followed by 32 H's, followed by 16 H's, 8H's 4H's, 2H's, 1H leading to 6 as the most likely value for the maximum run length.

To get a 95% confidence interval for the maximum run length, you use the distribution for the difference between the expected value of the longest run and the observed longest run. Rather remarkably, this distribution does not depend on the number n of tosses, but only on the probabiliy p of getting an H.

The author uses these ideas to check if standard run records, such as Dimaggio's 56 consecutive games with a hit and the roulette wheel at Monte Carlo that stopped 26 times on a red number, are outside the 95% confidence limits for the maximum run in the appropriate long coin tossing sequence -- none are. <<<========<<

The Boston Globe, 31 March 1994, p1. Richard A. Knox

Here we find more on the findings of falsified data in the research study that led to the recommendation of lumpectomy and radiation as an equally effective but less disfiguring alternative to mastectomy in treating breast cancer.

On 29 March, Dr. Bernard Fisher was removed as head of the National Surgical Adjuvant Breast and Bowel Project (NSABP). Although Fisher himself has not been accused of any fraud, he has been criticized for not acting quickly enough to correct the public record when the falsified data was first discovered.

DISCUSSION QUESTIONS:

(1) Was it essential for Dr. Fisher to make an immediate announcement of the fraud, even though the original incident concerned only one of 5000 physicians in his project, and the tainted data did not affect the overall conclusions of the study?

(2) The issue seems to be the harm done in shaking the public's confidence in the study's recommendations. Do you think media coverage has exacerbated or helped to alleviate this problem?

(3) In the original study there were three treatment groups: total mastectomy (n = 590), lumpectomy (n = 636), and lumpectomy and irradiation (n =629). At the end of the observational period (average 81 months), the numbers alive with no evidence of disease were: total mastectomy 373 (63.2%), lumpectomy 371 (58.3%) and lumpectomy and irradiation 412 (65.5%).

(a) What would be a reasonable margin of error for thepercentages alive with no evidence of disease?

(b) What would be the effect on this margin or error if 16% of the data were removed? <<<========<<

>>>>>==========>> "Campus Racial Lines May Be Blurring: Study Counters"

Notion That Minorities Segregate Selves The Boston Globe, 5 April 1994, p1. Alice Dembner

This article describes the results of the first national study of student interactions across racial lines. While people often perceive strong tendencies towards segregation, particularly among minorities, the study offers evidence to the contrary. For example, the following are percentages of students, broken down into racial/ethnic group, who reported that they frequently studied with members of another group: 15% of whites, 49% of blacks, 60% of Asians and 72% of Chicanos.

A survey was conducted of a "nationally representative" sample of 6107 students. One of the investigators is quoted as saying that the goal of the study was to inject some real data into an otherwise highly politicized debate. The article raises several possible criticisms of the study's methodology. First, the results are based on student's own accounts of their behavior, not on outside observation. Also, the study did not assess whether the interaction was greater than would be expected, given the small numbers of minorities on most campuses. It was noted, however, that the interaction with other races did not increase significantly as the minority population grew. The wording of the questions does not allow one to determine whether minority students' interactions outside their own group was, in fact, with whites or with other minorities.

DISCUSSION QUESTION: What strategy -- other than simply asking students about their behavior -- could be used to get a representative view of group study habits? [I (Bill Peterson) went to a talk by Uri Treisman who reported having investigators move into dorms at Berkeley to find out whether students were studying math alone or in groups] <<<========<<

>>>>>==========>> Drug For Alzheimer's Disease Gets Mixed Reviews.

The Boston Globe, 6 April 1994 Richard Saltus

Richard Saltus reports that a study on tactrine, the only Federally approved drug for treating Alzheimer's disease, shows statistically significant benefits for some patients. The study, appearing in the Journal of the American Medical Association, was led by Dr. Margaret J. Knapp, a clinical scientist with the Parke-Davis division of the Warner-Lambert Company, the maker of tactrine.

Dr. Knapp and her colleagues said that 48% of patients who took the drug for 30 weeks benefited, compared to 23% who took a placebo for that period. Paul Solomon, an author of the study and a Williams College psychologist, said that "seven out of ten patients who completed the study either held their own or improved" on tests of memory and mental functioning. However, the drug only postpones the advance of the disease by about six months, after which decline continues.

