Medical Hypotheses (2005) 64, 665–668

http://intl.elsevierhealth.com/journals/mehy

Editorial

How to write up a hypothesis: the good, the bad

and the uglyq

Summary Medical Hypotheses exists to give ideas and speculations in medicine a fair hearing. Doing this is not easy.
Most conventional journals would regard some of what is published here as questionable, most referees would reject it
as ÔunprovenÕ. We have more liberal standards, for reasons we have presented before. But we still require ÔgoodÕ
science – logical argument that is supported by fact and comes to interesting, even useful, conclusions.
Alas, not everything received comes close to even this liberal standard. Since I joined the Editorial Board I have read
about 130 submissions to Medical Hypotheses. They range from exciting and insightful papers that might be substantial
advances in their field, to complete rubbish. I want to lay out what I believe to be the essence of the former so as to
avoid having to read so much of the latter.
Ó 2004 Elsevier Ltd. All rights reserved.

Clear argument 1. Cream cakes contain high levels of oxidized

lipids (particularly cholesterol).
A hypothesis is, by definition, something that is not 2. The oxidized lipid is absorbed like other food
yet proven. So you, the author, cannot lay out ta- ingredients.
bles of data or pictures of tissue to prove your 3. Oxidized lipids are taken up by cells.
case. You must lay out an argument in a clear 4. Within cells they cause cellular damage.
and convincing way. 5. Slowly dividing and non-dividing cells do not
The simplest way to do this is to make the steps take up oxidized lipids: only rapidly dividing
explicit. Usually, there are several statements in cells take them up.
even the simplest syllogism. Each step should 6. Cancer cells are the only cells in the body
either show that your conclusions are the only ones which divide rapidly.
consistent with the facts and with basic logic, or, if 7. So eating cream cakes will load up rapidly can-
there are other conclusions, why you prefer the cer cells with cell-killing oxidised lipids, and so
ones you have drawn. Do not simply ignore the kill them.
other conclusions. Your readers are not stupid,
and will take your dismissal of the obvious as evi- This allows the reader to see where you
dence that you are. started from, and come along with you every
So, lay out your argument in small, simple steps. step of the way. It also allows you to focus on
You think cream cakes are a cure for cancer? The the key elements. Steps 1–3 might be quite
argument might be: uncontroversial, so you can skip over them
quickly. Step 4 is well established in the litera-
ture, but might not be known so well outside a
q
The views expressed in this article are the personal opinions
specialist community, so it needs a paragraph
of the author, and do not necessarily reflect the views of other and some good references to establish it as true
Medical Hypotheses editors. (see below on references).

0306-9877/$ - see front matter Ó 2004 Elsevier Ltd. All rights reserved.
doi:10.1016/j.mehy.2004.10.003
666 Editorial

