MAKING SENSE OF NUMBERS How statistics help us understand

advertisement
number crunching
ISSUE 18 | Summer 2013
bringing CUTTING-EDGE SCIENCE INto THE CLASSROOM
MAKING SENSE OF NUMBERS
How statistics help us understand the world
vital statistics
Statistics can seem daunting,
but don’t panic! This issue
of Big Picture shows how we can
use maths to understand more
about the world around us. Join us
as we explore how to use stats to
summarise data, see whether our
figures are significant and put our
findings into context, so we can
make decisions based on evidence
rather than opinion. Read this
alongside our online content at
www.wellcome.ac.uk/
bigpicture/numbers.
Some thought-provoking numbers from the world around us
NUMERACY skills equivalent to gcse grade G or higher by age
Source: www.gov.uk/government/uploads/system/uploads/attachment_data/file/32277/11-1367-2011-skills-for-lifesurvey-findings.pdf
Vital statistics
Some interesting numbers from
the world around us.
2
How science works
How researchers use science
to explore and understand.
4
Making sense of stats
A look at how we use statistics
to interpret data.
6
Risky business
Exploring how different people
understand risk.
Stats Q&A
Putting to rest some common
statistical myths.
When stats go bad
Real-life examples of how
statistics can be misused.
real voices
Three people talk about how
they use science and stats.
number of drivers
car drivers killed or seriously injured by age IN GREAT BRITAIN
INSIDE
2005–09 – average
2011
8
10
12
14
ONLINE
Go to www.wellcome.ac.uk/bigpicture/
numbers for more teaching resources,
including extra articles, useful web links, lesson
ideas, curriculum links and more. You can
also download the PDF of this magazine and
subscribe to the Big Picture series.
driver age
Source: www.gov.uk/government/statistical-data-sets/ras30-reported-casualties-in-road-accidents
2 | BIG PICTURE 18: number crunching
data storage
250
32
GB
GB
portable
mp3 player
games
console
18 000 000
200 000 000
GB
data amassed by CERN by mid-2012
200 petabytes
100 000 000
GB
usable storage at wellcome trust
sanger institute by 2011
GB
facebook photos and
videos by mid-2012
Source: www.amazon.com, www.itbusinessedge.com/cm/blogs/lawson/the-big-data-software-problem-behind-cerns-higgs-boson-hunt/?cs=50736,
www.extremetech.com/computing/129183-how-big-is-the-cloud, Wellcome Trust Sanger Institute
worldwide email accounts
2012
3.3
billion
earnings premium for those
who took a-level maths
2016
4.3
(estimated)
10%
billion
Source: www.reform.co.uk/client_files/www.
reform.co.uk/files/the_value_of_mathematics.pdf
Source: www.radicati.com/?p=8417
coin tossing probabilities
0.25
Probability of getting two heads
if you toss two coins in a row.
60%
of (97) Members of parliament tested couldn’t correctly
predict the probability of getting two heads in a row.
finding data
Putting this diagram together, we
found that different sources gave
different numbers for the same thing.
Why don’t they match?
Well, data can be interpreted in
different ways, and estimates can be
made using different methods and/or
baseline data. Definitions matter,
too – different sources might define
‘numeracy’ or ‘adult’ differently.
Which should you choose? The source
itself is important – is it reliable?
Are the figures recent? How might
an organisation’s ‘agenda’ affect
how it calculates and presents data?
Source: Royal Statistical Society and Ipsos MORI
Summer 2013 | 3
how science works
Scientists work to investigate, interpret and understand the
world around us. They use a set of tools and techniques known
as the scientific method and produce data.
iStockphoto
Beating bias
Researchers try to keep things objective
Bias is anything that introduces
errors into research and distorts your
findings. Good design means trying
as much as possible to eliminate bias
throughout the experiment – from
the initial research through to the
publication of the results.
Researchers try to reduce bias
in several ways. These include
using blind trials, in which certain
information is kept from people in a
study or even the investigators (e.g.
patients not being told whether they
are receiving an experimental drug
or proven drug). They also use control
groups: the control group is treated
the same as the experimental group,
except in the one variable you are
investigating.
If a population is being sampled,
the sample size needs to be big enough
to reflect the overall population as
precisely as possible. This increases
the study’s reliability (how likely
it is that someone repeating the
experiment would get results similar
to those of the initial investigator),
but it often adds to the cost.
How the sample is chosen is
also important. Choosing the
sample randomly or systematically
helps to eliminate investigator and
other biases. As the name suggests,
systematic sampling uses a system.
You break a population into elements
that are then selected at regular
intervals to form the sample – for
example, from a list of everyone in
year 12, start with a randomly selected
student and then pick every 20th
student from the list.
The null hypothesis
A hypothesis is an explanation you can test
It’s human nature to look for patterns
and draw conclusions from what
we observe – for example, to argue
that there is a link between x and y.
However, science can actually
never prove anything with absolute
certainty. Instead, researchers assume
that no link exists and explore how
likely it is that they would still see
the same result because of chance or
other unknown factors.
In science, a hypothesis is the
explanation you think is behind an
observation. To show a scientific
hypothesis to be true, you actually
need to show that the null hypothesis
– a ‘non-event’ where the effect is not
seen – is false. Statistical analysis can
then be used to assess the support
for the alternative hypothesis. For
example, you might think that
bumblebees prefer one colour flower
over another. In this case, your null
hypothesis (H0) is ‘There is no
difference in the number of visits
to each colour of flower’, and the
alternative hypothesis (H1) is ‘There is
a difference in the number of visits to
each colour of flower’.
MORE ONLINE: www.wellcome.ac.uk/bigpicture/numbers
4 | BIG PICTURE 18: number crunching
sturdy studies
How to recognise good research
When it comes to testing a hypothesis,
high-quality results come from a study
that is well designed and limits bias
(see ‘Beating bias’, left). Often, this
can mean changing only one variable
at time, although this can be hard
in real-world situations. So-called
‘multivariate statistics’ can be used
where several different variables are
being observed at once – for example,
when assessing the effects of fertilisers
on plant growth, where the variables
might include plant height and
crop yield.
