http://istpp.org/crime_prevention/voodoo_rebuttal.html#note1
Someone brought up the name Maxwell Rainforth so I pulled up this
article. Some interesting points.
However, look at the graphs. In the link. See how many flaws you can
find in his argument?
Among some -- reaching conclusions based on:
Comparing a five year graph with a one year graph.
The five year graph is averaged. look at HRA scale. Much lower than
1993 scale, as would be expected -- crime growing over time. But the
averaging cancels out variations in each year. Without comparing the
individual 1988-1992 annual graphs, with their inherent fluctuations,
to the 1993 graph, his argument is baseless. The fact that he does not
do that annual to annual comparision makes me assume he is hiding "the
obvious" -- annual variations will be much greater than a five year
averaged one and disprove his point.
And of course, eye-balling, as he is asking us to do, is always good
to confirm reasonability of statistical findings. But it is not in
iteslf a statistical conclusion. He manually "centered" temperature on
top of crimes. Lots of lattitude in that to make it look "good". Thats
why statistical regression is used to find the "best" fit, not an
eyeballed fit.
And look at march, may and oct of 93. These months also have crime to
temperature variations, although not as big as the DC project. What
explains those variations. Unaccounted factors. As may well explain
the DC variation.
And later, he dismisses a doubling of the murder rate during the
course from 10/mo to 20/, as an "outlier". Thats convenient.
Thus far, I am not particularly impressed with the objectivity of this
guy.
Further for most of the experiement, there appears to have been, from
the graphs, about a 10% violent crime drop. How often does violent
crime fluctuate 10% or so from its trend. ALL THE TIME.
Since MUM is not making the data public, its hard to fully dissect
their conclusions and to try to replicate their results -- as well as
other results using differnt model specifications and assumptions.
But they certainly make the case themselves, that the ME if it exists
-- their graphs don't tell a compelling story, per above unexplained
factors, and natural fluctions -- it is small, and not what would be
expected from all the PR. A larger, 50% drop in violent crime, or a
complete cession of it, would be "predicted" by the theory. Or
perhaps, instead of 1000-2000? YF in DC, 10,000 or 50,000 would be
required to create such noticable effects.
========================================
A Rebuttal to "Voodoo Science"
by Maxwell Rainforth, Ph.D.
The Skeptical Inquirer recently published an article by Robert Park
("Voodoo Science and the Belief Gene" (Park 2000a) which he excerpted
from his book, "Voodoo Science: The Road from Foolishness to Fraud"
(Park 2000b). In his book and his article, Park lampooned the
scientific research of Dr. John Hagelin and collaborators (Hagelin
1994, 1999), myself included. Based on 41 previous studies, we
predicted publicly that a large group practicing the Transcendental
Meditation program would lower violent crime levels in Washington, DC,
by reducing stress and tension in society. During the 8-week
experiment in the summer of 1993, violent crimes against the person
(homicides, rapes, and assaults) decreased by 23% and closely tracked
the rise in the number of participating meditators. The results were
published in Social Indicators Research, a respected, peer-reviewed,
scientific journal (Hagelin 1999).
Park's objection to our use of time series analysis is not based on
any scientific argument, but merely echoes the comments of a reporter
regarding the use of time series analysis to predict levels of violent
crime: "How could you know what the rates would have been?" But, there
is no mystery here. Violent crime levels are predictable on the basis
of temperature a fact that is well known among criminologists, and
was clearly explained at the press conference to present the research
report that Park attended, and in both the report and the published paper.
Figure 1
This shows up clearly in Figure 1, which is a graph of the Washington
data over the five years prior to the experiment (1988-1992). When
average levels of temperature and average levels of homicides, rapes
and assaults are plotted over weeks of the year, the crime and
temperature curves are right on top of each other, if the vertical
axis scales are appropriately chosen. This shows that usually the
violent crime levels were directly proportional to temperature and
therefore that violent crime could be accurately predicted from the
previous pattern in the data. The same thing happens in the first
months of 1993, but then in the middle of the experimental period
(when the meditating group was approaching its maximum size) the
violent crime curve drops well below the temperature curve and stays
down for several weeks (see Figure 2). In other words, during the
experiment in 1993, a drop in violent crime was clearly evident in the
raw data, even without using time series analysis.1
Figure 2