During the study, 29% of the 663 patients involved developed liver abnormalities and had to discontinue taking tactrine for several weeks, but 87% of these patients were later able to resume treatment without the same liver difficulties.

In all, more than one-third of the patients dropped out of the study because of side effects which included, in addition to the liver abnormalities, nausea, vomiting, diarrhea, and abdominal pains. Solomon said the large dropout rate does not reflect actual practice, because researchers involved in the study were not allowed to relieve the side effects with other drugs.

A separate study found a correlation between Alzheimer's and economic status that suggests that more highly educated people with better jobs are at a lower risk of developing the disease.

DISCUSSION QUESTION:Should the dropout rate affect the findings of this study? <<<========<<

The Washington Post, 5 April 1994 Sandy Rovner

While high blood pressure can damage or even destroy the blood vessels that are essential to the kidney's proper functioning as a blood filter and cleanser, at the same time certain types of kidney disease, affecting about 25% of all kidney patients (the percentage is higher among African Americans) can raise blood pressure not only within the kidney, but throughout the body. Thus hypertension and kidney disease can each reinforce the other with potentially devastating results.

Although diabetes is the major cause of kidney disease in the population as a whole, among African Americans high blood pressure is the leading contributor, and both African and Native Americans have higher rates of kidney disease than the general population.

In a study of 840 patients published in the New England Journal of Medicine, researchers found that rigidly controlling the blood pressure of certain kidney patients slowed the progression of the disease. In patients whose blood pressure was kept at the lowest levels, the progression of the special form of disease was half as fast as that of the patients whose blood pressure was maintained at "normal" levels. The same treatment, however, was not significantly helpful for kidney patients as a whole.

At the same time, African Americans in the study appeared to progress to end-stage kidney disease (requiring either hemodialysis or a kidney transplant) much faster than Caucasians with the same blood pressure. A related study published in the March issue of the American Journal of Hypertension suggests that this higher rate among African Americans may be attributable to the tendency of blood pressure to remainhigh for this group while sleeping.

In the study, which involved black and white children and adolescents with normal and roughly equal daytime blood pressure, the pressure dropped during sleeping hours for whites, but remained constant for blacks. <<<========<<

>>>>>==========>> How now Dow Jones? Dubious Reporting on Why the Market Moves.

Columbia Journalism Review, March/April 1994 Trevor Nelson

Treveor Nelson writes that "the actual reporting for a daily market piece generally consists of a series of phone calls to analysts, who offer their opinion" on the reasons for daily fluctuations. However, John Dorfman, a staff writer at the Wall Street Journal, admits that "the truth is the market is such a vast place no one knows for sure why it rises and falls". In particular, the explanations given for day-to-day fluctuations may be nothing more than after-the-fact rationalizations.

Richard Sylla, a professor of economics at New York University, says that "most of us think the short- term fluctuations of the market--from hour-to-hour, minute- to- minute--are basically random."

But no one wants to report on market activity without offering a reason for its behavior. Rick Gladstone, deputy business editor at the AP in New York, concedes "we are obliged to say what might have been at work,"even if there is no obvious connection to other news. Thus reporting inevitably contains some "informed speculation" and other simplifications.

DISCUSSION QUESTIONS:

(1) What does it mean to say that the short-term fluctuations of the market are basically random?

(2) Can reporting on the market affect market behavior? <<<========<<

>>>>>==========>> A Disabilities Program That "Got Out Of Hand"

New York Times, 8 April 1994, A1 Michael Winerip

Emily Fisher Landau was 56 years old before she was diagnosed as dyslexic. This led her to give over \$2 million to the Dalton School in Manhattan to create a learning disability program. Over the years this paid for 14 full-and part-time learning specialists in Dalton's kindergarten-to-third grade students. It financed research to develop a screening test that would identify learning disabilities at an early age and provided money for Columbia University Teachers College to evaluate the program and publish the results.

The article describes how this program got out of hand. By 1992 the specialists had almost taken over some of the classes. Half of the students entering fourth grade had already received remedial help.

This all ended in the fall of 1992 when the five kindergarten teachers revolted and refused to use the screening test.