Care should be taken when
extrapolating the results of a study.
Extrapolating means stretching
information beyond the specific
group you were studying – for
example, applying the findings of
animal research or in vitro research
(i.e. research in test-tubes) to humans.
When results are published in a
peer-reviewed journal, this means the
articles have been reviewed carefully
by scientists working in the same
field. This lends credibility to the
findings and indicates that they
reflect research that has followed
the scientific process.
FAST FACT
Drowsy driving is dangerous: if you’re awake for 24 hours, the
effect on driving is equivalent to a blood alcohol level of 0.1 per
cent, greater than the UK legal limit. Source: well.blogs.nytimes.
com/2013/01/04/drowsy-drivers-pose-major-risks
look at the evidence
grand designs
Evidence is central to science
Evidence is what science is all
about. It’s all well and good
having an idea about how the
world works, but you need
something to back it up. The
scientific method allows
researchers to test their ideas
through investigations.
First, you need a question.
This normally comes after
some initial observations that
suggest something interesting
is happening or from a problem
you want to solve. You can
construct a hypothesis that
could explain what you’ve seen
and carry out an experiment to
Not all experiments are the same
test it, then analyse the data and
draw conclusions. The overall
quality of any study depends on
making sure each step is done
properly.
Not all evidence gathered
using the scientific method is
the same. Quantitative data
are measurable (e.g. length or
height), and because they are
numerical they can be analysed
statistically. Qualitative
data are descriptive (e.g. hair
colour). Analysing qualitative
data is more difficult and
interpretations can be subject to
personal opinion.
How you design a study depends on the question you’re
asking. In medicine, the most appropriate type of study
depends on whether you are trying to diagnose, treat or
calculate the likely outcome of a condition. For more on
this, see the diagram below.
In ecology, as in medicine, samples are taken: examining
an entire population can be time-consuming and damage
the environment you’re looking at. The design of the
study depends on what you’re investigating. For example,
to estimate the size of an animal population, researchers
often use a mark–release–recapture method. Marking
and releasing a set number of individuals, then capturing
another set number and counting how many individuals
got caught twice gives a good indication of how many
animals there are altogether. To sample plant populations,
quadrats are used so that each sample comes from a specific
area of ground.
types of medical study
This diagram summarises some of the different types of study and analysis used in medical
research. In general, the higher up the pyramid an approach is, the higher the quality of evidence
produced by that approach (and the smaller the amount of evidence available).
Laboratory
work
Case series/case
report
Case-control
study
Cohort
study
Work is done in the
lab, using test-tubes
or animals, to hone
the research method
and question before
moving on to more
advanced studies.
Based on the
treatment of
individual patients
and on observation
rather than
experiment.
Even if the data
are quantitative
(numerical), it is
difficult to make
any generalisations.
An observational
study that compares
individuals who have
a particular condition
(‘cases’) and
those who do not
(‘controls’). You can
see any correlation,
or link, between a
particular factor and
the disease, but you
cannot draw reliable
conclusions about
any cause.
A group of people
are monitored over
an extended period
(often years) to
see how changes
in one variable
affect another – for
example, smoking
and lung cancer.
Randomised
controlled
trial (RCT)
Systematic
review and
meta-analysis
Often used in drug
testing, RCTs involve
the participants
being randomly
assigned to receive
either the treatment
under investigation
or a placebo (dummy
treatment). Known
as the ‘gold standard’
for clinical research.
The strongest
evidence. A
systematic review
collects all the
available literature
on a particular
topic, and metaanalysis is used
to combine the
numerical outcomes
of many separate
RCTs.
Source: Based on a diagram from the UNC Health Sciences Library
Summer 2013 | 5
making sense of stats
Experiments yield data. How can we interpret this information using
statistics? What are some of the common pitfalls in data analysis
and interpretation?
Graphically thinking
Just about average
There are different types of average
When we talk about ‘an average’,
what we’re really trying to do is get
some sense of where the middle
is. We can then use that as a way
of comparing two groups of data.
Unfortunately, there isn’t just one
type of average – there are several.
To get the mean, add the data
together and divide the total by
how many data points there are (see
equation below). Beware, though
– outlying data can often skew the
mean to be artificially high or low.
Take the following number list:
1, 3, 6, 9, 9, 11, 14. The mean here is
7.57. However, add a much higher
number to the end of the list, say
50. The mean is now 12.88. Just one
particularly large outlier has almost
doubled the mean, and the majority
of the numbers are below the mean.
Imagine how the mean wealth of
biology teachers in a room might
change if Bill Gates joined them.
If you place the numbers in
ascending order and look for the
middle value in the list, you have
the median. If there are an even
number of values, you take the
mean of the middle pair. For the
original list, this is 9. Outliers
have a much smaller effect on the
median than the mean, so adding
50 again does not alter the median.
The mode is the value that
occurs most often in a list. For this
list, the mode is 9.
What’s your type?
Not all data are the same
iStockphoto
Researchers define data in different ways. For example, data are categorical if the
values can be sorted into non-overlapping categories (e.g. by blood type, species
or sex). Every value should belong to only one category, and it should be clear
which one it belongs to. Categorical data are also known as ‘nominal data’, or
‘frequencies’, as the research looks to find out how frequently data fall into each
category. Ordinal data, by contrast, can be ranked or have some sort of rating
scale. Ordinal data often come from surveys and questionnaires.
Data can also be defined as discrete
or continuous. Data are discrete, or
discontinuous, if they can take only
isolated values. Continuous data can
take on any value and are limited only by
how accurate your measurements are.
