Although I agree with Chris on a number of points I think that a
critical point is being overlooked.  To call research or results
"scientific" is vague and indeterminate.  A more appropriate
distinction is whether the research is experimental or nonexperimental.
And if it is experimental research, to what degree are we confident
that internal validity (i.e., that we have established causal relationship
between variables) has not been compromised.  Ultimately,
"scientific research" is concerned with being able to make statements
about the relationships among variables that are true.  Experimental
research is to be preferred but sometimes we can only do correlational
research.

Much of the research on the effect of salt on health is correlational
and we all know the problems associated with that type of research.
Below is a little SPSS program with inline data that came from a
1991 NY Times article that reported on a cross-national study of
sodium intake and its effect on systolic and diastolic blood pressure.
I have used this dataset in statistics classes to illustrate correlational
research and its problems.  One can copy the text into a syntax window
and run it.  The key results to notice would beL

(1)  When all nations are included, there is a significant correlation
between sodium intake (as measure by sodium in urine) and the
blood pressure measures (i.e., systolic and diastolic BP).

(2)  When one leaves out the Yanomamo Indians from Brazil
who have the lowest urinary sodium, the correlation drop in
value and become nonsignificant.  NOTE:  the data I have
may not represent the complete data in the original research
but the point of the NY Times article was the elimination
of outliers produces nonsignificant results.

(3) Exclude the nations next two lowest sodium intake levels
(i.e, New Guinea and Kenya), the correlation continues to drop.
The correlation between sodium intake and diastolic bp turns
negative when Brazil, New Guinea, and Kenya are excluded.
Statistical power is not the issue though representativeness
of the sample might be.

What does this "prove"?  Well, one could say that if you leave
out nations with extremely low sodium intake, there is no relationship
between sodium intake and blood pressure.  But I wonder to
what extent has this result been replicated -- I would like to see
many replications before I would say there is no relationship.

As I tell my research methods course on the first day of class:
"Science is hard, m'kay?"

-Mike Palij
New York Unviersity
[email protected]

title 'New York Times (1991):  Sodium and Blood Pressure Data'.
data list  /
  rank  1-3
  center  4-23  (a)
  bp.sys  24-28 (1)
   bp.dia 30-33 (1)
   na.uri 36-40 (1).
var labels rank 'Rank based on Sodium usage-Highest to Lowest'
 center 'City/Country study conducted'
 bp.sys 'Systolic Blood Pressure'
 bp.dia 'Diastolic Blood Pressure'
 na.uri 'Sodium in Urine'.
*1234567890123456789012345678901234567890.
begin data.
27 Argentina           115.1 72.8  155.8
41 Belgium (Chareroi)  125.1 77.8  141.2
52 Brazil (Yanomamo)    96.0 60.6    9.0
31 Canada (Labrador)   118.1 73.0  151.2
01 China (Tianjin)     119.4 70.5  245.6
06 Columbia            120.7 67.6  201.4
43 Denmark             124.0 81.6  140.2
28 Finland (Turku)     127.0 78.4  154.8
18 Germany Heidelberg  117.6 73.6  172.9
08 Hungary             125.7 80.3  198.3
26 India (New Delhi)   113.7 73.9  160.6
23 Italy (Naples)      114.8 72.5  167.7
02 Japan (Toyama)      117.3 72.1  212.4
49 Kenya               113.3 66.0   56.8
38 Mexico              112.7 73.6  144.1
50 Papua N. Guinea     108.0 62.7   36.8
13 Poland (Warsaw)     123.3 77.1  181.3
12 Portugal            132.6 76.6  181.9
03 S. Korea            112.2 71.9  208.2
25 Soviet Union        117.7 73.4  161.7
16 Spain (Manresa)     119.7 71.8  174.6
40 Taiwan              116.3 76.2  141.4
29 U.K. Birmingham     119.6 71.2  153.1
44 U.S.A. (Chicago)    114.0 70.5  140.1
37 U.S.A. (Hawaii)     123.3 73.8  144.1
end data.

