How To Read A Scientific Paper: Case Study: Buy My Oranges .. ..9
How To Read A Scientific Paper: Case Study: Buy My Oranges .. ..9
What is a scientific paper? ………………………..…….. 2
Scientific papers are peer‐reviewed....……….....… 3
Anatomy of a scientific paper ……………………….... 4
How to read a scientific paper ………………………… 5
Ethics in research and publishing ………………….... 6
Understanding numerical data ……….…………….… 7
What does “statistical significance” mean? ....... 8
Case study: Buy My Oranges ……………..……………..9
Appendix: Numerical analysis .....……………….... 12
Additional resources and credits ………………… 13
What is a scientific paper?
Scientific papers go straight to the use in new ways to address different
source questions.
If someone asks you about a new
movie you haven’t seen yet, what do The authors of scientific papers also
you say? Maybe, “I haven’t seen it, but provide an interpretation of what they
I’ve heard it’s good.” We generally try to think their new information means and
distinguish first-hand from second-hand how it contributes to our understanding
of how the natural world works. By
presenting the data itself as well as the
analysis, other authors can evaluate
these interpretations for themselves.
Because our understanding is always
changing, sometimes the
interpretations of the data can be re-
evaluated in light of new ideas and new
data.
Copyright 2013 American Society of Plant Biologists. www.aspb.org
Sc
cientiific paper
p rs are
e pee
er-rev
view
wed
Peer revieew is a trad
dition in sccholarly e performe
be ed or that the analysis of the
publication. Prior to publication n, an article re
esults be re
evised.
is evaluated by othe er experts, usually
anonymou usly, and these evalu uations are e Most of the articles of the “masss media”
M
used to im
mprove the e paper. Th he reviewers (n
newspaperrs, magazin nes, and blogs)
b are
may recom mmend tha at additional data be noot subject to
t peer revview. Altho ough they
collected and analyzzed or thatt claims no ot an be effecctive at intrroducing scientific
ca
well suppoorted by thhe data be removed. to
opics to a broader
b audience, the ey are not
allways sufficiently cauutious in th
heir
In the stan
ndard
process, sshown
to the righ
ht, the
authors suubmit
their papeer to a
journal edditor,
who evalu uates
whether thhe topic
is a good match
for the jou
urnal. If
so, the paaper is
sent to tw
wo or
more expe erts,
who read it and
provide thheir
frank evalluations, annalysis of the
t results of the stud dy and
including wwhether thhe conclusiions are ometimes overstate the
so t conclussions. For
supported d by the evvidence andd whether exxample, the mass me edia often states
the study is novel, immportant, and
a th
hat high fruuctose cornn syrup cau uses
g. The edittor forward
interesting ds the obbesity, alth
hough there e is little co
onvincing
commentss to the authors, alon ng with a daata to supp aim. Similarly, a
port this cla
decision to accept th he paper, to requestt re
esearch rep port might show that under
revision, o
or to rejectt it. la
aboratory conditions,
c a new dru ug slows
th
he rate of growth
g of cultured
c cancer cells,
The role oof the revie
ewer is to evaluate
e th
he buut the med dia may describe it ass a new
experimen ntal design
n and the data
d cu
ure for can ncer.
presented d, as well as
a the interrpretation of
o
the resultss. Sometimmes, the reeviewer will If you want to
t know more about a news
find that th
he experimmental desiign was no ot re
eport on a scientific breakthroug
b gh, find
rigorous eenough to support
s the
e th
he original, peer-revie
ewed articlle and
interpretations made e by the au uthors, in re
ead what thhe researcchers actuaally
which casse the revieewer mightt diiscovered.
recommend that add ditional expperiments
Anatomy of a scientific paper
TITLE
AUTHOR INFORMATION
ABSTRACT: A summary of the study and findings,
written by the author.
Results are presented as tables,
large datasets, and figures, which Figure legend
can include graphs, videos,
diagrams, and photographs.
Some papers include additional
supporting data as a supplement.
DISCUSSION: An analysis and interpretation of the
data presented that integrates the new information
with prior findings, states the implications of the
work, and sometimes generates new hypotheses
tobe tested.
REFERENCES: The list of the articles cited in
the paper that provide information on the research
topic and the methods used.