So, while foot length is continuous, shoe
size is discrete because you can’t be a size
7.234434 – you have to be a 7 or a 7.5.
MORE ONLINE: www.wellcome.ac.uk/bigpicture/numbers
6 | BIG PICTURE 18: number crunching
Using graphs and diagrams to show data
When you have your data, you may want to
represent them graphically – for example, to
show whether two variables are correlated
(e.g. a scatter graph plotting duck egg length
against duck egg width) or to show different
proportions (e.g. a pie chart showing the
prey items of a lion). Which chart or graph
you use will depend on the type of data you
have. Common diagrams include bar charts,
pie charts, line graphs, scatter graphs and
histograms.
See www.wellcome.ac.uk/bigpicture/
numbers for a how-to guide on histograms
and weblinks to other resources on graphs
and charts.
A number of significance
Significance has a special meaning in stats
If you want to accept your alternative
hypothesis, you must first reject your null
hypothesis. There is a chance, however, of
rejecting the null hypothesis when it is actually
true. It is usually possible to calculate the
probability (p-value) that what you observed in
an experiment was due just to chance.
You use a significance level to decide
whether you will reject the null hypothesis,
and this is often set at the 0.05 or 5 per cent
level. If your measured p-value equals 0.04, for
example, then this is less than 5 per cent, so you
can reject the null hypothesis and accept your
alternative. Still, this doesn’t mean that you
have proved the alternative hypothesis: if the
null hypothesis were true, there would still be
a 4 per cent chance of getting your result.
If an unscrupulous investigator keeps on
doing experiments on useless treatments, they
will still get results ‘significant at the 5 per cent
level’ on 1 in 20 occasions. If only those ‘positive’
trials are reported, we will get a very misleading
impression. This is why it is essential to have
access to all the evidence, whether positive
or negative. ‘All trials registered, all results
reported’ is a campaign started by researchers,
doctors and others to try and get all clinical
trials past and present to be registered and to
have their results reported.
FAST FACT
In a class of 23 people, the chance
that two people will have the same
birthday is just over 50 per cent.
Source: mathforum.org/dr.math/
faq/faq.birthdayprob.html
iStockphoto
What is normal?
Many things follow a normal distribution
Datasets can be spread
out in many different
ways. The majority of
the data can sit above
the mean or below it. In
many datasets, however
– particularly large
ones – the data points
seem to settle equally on
either side of the mean.
Plotted on a graph, the
shape of the distribution
resembles a bell and so is
sometimes called a ‘bell
curve’. This is also called
a normal distribution
(see graph below).
Standard deviation
is a measure of how
spread out the numbers
are around the mean.
If a dataset follows a
MEAN
SD = STANDARD
DEVIATION
-3SD
-2SD
-1SD
0
68%
95%
99.7%
+1SD
+2SD
+3SD
normal distribution,
approximately 68 per
cent of the data will fall
between one standard
deviation on either side
of the mean. Around 95
per cent will fall within
two standard deviations
on either side. In such
circumstances, the mean,
median and mode of the
data are all equal.
There are many
everyday biological
examples that follow
a roughly normal
distribution, including
blood pressure, height
and foot length. Along
with these examples, you
could also try looking at
stalk height in daisies,
the length of holly
leaves or the diameter of
lichens (commonly found
on gravestones).
Choose your method
Exploring different statistical tests
There are several different statistical
tests that you can use, depending on
the type of data you are dealing with.
Two examples are given here, and
there are more online – plus a workedthrough example for chi-squared (‘chi’
is pronounced ‘ki’ to rhyme with ‘eye’)
– at www.wellcome.ac.uk/bigpicture/
numbers.
The chi-squared test is used with
categorical data to see whether any
difference in frequencies between your
sets of results is due to chance. For
example, you could use the test with
the null hypothesis that ‘there is no
difference in the frequency of worms
on different types of ground’.
In a chi-squared test, you draw a
table of your observed frequencies
and your predicted frequencies and
calculate the chi-squared value. You
compare this to the critical value to
see whether the difference between
them is likely to have occurred by
chance. If your calculated value is
bigger than the critical value, you
reject your null hypothesis.
The t-test enables you to see
whether two samples are different
when you have data that are
continuous and normally distributed.
The test allows you to compare
the means and standard deviations of
the two groups to see whether there
is a statistically significant difference
between them. For example, you could
test the heights of the members of two
different biology classes.
Jumping to
conclusions
Take care with correlation
So you’ve collected your data and
noticed a strong correlation between
two of your variables. It might be very
tempting to assume that a change
in one is causing a change in the
other, but don’t fall into the trap.
A correlation shows that there is a
relationship between your variables;
it doesn’t prove there is a causal
relationship.
Think about Andrew Wakefield’s
1998 claim that using the MMR
vaccine can result in autism. It’s true
that use of the MMR vaccine had
increased up to that point, as had the
number of cases of autism recorded,
so there’s a correlation between the
two. This doesn‘t necessarily mean
that the jab is causing the increase in
autism – there could be a third factor
(called a confounder) causing one
or both of the variables to increase,
such as an increase in maternal age.
Wakefield’s work has been discredited
and he was struck off in 2010,
meaning he can no longer practise
as a doctor.
True causation can only be tested
with carefully controlled studies,
which often compare two groups who
are matched in every way except for
the variable of interest. This limits
the role of confounders as much as
possible.
Oh, and watch your ‘u’s and ‘s’s – a
causal relationship is very different
from a casual one!
Summer 2013 | 7
Risky business
It is impossible to live a life without risk. So, how do we understand
and weigh up the risks associated with different activities, behaviours
and events?
Take a chance Nobody lives a risk-free life
responsibility for reducing
risk, from individuals to
governments. On the roads,
for example, laws about
speed limits, seat belts and
drink-driving are intended
to reduce the risk of accidents
and injuries, and we all have
responsibility for our
own behaviour as a driver
or passenger.