Step 5 however is quite unexpected. Why do I ment on water, very few insoluble compounds are
say that? What is the evidence? Is there counter- poisonous, the elements making up water are in
evidence? Is it open to question? Maybe this is the the most reactive 10% of the periodic table, and
central unknown in your hypothesis, in which case so on. But each statement is either irrelevant (like
pointing it out is of critical importance, because the last), or consistent with other hypotheses (that
measuring cellular uptake of oxidised lipid be- the toxins are poisonous, not the water). So none
comes a key test of your idea. of them make a useful contribution towards sug-
And Step 6 is just plain wrong, so what am I going gesting your hypothesis is more likely to be true
to do about that? Remove it, thatÕs what, and if that any other wild speculation.
that bursts my bubble, well, what distinguishes sci- Related is ignoring well-known facts that are not
ence from pseudoscience is the sound of bursting consistent with your hypothesis. I recall a long
bubbles. argument about an alternative structure for DNA
Most of the really ugly arguments I come across with one hypothesiser. The structure was very
can be un-masked by this approach. Is there a step beautiful, but suggested that DNA should fall into
that cannot be supported by a direct quote from a its component bases when you heated it. But it
standard text-book? Then you need to provide evi- does not – I have done the experiment, and human
dence. The further it is from what is taught at DNA gets more viscous as you heat it, not less. For
undergraduate-level science (today, not when you me that demolished the hypothesis.
were an undergraduate), the more you have to sup- Secondly, the argument ÔX is correlated with Y,
port it. Stating that the majority of chemical trans- and so must be linked to YÕ (although it is rarely sta-
formations occurring in living systems are catalysed ted this honestly). This is usually followed by a long
by enzymes needs no further amplification (indeed, and tortuous ÔmechanismÕ, which has been worked
it does not really need to be stated). Stating the out post-hoc to explain the correlation. This is also
opposite needs major back-up, as it flies in the a rubbish argument, if only on numerical grounds.
experience of something like 10,000,000 person- The world has uncountable facets. You can trawl
years of biomedical research over the last century. databases of numbers until the most improbably
This may seem like teaching grade school logic, correlations turn up: this is what ÔimprobableÕ
what my children learned in their Ôwhat is a fair means – something that will turn up eventually
testÕ sessions at six years of age. But literally doz- by chance. One can, for example, show an excel-
ens of submitted papers flit over steps in their lent correlation between the cumulative deaths
arguments that are simply wrong, either because
they do not recognise their own assumptions, or
(less charitably) because they are hoping I will
Amazon sales cause vCJD?
not recognise them.
Amazon.com net sales
250
Cumulative vCJD cases

Bad evidence 200

A critical part of this is evidence. You start from a 150

base of fact, take some logical steps, make a con-
clusion, and then see if the world fits with your 100
conclusion. The first and the last of these need evi-
dence. This is the main failing of many other 50
submissions.
First, a long list of examples which are consis-
0
tent with their hypothesis is not evidence that
1995 , Q1

1996 , Q1

1997 , Q1

1998 , Q1

1999 , Q1

2000 , Q1

2001 , Q1

2002 , Q1

2003 , Q1

2004 , Q1
, Q2
, Q3
, Q4
, Q2
, Q3
, Q4
, Q2
, Q3
, Q4
, Q2
, Q3
, Q4
, Q2
, Q3
, Q4
, Q2
, Q3
, Q4
, Q2
, Q3
, Q4
, Q2
, Q3
, Q4
, Q2
, Q3
, Q4
, Q2
, Q3
, Q4

the hypothesis is true, unless the evidence is not

consistent with the alternative hypothesis, and
particularly with the ÔmainstreamÕ hypothesis you
Figure 1 Amazon sales and cumulative vCJD deaths as a
are trying to supplant. This is because there are function of time. Amazon data from www.amazon.com.
innumerable facts in the world, and nearly all of vCJD data from http://www.cjd.ed.ac.uk/vcjdq.htm. X-
them are consistent with nearly everything. I could axis – date by quarter. Y-axis – vCJD cumulative cases/
provide a list of things that are consistent with the Amazon net sales in $10 Ms. Amazon data prior to 1997 is
hypothesis that water is poisonous: everyone who not available, as Amazon was a private company then.
drinks water dies, no one has done a safety assess- Note: the answer to the title is Ôno, of course not.Õ
Editorial 667

from vCJD and sales at internet bookseller ama- Good evidence

zon.com (Fig. 1). But this is meaningless unless
there was a reason to look at Amazon.com to start There are therefore four kinds of evidence you can
with as a potential cause of vCJD (and I emphasise, use to support your argument.
for the lawyers, that there is absolutely no such
reason). Otherwise all you are showing is that if
you look at enough statistics you will find two lines
Evidence that the problem you are talking
that look the same (which is what I did to find Fig. 1 about is real
– it took me about 30 min.).
Both of these types of arguments turn up reg- An amazing number of manuscripts fail to do this. A
ularly in newspapers and pressure group leaflets paper titled ‘‘Cream cakes as a curative treatment
to ÔproveÕ that vaccines cause autism, that micro- for the effect of low frequency electromagnetic
waves cause cancer, that vitamins from plants fields on human cognition’’ rather assumes that
are different to vitamins from chemists. You there is an effect of electromagnetic fields on
might be able to batter a journalist or a politi- brains. This is by no means established fact – so
cian into submission with such arguments, but establish it. And not just by reference to your last
readers of Medical Hypotheses are made of ster- paper in Medical Hypotheses (see below). More
ner stuff. And so are the editors, so please do mundanely, some recent evidence that the prob-
not waste my time! (Any or all of these proposi- lem is still outstanding is useful – a paper theoriz-
tions might be true, but this is not the way to ing on the cause of AIDS must first argue why HIV is
prove it.) not a cause of AIDS, or it has nothing to argue
Thirdly, there is the argument that goes ‘‘sci- about.
ence cannot explain X, I have an explanation for
X, therefore my explanation may be right’’. I hope Evidence that the foundations for your
this does not need to be debunked further here. argument is solid