How
w to read
r a sciientiffic pa
aper
Although scientific papers
p see em short, off the Results. Once you’ve
y read
d the
they are qquite dense e, and it takes a bit of
o auuthors’ con
nclusions and
a interprretations,
time and eeffort to rea
ad one! He ere are a goo back to the
t Resultss section to
o examine
few tips to
o help you get used to t the format th
he data on which theyy based their
and make e sense of the paper. co
onclusionss.
5 10 15 20 25 30 35 40 45
5 50 55 60 65 70
0 75
Wha
at doe
es “sttatistiical siignific
cance
e” mean?
whhich is the probability
p that the nu
ull
Significan nt, but stattistically significant?
s ? hyypothesis iss true. By convention, biologists
When com mparing sam mples or po opulations, typ
pically rejecct the null hypothesis
h if p<0.05.
the differences in me easured vallues can be e Thhat is, unlesss p<0.05, there is no o differencee
small but nevertheles
n ss meaning gful. To beetween the samples, and a the alte ernative
determine if observed difference es are hyypothesis (tthat the sam mples are different)
d iss
meaningfu ul, scientists use statisstical tests. If rejjected.
a measure ed differenc ce is suppoorted by thee
statistical ttest, it can be said tha
at the W
When we apply the t-tesst on our height
difference between th he sampless is sa
amples, the e calculated d p-value iss 0.5.
“statistica ally signifiicant.” Notte that this Thherefore, we do not re eject the null
does not m mean that the t differen nce is real, hyypothesis, and
a we con nclude that there is
just that th
here is a strrong probability that th he noo statisticallly significan
nt differencce between
samples a are differentt. thee heights of
o people att the west-sside
resstaurant (WWR) and the e east-side e
Taller, or not? resstaurant (EER).
As an example, let’s see if peop ple who livee
on the easst side of toown are talller than tho
ose W
We can also measure the height of o people
on the wesst side of toown. It wouuld be in a school ca afeteria in the
t west side of town
extremely difficult to measure th he height of
o (W
WS). In this s sample, th he mean an nd
every persson in the town, so insstead we willw staandard dev viation of th
he heights measured
m
measure a subset off people and then apply at the school are 145.4 cm (18). We W can
a statistica
al test to he pret the data.
elp us interp coompare thes se values tot those me easured at
thee west-side e restauran nt, which haad a mean
We measu ure the heig
ght of 20 pe eople at fasst- of 163.0 cm (11).
( Now when
w we apply the t-
food restaurants on the
t east or west side of o tesst to these samples, the result iss p<0.05.
town and ccalculate thhe mean an nd standard d Thherefore, we reject the e null hypotthesis and
deviation of
o each sam mple (the standard
s saay that the difference
d in height be
etween
deviation is
i often indicated in pa arenthesess peeople at WR R and at WSW is
following tthe mean). On the west side of staatistically significant. In
I the *
town, the m mean of measured he eights is graaph to the right, the mean
m
163.0 cm (11) and on n the east, 163.3 cm vaalues and standard de eviations
(11). Theyy don’t look
k different, but
b are theyy? forr the three samples are
To find out, we can use
u a statistical test th hat shhown. To indicate that
is common nly used to compare twot sets of me easured va alues in thee WS
related but unlinked data,
d called
da saample differr significanttly, we
Student’ss t-test (whhich can be done in higghlight them m with a staar.
Excel; seee the Appen ndix).
Sttatistically
y significan nt, but sign nificant?
By conven ntion, a t-test calculate
es the Ju
ust because e a result iss statisticallly
probabilityy that the null hypoth hesis is true
e. meeaningful doesn’t
d mak ke it important. Is it
The null hyypothesis states
s that the two ne
ewsworthy to t find that school kids are
samples a are the sam me. The alteernative horter than adults?
sh
hypothes sis is that th
he two sam mples are
different. The outputt of a t-test is a p-valuue,
Cas
se stu
udy: Buy my oran
o ges
You like frresh orangge juice, an
nd you also
o Iff I haven’t convin
nced youu to buy
like to gett some vita
amin C in th
he morningg. m orang
my ges, how can I coonvince
I have devveloped a new varietty of
y
you? Whaat other question
q ns do
oranges, BETTER (B), that I’m m marketing
based on their highe er-than-usuual vitamin
n
y
you have? How would
w youu get
C contentt. y
your ques
stions an
nswered d?