In the home, people install
smoke and intruder alarms to
reduce the risk of fire damage
or burglary. The same is true of
carbon monoxide detectors. In
addition, risk in the workplace
has become a hot topic in
recent years.
Along with conventional
risk reduction devices such as
fire extinguishers and alarms,
attention is increasingly
turning to reducing other,
less obvious risks at work.
This can include supplying
appropriate office furniture
to reduce the risk of back pain
and taking steps to reduce the
risk of stress. On a larger scale,
national and international
government organisations
spend a lot of time and money
on health campaigns to warn
about risks in many areas, from
unprotected sex to smoking and
from drinking while pregnant
to the dangers of drugs.
iStockphoto
Risk is generally understood
as an exposure to the chance
of injury, loss or harm.
Throughout our lives we come
across countless situations
that present such risks; they’re
impossible to avoid. We often
act to limit risks by undertaking
a particular action or by
reducing or stopping a certain
behaviour. Sometimes we do it
without giving it much thought
– by taking an umbrella if the
forecast says it’s likely to rain,
for example.
Everyone has some
Life-changing findings?
We don’t all respond to risk the same way
Living the (micro) life
One way to quantify risk
MORE ONLINE: www.wellcome.ac.uk/bigpicture/numbers
8 | BIG PICTURE 18: number crunching
deal with short-term choices, where the
potential benefits are immediate, better
than long-term ones.
In some situations we have to weigh
up relative risks – for example, with
preventative medicines such as antimalaria drugs. Drugs like doxycycline
and mefloquine can have unpleasant
side-effects, but if you don’t take them
when travelling to malaria hotspots,
you run the risk of contracting the
potentially fatal condition.
Research into our perception of
risk has tried to place a monetary
value on how much the average person
would have to be paid to willingly
accept a one-in-a-million chance of
death – see ‘Living the (micro) life’. You
might think it would be high, but the
findings estimate that it is just $50.
iStockphoto
US scientist Ronald A Howard first
introduced the concept of ‘micromorts’
– a unit of risk measuring a one-ina-million probability of death. This
allows us to compare the risk of day-today events. For example, in the medical
world, going under anaesthesia for a
non-emergency operation exposes you
to an average of ten micromorts. In the
UK, giving birth (all births combined)
is worth 120 micromorts; a Caesarean
section increases this to 170. Skydiving, rock-climbing and hang-gliding
come in at ten micromorts or lower.
Calculations suggest each mission
flown by a member of Bomber
Command in World War II carried
27 000 micromorts – a 2.7 per cent
chance of death.
People perceive and respond to risks
in different ways. How we think and
behave (psychological factors) and how
society works (sociological factors)
play a part in this – we often go with
our gut feeling or are affected by the
behaviour of people around us. It seems
most people underestimate the risk in
activities where they have control and
overestimate the risk of things they can
do little about.
The timescale involved can also be
a factor, particularly when it comes to
health. It can be hard to be motivated
to change short-term behaviour – such
as eating or drinking habits – because
of a risk of health problems in the
distant future. This can often lead
people to put off an action to another
day, saying things like “I’ll stop
smoking when I get older”. When it
comes to long-term benefits, it is also
potentially easier to take a ‘positive’
action, such as joining a gym, than a
‘negative’ one, such as drinking less. On
the whole, it seems as though people
FAST FACT
On average, one in five in vitro fertilisation (IVF) pregnancies
is a multiple pregnancy, compared to one in 80 for women
who conceive naturally. Source: www.hfea.gov.uk/Multiplebirths-after-IVF.html
iStockphoto
Calculate your odds
Our gut reactions to probability aren’t always correct
Are you absolutely
sure?
Why risk should be reported
responsibly
The way numbers are presented
can be misleading, so watch out.
In 1995, UK news outlets reported
advice from the Committee
on Safety of Medicines, which
suggested that a new version of
the contraceptive pill doubled
a woman’s risk of venous
thromboembolism (VTE) – a
condition in which blood clots
form in the veins around the
legs. In the wake of the news, the
British Pregnancy Advisory Service
estimated that the number of
abortions in the UK rose by 13 000,
reversing a previously downward
trend, as a result of a falling trust
in the Pill. The number of births to
teenage mothers also increased.
The actual findings relating to
the Pill and VTE are as follows: for
every 7000 women that took the
previous Pill, one would have VTE.
For every 7000 women that took
the new Pill, two would have VTE.
The claim of ‘a doubling’ of
the risk of VTE was not wrong:
the number of women affected
had doubled from 1 in 7000 to 2
in 7000. This is the relative risk.
But was this the best number to
publish? The difference in terms
of women affected is just 1 in 7000
(the difference between 2 in 7000
and 1 in 7000 women), or 0.014
per cent.
We’re not always good at doing maths
quickly. Often we go by gut feeling, rather
than what the numbers tell us. Take
childbirth, for example; imagine a woman
has given birth to three children, who are
all boys. If she becomes pregnant again then
people might say – because all of her current
children are boys – that there is a strong
chance her next child will also be a boy, or
that she must be ‘due’ a girl next. And yet
biology tells us there is still a 50 per cent (or
1:1) chance of her having a boy because each
conception is an independent
event and is unaffected by
the existence of her previous
children.
The same is true of coin
tosses. Just because a coin
comes up heads ten times
in a row, a head is no
more likely than a tail
on the 11th flip
(provided the coin
is not fixed). The
probability of
having a boy, or
the coin coming
down heads,
are both 1/2, no matter what’s happened
previously. The situation is different if the
events are dependent – if you pull an ace
from a pack of cards without returning it,
the probability of picking another ace goes
down from 4/52 to 3/51.
This is especially true when unlikely
events occur. Take natural disasters, for
example. It is often said that events like
floods, tsunamis and hurricanes are ‘once
every 100 years events’, and people are
surprised when they occur more often.