The Columbia researchers said that it was hard to say if those who had special learning assistance did better than they would have without the help since all those who were tested to see if they needed help were given it. The screening test is still be given for the purposes of the study, but the results are not being used by the teachers. <<<========<<

>>>>>==========>> Smoking Signals And Uncritical Reporting; Secondhand Doubts

Washington Times, 5 April, 1994, A14 Matthew Hoffman

A report of the Congressional Research Service "Cigarette Taxes to Fund Health Care Reform: An Economic Analysis", claims that the Environmental Protection Agency's "Passive Smoking" study, which concludes that 3,000 people die annually from exposure to other people's tobacco smoke, was seriously flawed. The report criticizes the agency for relying completely on other people's studies, changing the significant level in one of these studies to give the desired outcome, using subjective judgments as to which studies had statistical problems, and ignoring two important studies that did not show a significant effect.

These are pretty much the criticisms made by Alan Gross and refuted by Howard Rockette in the pair of articles pro and con that appear in the Winter 1994 issue of Chance Magazine -- see Chance News 2.20.

DISCUSSION QUESTION: What information would you need to have to judge if this attack on the EPA decision by the CRS is justified? <<<========<<

>>>>>==========>> Investigating A Medical Maze: Virus Transmission in Surgery

New York Times, 22 March 1994, C3 Lawrence K. Altman

Transmission of hepatitis from a patient to a health care worker is fairly common, but since 1987, there have been only seven clusters reported where the transmission is from doctor to patient. The most recent cluster occurred at two hospitals associated with the University of California at Los Angeles, where it was found that 18 of the patients of a surgeon, who had operated on 142 patients, had developed hepatitis, while none of at least 155 patients operated on by other surgeons had developed the disease.

The surgeon involved had apparently been infected by a patient. It appeared that he had followed standard procedures for his surgery, but reported that when he operated, on occasion, his fingers would become irritated; and he speculated that perhaps the pressure of tying sutures might have something to do with transmission. In a simulation of his surgical technique, he tied knots for an hour. Then the inside of each glove was washed and the washing liquid sent to a lab for analysis. The virus was found in the liquid, offering a theory on how it was transmitted. <<<========<<

>>>>>==========>> Study Finds Hope in Immune Therapy for Cancer

New York Times, 23 March 1994, A15 Gina Kolata

The current issue of the Journal of the American Medical Association reports on a study to see if certain cancers can be attacked indirectly by giving patients interleukin 2 with the aim of stimulating white blood cells. Success with interleukin 2 can come only through improving the immune system, since it has no effect on the cancer cells themselves.

In this study, 283 patients who had the skin cancer melanoma or kidney cancer that had spread, and who were no longer helped by conventional treatments, were treated with interleukin 2. Their progress was monitored for several years. Of the 283 patients, 8 were apparently cured and an additional 10 percent of the melanoma patients and 13 percent of the kidney cancer patients had their tumors shrink temporarily. The report noted that side effects can harm the heart, the kidneys, the liver and the nervous system, and that three patients died after receiving interleukin 2. Experts say the importance of the study is that it appears to indicate this indirect approach has some positive effect, but the results are not very striking and the side effects are worrisome. And the fact that it was not a controlled study makes it hard to be sure that the successes were the result of the treatment. <<<========<<

>>>>>==========>> Gene Experiment to Reverse Inherited Disease is Working

New York Times, 1 April 1994, A1 Natalie Angier

There are a number of studies underway to see if certain inherited diseases can be prevented by modifying the genes of a person with gene defects that cause the disease.

The April 4 issue of Nature reports the first success for this kind of genetic engineering. Researchers report that they have partly corrected a cholesterol disorder, called familial hypercholesterolemia, by supplying a 30-year old women with copies of a gene she lacks. Results to be published on Friday in the journal "Nature Genetics", researchers said that, in the two years since adding the new genes, the patient's blood cholesterol has fallen from 525 milligrams per deciliter to 410 (still twice the normal level), and the desirable cholesterol level has increased -- though it is still below 200.

Researchers hope they have brought the unidentified woman a reprieve from the early death she faced from premature heart disease. She suffered a heart attack at age 16, underwent coronary bypass surgery at age 24, and still her heart disease worsened. Her two brothers died of the high-cholesterol disorder in their 20s.

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! CHANCE News 3.05 (21 March to 12 April 1994) !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! Please send suggestions to: dart.chance@dartmouth.edu

>>>==========>>|<<==========<<< .