This must refer to fact. Your previous speculation

that cream cakes cure cancer cannot now be ci-
Wrong evidence ted as evidence that it is true. There is no prob-
lem with you saying ‘‘I previously speculated that
Lastly, there is evidence that does not even exist, the preferential uptake of toxic oxidised choles-
which brings me to references. Papers have refer- terol by rapidly dividing cells might make cream
ences in them to save space when you present your cakes a cure for cancer.’’ But that was a specula-
evidence and your arguments. If Smith et al. did tion. Hypotheses built on speculation is called
wonderful work, rather than describe everything ÔphilosophyÕ or ÔfictionÕ, and is not what we are
they did, you just state their conclusions and refer about.
to Smith et al. The reader can then go and read
what Smith et al. actually said if they want to.
Evidence that your argument is logical
But your statement of what a paper is about
must be accurate – to say anything else is decep-
Do this by showing what it is, as above.
tion. Saying ‘‘Bains et al. (2006) showed that cream
cakes cure cancer’’ is misleading if the paper just
correlates cream cake consumption with cancer. Evidence that your conclusions are right
Saying ‘‘Wood is a well-known toxin (Smith and
Jones 1942)’’ when Smith and Jones were talking It is easy to get mesmerised by a chain of argu-
about the number of people killed by falling trees ments that leaves you stranded in a logical cull-
is downright dishonest. This example might seem de-sac. Your hypothesis has some implications. Do
ludicrous, but I have read several manuscripts that they agree with the real world? If not – trash the
make a striking assertion, and then cite a reference hypothesis, please, because the real world is the
that does not support it all – on one occasion it ultimate judge. If oxidised lipids cured cancer,
actually proved the opposite! Finding just one such then hamburgers should be even better than cream
horror in a paper means that you do not trust a sin- cakes, and the rise in fast food consumption should
gle thing the author says, and thus it destroys the be paralleled by a near-abolition of cancer. This
whole point of our good publishers felling trees has not happened. Ooops. Amazon.com trades
(entirely safely, of course) to get your opus into mainly in the US, but nearly all vCJD cases are in
print. Europe. And so on. It sounds obvious. But the
668 Editorial

hypothesis that leads to an absurd conclusion is, of your hypothesis to have relevance to the prac-
well, absurdly common. tice of medicine. Otherwise it is ÔjustÕ theory, an
under-rated part of biology but not really appropri-
ate for this journal.
So what? None of this is prescriptive. There are excel-
lent papers in Medical Hypotheses that read like
Lastly, I believe strongly that hypotheses must be letters from home, and frankly appalling ones
testable, otherwise they are just waffle. Is there, that read like, well, like every other paper. But
even in principle, any way to see if water is toxic? if you could eliminate papers with no structure,
If not, you are wasting my time – you might as well no argument, no evidence and no appropriate ref-
say that water is vootick, (a property I just in- erences, I, and I believe you, would be much
vented, which has no measurable characteristics happier.
at all). So, suggest a test. Maybe you would expect
mice raised on grain alcohol instead of water to William Bains
live longer? Then that is a potential test. If some- Editorial Board Member
one does the experiment and it works, you will be Medical Hypothesis
famous. 37, The Moor, Royston
It would be nice, in a journal called ÔMedical Herts SG8 6ED, UK
HypothesesÕ, for at least some of the implications E-mail address: [email protected]