What it actually means is there is
a 1 per cent (1 in 100) chance of
the event happening in any
given year. Again, however,
these are independent events:
if a ‘once in 100 years’ flood
happened last
year, that
doesn‘t mean
that it can’t
happen again
this year. It is
unlikely – but
unlikely things
happen all
the time.
Summer 2013 | 9
stats q&a
From lottery mythbusting to understanding medical test results, your
niggling number questions are answered using science and statistics.
L
?
Q: My dad refuses to
pick 1, 2, 3, 4, 5 and 6
on the Lotto draw.
Is this sensible?
Q: My teacher says there’s
no such thing as ‘truth’
in science. So why do we
bother?
A: Those six numbers are just
A: Think for a second about
as likely to come up as any
other six numbers. In the UK
Lotto, six numbers are drawn
at random from numbers 1 to
49, giving 13 983 816 possible
combinations. The chance of
winning is approximately 1 in
14 million, no matter which
numbers you pick. However,
there is a very good reason why
your dad shouldn’t play 1, 2, 3,
4, 5 and 6 – if he did win, he
would win less money than with
another combination.
An estimated 10 000 players
a week choose the combination
of balls 1–6. If you picked those
numbers too and they came up,
you’d have to share your cash
with all those other people,
which would severely dilute
your winnings.
Although you can’t increase
your chances of winning,
there are other ways you might
increase the potential amount
you’d win. People pick numbers
between 1 and 31 more often,
for example, because they
correspond to the birthdays of
people they know. Selecting six
numbers between 32 and 49 is
more likely to set you apart from
the crowd.
how you’d prove something
to be true. You could say,
for example, that all swans
are white – but you can’t
prove that’s true because it’s
impossible for you to observe
every single swan. However,
just one observation of a black
swan would prove that you were
wrong. It’s much easier to prove
something is untrue than to
prove it’s true.
Accordingly, the scientific
method works by trying to
falsify a statement. When
a scientist puts forward a
theory, the rest of the scientific
community try to disprove it.
It’s only those theories that
survive the onslaught – the ones
that can’t (yet) be proven wrong
– that we stick with. That
doesn’t mean that one day a new
piece of evidence won’t come
along to disprove it, but you
trust it for now. For example,
we think that dropping a brick
will mean it falls to the ground,
but we can never prove that it
wouldn’t one day float off. So
far, though, the theory that
it will fall has withstood all
attempts at falsification.
MORE ONLINE: www.wellcome.ac.uk/bigpicture/numbers
10 | BIG PICTURE 18: number crunching
Q: I just read online about
a man who cured his
cancer with carrot juice.
Why can’t we all use
natural remedies?
A: Remember a key saying
in statistics: correlation does
not equal causation. In other
words, just because there
is a relationship between
the volume of carrot juice
the man consumed and the
disappearance of his cancer, it
doesn’t mean that one caused
the other.
Even if it did, think what
you’re not hearing about. If
he had died in spite of all the
carrots, would an article have
been written about it? Maybe
there are thousands of failures
out there that are not being
reported. This is an example
of ‘selection bias’ – where you
only hear about the ‘successes’.
A related phenomenon,
‘publication bias’, happens in
scientific research. Journals and
some researchers are more likely
to publish new, exciting results
than ‘negative’ findings that
a particular variable does not
cause an effect.
Medical science does not
work on the basis of anecdotal
evidence (based on personal
experiences), but through
carefully controlled trials.
Q: I’ve passed my driving
test. I’ve been told by car
insurance companies that
there’s no discount for
being female, even though
women are statistically less
likely to have an accident.
Why?
A: You’re right that women are
statistically less likely to have
an accident. In fact, young men
under the age of 22 used to pay
an average of £1000 more a year
in insurance premiums than
women of the same age. This
was because the statistics show
that young men are twice as
likely to suffer a serious collision
as young women (and ten
times more likely to be killed
or seriously injured in one than
those aged 35 or over).
However, an EU ‘gender
directive’ that came into force
on 21 December 2012 made it
illegal to discriminate according
to sex when pricing financial
products, such as pension
annuities and insurance. The
change has seen premiums for
women under 40 rise and those
for men of a similar age fall.
Some have argued that this
is unfair, but others have said
equality can’t be selective. The
change is unlikely to affect the
premiums of those aged over 40
– the age at which the statistics
show that men and women
become equally likely to have
an accident.
FAST FACT
More than 99 per cent of the population has a greater
than average number of legs. Source: plus.maths.org/
content/all-about-averages
Q: My mum doesn’t
like that I smoke, but I
don’t think it’s a big deal.
My great-grandma has
smoked since she was 15
and is still going at 90!
A: A common misconception
is that your own experience
– or those of your friends or
relatives – can be generalised
to everyone else. As a famous
quote by English statistician
and biologist R A Fisher says,
“That is an experience, not an
experiment”. Just because your
great-grandma has survived
to that age doesn’t mean you
will; you can’t get meaningful
data from a sample size of one.
This is why most scientists use
experiments that collect large
volumes of data, rather than
relying on anecdotal
case studies.
Q: My doctor has told me that if I get a ‘positive’ result in a medical
test, it doesn’t definitely mean that I have the disease. Why not?
A: No test is 100 per cent accurate.
Not all positive test results mean that
someone has a particular condition; not
all negative results mean that a disease
is absent. Positive and negative test
results can be described as true or false,
depending on whether they classify
correctly the person tested.
The diagram shows the outcome of
breast screening by an X-ray technique
called mammography. For every 1000
positive
test
result
41
1000
women
screened
women screened, 41 have a positive
result and are called back for more tests.
Of these, 8 women will be found to have
cancer, and 33 will be found not to. So,
an initial positive result means just a
20 per cent chance of cancer. Overall,
966.8 (8 + 958.8) women per 1000
screened will get a ‘true’ result, making
this technique 97 per cent accurate.
