Sandbox

Crowd estimates

Crowd counts: When The Post did it right
by Andy Alexander, Washington Post, Ombudsman Blog, November 10, 2010

Recent months have seen two rallies on the National Mall by media personalities: the "Restoring Honor" rally by Glenn Beck of Fox News, and the "Rally to Restore Sanity and/or Fear" by Jon Stewart and Stephen Colbert of Comedy Central. For any rally, the estimated crowd size is a widely quoted--and disputed--measure of success. In the wake of controversy surrounding Louis Farrakhan's "Million Man March" in 1995, the National Park Service stopped making official estimates.

Beck claimed that his rally attracted 300,000 to 500,000. CBS News announced an estimate of 87,000, based on an analysis of aerial photos they had commissioned from AirPhotosLive.com. The same firm provided an estimate of 215,000 for the Stewart-Colbert rally.

Absent the politics, the mathematics for converting an aerial photo to a crowd estimate is relatively simple: one simply divides occupied area by an estimate of crowd density (people per unit area). The Ombudsman article here describes how the Post produced its estimate of 1 million for the inauguration of President Obama. A satellite picture with accompanying analysis is available here. As described in the article, this analysis

...not only provided the satellite image online, but The Post also gave readers a step-by-step explanation of how the photo was analyzed in order to make an estimate. It noted those who were not included (people along the parade route or in side streets, for instance). Then it explained what experts calculate to be the area occupied by a single person in a crowd that is jam-packed (2.5 square feet), reasonably packed (up to 4.5 square feet) or loosely packed (10 square feet). The giant crowd was divided into zones, and each was assigned a density. Next, The Post noted what percentage of each zone was occupied by people. Finally, all the zones were totaled to produce the estimate.

Submitted by Bill Peterson

Forecasting Campaign 2010

T'he Wall Street Journal
October 23, 2010
The Numbers Guy (Carl Blank)] says.

My print column this week examines the tough task facing forecasters of the House of Representatives races this fall. Political scientists, veteran pundits and number crunchers have advanced the field considerably from several decades ago, thanks to freely available poll data and advanced computer models, but they still run up against inherent challenges of quantifying public mood and the small number of observations: There is only one midterm election every four years, after all.

Of course it is interesting to see how the predictions were

freakonomics.blogs.nytimes.

the prediction of the model is that the Democrats will end up with 199 seats, or a loss of 56 seats.

Submitted by Laurie Snell
To be continued

Placebo contents

According to Wikipedia,

A placebo (Latin: I shall please) is a sham or simulated medical intervention that can produce a (perceived or actual) improvement, called a placebo effect.

[The origin for the term placebo] dates back to a Latin translation of the Bible by Jerome. It was first used in a medicinal context in the 18th century. In 1785 it was defined as a "commonplace method or medicine" and in 1811 it was defined as "any medicine adapted more to please than to benefit the patient," sometimes with a derogatory implication.

Nowadays, so entrenched is the necessity of a comparison to a placebo, any medical treatment trial without a control arm containing a placebo would be viewed skeptically both statistically and medicinally. But, consider the provocative title of Golomb: “What's in Placebos: Who Knows? Analysis of Randomized, Controlled Trials.” Surprisingly,

No regulations govern placebo composition. The composition of placebos can influence trial outcomes and merits reporting.

The study looked at four prestigious journals: New England Journal of Medicine, JAMA, The Lancet and Annals of Internal Medicine. Included were 176 journal articles:

Most studies did not disclose the composition of the study placebo. Disclosure was less common for pills than for injections and other treatments (8.2% vs. 26.7%; P = 0.002).

Conclusion: Placebos were seldom described in randomized, controlled trials of pills or capsules. Because the nature of the placebo can influence trial outcomes, placebo formulation should be disclosed in reports of placebo-controlled trials.

Discussion

1. Golomb cites the following example: “For instance, olive oil and corn oil have been used as the placebo in trials of cholesterol-lowering drugs.” Under the assumption that these oils might be beneficial, rather than inert, why does this understate the positive benefit of the treatment?

2. Golomb cites another example where a lactose placebo was used in a gastrointestinal trial. Under the assumption that the lactose was harmful, why does this overstate the positive benefit of the treatment?

3. Why is modern communication, e.g., the internet, facebook, etc., a cause for concern when conducting a randomized control trial (with or without a placebo arm)?

4. Golomb further alleges, “failure to describe placebo ingredients breaches basic scientific standards of rigor.” Why would describing the placebo ingredients “disadvantage” the “publication prospects” of the researchers and “disadvantage” the publisher of the particular journal?

5. Medicine is not the only area of endeavor which should require a placebo arm. Name some others.

6. For the record, the term nocebo (I will harm) was coined in 1961 and refers to the negative effects of a sham or simulated medical intervention. An example sometimes given is a patient dying of fright due to being bitten by a non-poisonous snake. Give examples of some other nocebos.

7. Why would prayer be considered a placebo? Why would prayer be considered a nocebo?

8. Exorcism has been in the news lately. Justify exorcism as a treatment, a placebo or a nocebo.

Submitted by Paul Alper