desc bp.sys to na.uri.

corr var=bp.sys bp.dia with na.uri/statistics=desc.

sort cases by na.uri.

temp.
select if(na.uri gt 9.0).
*excluding Brazil from correlational analysis.
*subtitle "Corrs between Blood Pressure & Sodium intake - Excluding Brazil".
corr var=bp.sys bp.dia with na.uri/statistics=desc.

temp.
select if(na.uri gt 37.0).
*excluding Brazil from correlational analysis.
*subtitle "Corrs b/w BP & Sodium intake - Exc Brazil & New Guinea".
corr var=bp.sys bp.dia with na.uri/statistics=desc.

temp.
select if(na.uri gt 57.0).
*excluding Brazil from correlational analysis.
*subtitle "Corrs b/w BP & Na intake - Exc Brazil, New Guinea & Kenya".
corr var=bp.sys bp.dia with na.uri/statistics=desc.


---------------------- Original Message  --------------------------------------
On Sun, 03 Jun 2012 14:02:34 -0700, Christopher Green wrote:

This article is an excellent example, I think, of why so many people feel
justified in being skeptical of scientific results.
http://www.nytimes.com/2012/06/03/opinion/sunday/we-only-think-we-know-the-truth-about-salt.html?hp

Despite the caricature of "the method" often diagramed in textbooks, science is
never simply a question turned into a hypothesis, followed by relevant
observations, giving rise to definitive answers. First, the answers are always
tentative and liable to be modified or even reversed by future research (even
venerable ones like the alleged link between salt intake and hypertension).
Second, there is never a straight path from scientific results to the public.
There are any number of corporate, government, and media filters, none of which
may have the dissemination of "Truth" as their first and only priority. Third,
the science itself, these days, is all too often sponsored (and thereby
controlled) by interested corporate or gov't parties. As a New England Journal
of Medicine editor put it a few years ago, the medical journals have become
little more than branches of pharmaceutical companies' marketing campaigns.
Fourth, the explosive rise of fraud cases in science makes everyone wonder
whether a result that is too good (or bad) to be true might, in fact, be just
that.

We are all familiar, by now, of how corporations promote the idea of certain
conclusions remaining "controversial" as a tactic to keep people from making
advisable changes in their behavior. (If you are uncertain about this,
immediately read Conway & Oreskes' _Merchants of Doubt_ about the (linked)
coroporate campaigns to call into question the links between tobacco and cancer
and between CO2 emissions and climate change. Repeat until your uncertainties
fade.) One of the paradoxical outcomes of this is that when there really is
controversy about a certain scientific conclusion, parties that are interested
in there *not* being any question can intimidate doubters by threatening to
publicly accuse them of being part of a corporate "doubt" campaign (which is
what seems to be happening with putative salt-hypertension link now).

All this makes it virtually impossible for all but those who spend their lives
and careers learning all there is too learn about one of these topics (viz.,
scientists themselves) to know who to trust when there are a bunch of competing
claims in the public arena. (What is amazing is that I have found many
scientists to be just as clueless about topics outside of their own areas of
expertise as anyone you might meet in business or politics. The PhD does little
to inoculate one against nonsense.) The vast majority of people do not (nor
will they ever) read scientific journals. They won't even read Scientific
American. You'll be lucky if you can get them to read the science sections of
newspapers. And now that no one watches TV news anymore, but simply picks out
the stories they are interested in, according to headline, on the web, you
can't even forcefeed them 30-second science stories anymore, unless they want
to know about it in advance.

What are we (who believe in the potential of science give us answers) to do?

---
You are currently subscribed to tips as: [email protected].
To unsubscribe click here: 
http://fsulist.frostburg.edu/u?id=13090.68da6e6e5325aa33287ff385b70df5d5&n=T&l=tips&o=18160
or send a blank email to 
leave-18160-13090.68da6e6e5325aa33287ff385b70df...@fsulist.frostburg.edu

Reply via email to