8
true
positive
33
false
positive
further
testing
958.8
negative
test
result
Lottery and smoking
images: iStockphoto
Carrot juice image: 123RF
(cancer)
(no cancer)
true
negative
(no cancer)
959
0.2
false
negative
(cancer)
Summer 2013 | 11
when stats go bad
Actually, this is more a case of ‘when people using stats go bad’. Look at
these real-life examples to see how numbers can be misused, misrepresented
or poorly explained, and explore the implications of substandard stats.
Take care with your... wording
“Just 100 cod left in North Sea”
In September 2012 the
Daily Telegraph website
ran a story with a rather
astonishing
headline:
“Just 100 cod left in North
Sea”. Actually, there are
iStock
around 437 million cod in
that body of water. What
happened? The answer is
a lesson in being precise
with your vocabulary.
The original story
came from the Sunday
Times, who had reported
that an analysis of catches
at North Sea ports across
Europe in 2011 found
no cod over the age
of 13. Using statistical
sampling
techniques,
it was calculated that
this must mean fewer
than 100 such fish exist
(otherwise one would
have been caught). The
finding was reported
under the headline of
“Only 100 adult cod in
North Sea” – but what
constitutes a mature
cod? You’d assume from
the story that it was any
fish over the age of 13;
however, cod reach full
sexual maturity around
the age of six. According
to the Government, there
are more than 21 million
such fish in the North
Sea. It all comes down
what you classify as an
‘adult’, or ‘mature’, fish.
Consider…
Q:Can you find examples
of other headlines that
are misleading because of
the words used (or those
left out)?
Take care with your... calculations
iStockphoto
“Chance of cot deaths in brothers ‘1 in 73 million’”
Sally Clark served three years of a life
sentence for the murder of her two children
before her conviction was overturned in
2003. In the original case, the defence had
claimed that sudden infant death syndrome
(SIDS) – commonly known as cot death – was
responsible for the death of both boys, who
died just over a year apart. The prosecution
argued that such a double cot death was
exceptionally unlikely and claimed murder.
The prosecution’s assertion was based
on the expert testimony of Professor Sir
Roy Meadow, a researcher in paediatrics.
Meadow had said that the chances of one
child dying from SIDS in a non-smoking,
affluent family was 1 in 8543. When working
out the probability of two cot deaths in the
same family, he squared this probability
– multiplying 8543 by 8543 – to get 1 in
73 million.
MORE ONLINE: www.wellcome.ac.uk/bigpicture/numbers
12 | BIG PICTURE 18: number crunching
This would have been the right thing to
do if the two events were independent of
each other, like tosses of a coin. The chance
of getting two heads in a row is 1/2 x 1/2 (1/4).
However, two cot deaths in the same family
are not independent events; there could be
underlying genetic or environmental factors
that make them more likely. The Royal
Statistical Society deemed Meadow’s account
a “mis-use of statistics”.
Consider…
Q:Even if the chance of a double cot death
in the same family really was 1 in 73 million,
why would this not have meant there was
only a 1 in 73 million (0.0000014 per cent)
chance of the accused being innocent?
Search for “prosecutor’s fallacy” online to find
out more.
Q:What figure should this number have
been compared with to work out the relative
likelihood of guilt or innocence?
Q:Should statistical evidence in court only
be presented by experts in statistics, rather
than by experts in the field in which the
statistics are being used?
FAST FACT
Although Grenada won only one medal at the London 2012 Olympic
Games, it comes top in the medal table per capita (per person living
there). The USA, which won the most medals, is ranked 49th when
you count medals this way. Source: www.medalspercapita.com
Take care with your... claimS
“Lifescan,
like an
MOT for
your body”
Lifescan is a company offering
you a CT scan to provide an
“MOT for your body”. Their
original TV advert described it
as “a quick and easy scan that
could detect the early signs of
life-threatening diseases, way
Wellcome Images
before the symptoms begin”. It is
also described as a “check-up all
in one go”. In 2010, however, the
Advertising Standards Authority
(ASA) – the UK’s independent
regulator of advertising across
all media – ruled that the advert
could no longer be broadcast
in its original form after
complaints from two medical
doctors. According to the ASA,
the advert implied that CT
scanning in patients with no
current symptoms could pick up
any kind of underlying health
problem, and they ruled there
was no evidence to back up these
claims.
The ASA also ruled that
Lifescan didn’t provide enough
information about the potential
risks of exposure to radiation.
According to a 2007 report
from a governmental advisory
committee, a typical CT scan
carries a one-in-2000 lifetime
risk of developing a fatal cancer.
Although this risk might be
acceptable in high-risk patients,
it might be unacceptable for
people without symptoms.
In explaining their verdict, they
said: “We were concerned that
the...respondents entering the
competition were selected on
the merits of their competition
entry [the short story] and may
have been inclined to be less
than impartial in their survey
responses in order to stand a
better chance of winning.”
Why risk should be
reported responsibly
Consider…
Q: How much information can a
company reasonably be expected
to include in a short advert?
Q: To what extent is the
consumer responsible for doing
their own research before having
a scan?
Take care with your... sample
“Recommended by
93% of Red readers”
iStockphoto
The Advertising Standards
Authority (ASA) banned a TV
advert for hair product Nice
‘n’ Easy in 2009. The advert
included a voiceover that said
“93% of Red magazine readers
would recommend Nice ‘n’ Easy
to a friend. The other 7% probably
don’t
have
any friends”.
Afterwards,
some
text
flashed up on
the screen:
“Participants
in a survey
of 245 Red
magazine
readers,
April 2008”.
The problem wasn’t so much
the small sample (a professional
opinion survey typically covers at
least 1000 people) as the way the
data were collected. Participants
had volunteered to take part in
the survey and were sent a pack of
the hair dye, as well as the survey
(which included a question about
whether they would recommend
the product to a friend). If they
returned the survey, along with
a photo of themselves and a
short story, they had a chance
of winning a trip to New York.
The ASA banned the advert
because they believed the claim
of 93 per cent was misleading.
Consider…
Q: How many of a magazine’s
readers should you ask before
you can make claims about the
opinions of its readership?
Q: Is it OK to provide incentives
to people taking part in surveys?
Will you get the same results,
or will the incentive skew the
findings?
Q: Is there any guarantee that
the people filling in the survey
actually tried the hair dye they
were sent? How could it have
been done differently?
discuss
Is it ever morally acceptable
to ‘spin’ information about
risk to try to influence people’s behaviour? Imagine a
new report has been published, linking eating meat
to the development of cancer. Is it right for a newspaper to report the relative risk
without the absolute risk
(see page 9), to create a more
striking story and therefore
sell more papers? Would
you feel differently if this
were done by a government
health campaign or a cancer
charity trying to raise funds
for research? What if it were
used in adverts by a business
that produces vegetarian
food?
Summer 2013 | 13
real voices
Three people tell us about the role of statistics and science in their lives. Meet Vicky Peterkin, a
biostatistician at a pharmaceutical company; biology teacher David Colthurst; and Anthony
Underwood, who uses bioinformatics at Public Health England.
Vicky Peterkin
dr David Colthurst
Senior biostatistician at a
pharmaceutical company
What do you do?
I work as a statistician in the
pharmaceutical industry,
specialising in clinical trials
for new treatments.
How do you use statistics
in your job?
To find out whether the drug
we’re testing significantly
improves a measurement
of interest compared with a
control drug. For example,
if we’re testing a new drug
to treat high blood pressure,
the measurement of interest
might be the change in
blood pressure since starting
treatment; however, it’s not
enough to just look at the mean
blood pressure change in each
treatment group. We need to
adjust for other information
(such as age and weight) and
show whether the difference
is statistically significant.
Why is that?
Adjusting for other
information ensures that
any differences between the
treatment group results are
due to the drug, not due to
differences in disease severity
and demographics. Statistical
significance indicates that the
size of the difference between
treatment groups is too big to
have occurred by chance and
must be due to the treatments.
What is a p-value?
When we run a statistical
test, we get a p-value at the
end. It shows the statistical
significance – the degree to
Biology teacher leading a project to do
scientific research in schools
which we can be certain that
the effects we see are due to our
drug, rather than chance. It’s
the key piece of information
we want from the trial. It tells
us whether the drug is useful
and whether we should carry
on researching it. That’s what
I love about statistics: you
can boil a huge amount of
information down to a single,
clear, yes-or-no decision.
Why might people be wary
of maths?
I think people imagine
statisticians sit at a computer
working on their own all day.
In my job that’s just not the
case. I’m involved at every stage
of a clinical trial – I help design
it, check the data quality while
it’s running and analyse it all
at the end – and I work closely
with medics, scientists and
trial monitors on a daily basis.
What training do you have?
I studied maths at university,
then a Master’s in statistics.
There’s a lot of on-the-job
training, too.
I think students need to
be exposed to statistics at
an earlier age, so they can
see for themselves how useful
it is. The industry is in
desperate need of young
statisticians. You can become
involved in important trials
very quickly, and your opinion
is highly valued.
Find out more about working
as a statistician at
www.psiweb.org/newcareers.
MORE ONLINE: www.wellcome.ac.uk/bigpicture/numbers
14 | BIG PICTURE 18: number crunching
What do you do?
I am a secondary school
science teacher. I am also
the lead teacher on the MBP2
project, which gives sixth-form
students the opportunity to
carry out genuine research.
What is the MBP2 project?
Five years ago my wife was
diagnosed with multiple
sclerosis. I wondered if I would
be able to combine my 15 years
of experience as a biochemist
with my 15 years of teaching.
The Myelin Basic Protein
Project (MBP2) is investigating
the role of this protein in
multiple sclerosis using
genetically modified yeast.
The project is run in
collaboration with researchers
at the University of Kent.
When we first started,
they ran a workshop at the
University to teach DNA and
protein techniques to a small
group of students, who then
taught them to their teachers
and other students. It was a
nice turnaround to have the
students be the experts while
the teachers sat scratching
their heads, thinking “How
does this bit work?”
What do your students gain?
The real wake-up call for them
was around experimental
procedure – often, experiments
don’t work as planned or give
the results that are expected.
When this happens, you
have to tweak and change
the design. Very quickly the
students realised that while
you may do steps A, B, C and
D in the hope you will get
result E, nine times out of ten
it won’t work! It has also given
the students the opportunity
to try lots of techniques that
they normally wouldn’t get to
experience until university.
For example, our students
have developed a genetically
modified yeast strain that can
make the human myelin basic
protein. While doing this they
have learnt how to extract
DNA, use the polymerase chain
reaction (PCR) to amplify DNA,
and carry out Western blots to
study the proteins produced.
What is next for the project?
I believe MBP2 provides a model
for how students can carry out
research in schools. Inspired
by this, Authentic Biology is a
series of research projects led
by sixth-form students in five
schools across the UK. Each
one is drawing on the expertise
of researchers at their local
universities to investigate
topics relevant to them. In
London, for example, students
are researching diabetes, and
in Sheffield they are looking at
heart disease. We are currently
planning the second Authentic
Biology Symposium. This is
an opportunity for the schools
involved to share their research
with each other and with
academics from the partner
universities.
Find out more at www.
thelangtonstarcentre.org/
index.php/mbp-squared-link.
Big Picture Debates the Brain
Is being in love just a chemical reaction? Is technology
harming our brains? These are just some of the areas
of debate presented in our free app, which explores social
and ethical questions about the human brain. Find it at
brainapp.wellcomeapps.com or on the App Store.
dr ANTHONY UNDERWOOD
Bioinformatician at Public
Health England
What do you do?
I’m a bioinformatician, which
means I use computers as a laboursaving device to answer biological
questions more quickly than would
be possible using the traditional
biological methods. I work for Public
Health England (PHE), an executive
agency of the Department of Health
with a broad remit to protect the
community against health dangers,
including infectious disease,
chemicals and radiation.
How do you use statistics
in your job?
I work in the bioinformatics unit of
the Microbiology Services Division
of PHE. We are currently involved
in a project that aims to sequence
many of the genomes from bacteria
or viruses in patient samples sent
to us by hospitals and GP surgeries.
Obviously, that generates a vast
amount of data very quickly, and
it’s my job to crunch it so we can get
meaningful information from it.
What kind of information?
My lab colleagues need to identify
the species and characterise the
particular strain of the bacteria or
virus. They look for sets of genes that
might make it resistant to particular
drugs or produce variants of a toxin
that make it more harmful when
infecting people.
The revolution in sequencing
technology means we can now
identify differences between
bacteria at the level of a single DNA
nucleotide. That means we can do
some really neat stuff in terms of
tracking the infection to its source
– whether that’s a meat-packaging
factory, within a particular
community (a school, for example), or
a healthcare worker in a hospital.
Why do you think people can be
wary of maths?
I know my own kids think “What’s
the use of that formula I’ve just
learned?” I think most people like
to see how they can use numbers to
achieve something concrete.
Did you like maths at school?
I always enjoyed maths, but biology
was something I could reach out and
touch. I could look at a pond sample
under a microscope or tear apart a
leaf and see what was going on inside
it. It was only when I was working as
a molecular biologist that I saw the
real-world uses of maths.
How did you get into bioinformatics?
I was looking for amino acid patterns
in nematode worms, which was a
long, laborious process. Someone
mentioned that bioinformatics
might be able to help me out. When I
finally managed to get the computer
program I’d written to work, I got the
data in minutes instead of days. That
was a huge kick. I still get a thrill
now, when I write a program that can
do weeks of work instantaneously.
The big leaps forward in science
today come from combining
different skill sets, which is what
bioinformatics does. It’s really taking
off; our group has expanded from a
team of three to ten in a year, so it’s a
career for the future.
Find out more at www.gov.uk/
government/organisations/publichealth-england.
the team
Education editor: Stephanie Sinclair
Editor: Chrissie Giles
Assistant editor: Kirsty Strawbridge
Writers: Penny Bailey, Chrissie Giles, Emma
Rhule, Colin Stuart
Project manager: Rosie Cotter
Graphic designer: Malcolm Chivers
Illustrator: Glen McBeth
Publisher: Mark Henderson
Head of Education and Learning: Hilary
Leevers
Teachers’ advisory board: Peter Anderson,
Paul Connell, Alison Davies, Helen English,
Ian Graham, Stephen Ham, Kim Hatfield,
Jaswinder Kaur, Moss Newnham, Jonathan
Schofield, Robert Rowney
Advisory board: Sarah Allen, Graham
Currell, Fiona Davidge, Neville Davies,
Thomas Ezard, Marianne Freiberger,
Alexis Gilbert, Jenny Koenig, Nancy Lee,
Ross MacFarlane, Giles Newton, David
Spiegelhalter, Andrew Steele, Robin Sutton
Wellcome Trust: We are a global charitable
foundation dedicated to achieving
extraordinary improvements in human and
animal health. We support the brightest
minds in biomedical research and the
medical humanities. Our breadth of support
includes public engagement, education
and the application of research to improve
health. We are independent of both political
and commercial interests.
The future of science depends on the quality of
science education today.
All images, unless otherwise indicated, are
from Wellcome Images
(images.wellcome.ac.uk).
Big Picture is © the Wellcome Trust 2013
and is licensed under Creative Commons
Attribution 2.0 UK. ISSN 1745-7777.
Cartoon illustrations are © Glen McBeth.
This is an open access publication and, with
the exception of images and illustrations,
the content may unless otherwise stated
be reproduced free of charge in any format
or medium, subject to the following
conditions: content must be reproduced
accurately; content must not be used in a
misleading context; the Wellcome Trust
must be attributed as the original author and
the title of the document must be specified
in the attribution. The Wellcome Trust is a charity registered
in England and Wales, no. 210183. Its sole
trustee is The Wellcome Trust Limited, a
company registered in England and Wales,
no. 2711000 (whose registered office is at 215
Euston Road, London NW1 2BE, UK).
PU-5687/23K/05–2013/MC
Summer 2013 | 15
Veer
Want to bring
cutting-edge
science to the
classroom?
We’re looking for trainee and
newly qualified post-16 biology
teachers to join the Big Picture
advisory board.
You’ll be involved in reviewing content,
shaping the future of the magazine,
and much more. It’s a brilliant opportunity
to influence the Wellcome Trust’s education
work and could boost your career prospects.
For more information, contact the team
at bigpicture@wellcome.ac.uk
BigPicture free subscriptions
Sign up to receive free regular
copies of Big Picture at
www.wellcome.ac.uk/
bigpicture/order
Here, you can also order more
copies of this issue of
Big Picture, or (not all available)
past issues, which include
Inside the Brain.
Or you can contact us:
T +44 (0)20 7611 8651
E publishing@wellcome.ac.uk
Big Picture
Wellcome Trust
Freepost RSHU-ZJKL-LCZK
Feltham TW14 0RN
Are you a teacher in the UK? If so, you can order a class set. Email publishing@wellcome.ac.uk
Name:
Job title:
Organisation:
Address:
Email address:
Feedback
Questions, comments, ideas? Share your thoughts on Big Picture
by emailing us: bigpicture@wellcome.ac.uk
Big Picture is a free post-16 resource that explores issues around biology and medicine.
50%
This document was printed on material
made from 25 per cent post-consumer
waste & 25 per cent pre-consumer waste.
Download