Given tha
at the juice from B tasstes just ass
good, andd the orangges producce the sam me
amount off juice, butt B produce
es more
vitamin C, will you start
s buying
g B insteadd
of A? Whyy or why no ot?
Are they really diffferent? caan begin too figure out which of these
You proba ably are cooncerned th hat the sccenarios iss supportedd.
differencee between the amoun nts of
vitamin C in the two juices is faairly small A histogram m plot showwing the me easured
(B has abbout 8% mo ore than A)) and vitamin C va alues from type A (blue) and
wonder if that differe
ence is reaal or tyype B (orannge) for the
e two differrent
random. M Maybe I jusst got luckyy and sccenarios iss shown be elow. Just looking
l at
picked a B orange with
w a sligh htly higher- thhe data, wee can get a sense tha at the
than-averrage level of
o vitamin C.
C You vaalues for B tend to bee higher thhan the
probably wwant to knoow if the difference I vaalues for A in the botttom plot, but
b we can
observed is repeata able and wh hether a acctually asssign a certa
ainty to thiss
small diffe ess than 10% has an
erence of le ny in
nterpretatioon by analyyzing the numbers.
n
real mean ning. In
n the first dataset,
d the
e mean (wiith the
It is possible to answ
wer both thhese sttandard de eviation shoown in parrentheses)
questionss by measu uring the vitamin C off A and B respectivel
r ly are 18.9
95 (0.67)
content froom severa al oranges of each annd 19.09 (0 he second dataset,
0.67). In th
variety annd analyzinng the meaan and th
he values are
a
standard d deviation of
o both datasets. 188.39 (0.39)) and
199.44 (0.35)). The *
Let’s look at two sce enarios. In one, the Student’s t-ttest
two types of oranges are not really
r coonfirms tha at the
different a
and I just got lucky. We’ll
W call th
hat seecond data aset
the null hyypothesis, meaning thatt there is re
eally is diffe
erent:
no difference. In thee alternativve p<<0.05, whe ereas
hypothesiis, the two types of oranges are e fo
or the first set,
s
different. If we meas sure the vittamin C p==0.5.
content off lots of diffferent oran
nges, we 18 -
Therefore, by repeating the depend on the conditions in which it was
measurement, we can work out whether grown. Perhaps next year the vitamin C
the two samples are different or the level will not be different between A and
same. Because scientists regularly B. It would be nice to see a difference that
measure small but meaningful extends over more than one harvest
differences, replication, numerical season and more than one growing
analyses, and statistical testing are region.
essential tools for scientists.
Science provides us with a set of very
Other considerations powerful tools with which to understand
Which of these statements are supported the world, but our knowledge accrues in
by this study? small steps. Any single study provides an
incremental advancement, but also lays
B oranges are healthier. the groundwork for further studies.
If I eat B oranges I will get fewer colds.
B oranges are a better value. Caveat emptor
B oranges always have more If you wanted to buy a car, you wouldn’t
vitamin C. take the advice of the salesperson on the
car lot, would you? You would first do a
The answer is that none of these side-by-side analysis of the cars you like
statements are supported by the data to compare fuel efficiency, safety ratings,
presented in this study. Based on the and the reputation of the manufacturer.
distribution of values shown in the An educated consumer makes good
second, “alternative hypothesis,” dataset choices by looking at evidence. We are all
our study showed that there is slightly consumers of science, whether it informs
more vitamin C on average in the B our decisions about voting, health care,
oranges, but any claims beyond that are eating, purchases, energy use, or merely
unjustified extrapolations. To investigate how we understand the world. Although
these issues, we would have to design the scientific literature can be a bit
and perform additional experiments. We daunting, it is an important resource.
could also examine the scientific literature Learning to read scientific papers
to investigate the relationship between empowers you to cut through the
vitamin C consumption and the incidence hyperbole and hype, to see for yourself
of colds. Furthermore, the characteristics the evidence that lies behind the claims,
of a fruit are somewhat variable and and to draw your own informed
conclusions from the studies.
Appendix – Numerical analysis
Here are the numbers graphed on the Here are the numbers used on the page
page titled “Understanding numerical titled “What does ‘statistical significance’
data.” mean?”
West
Graph 1 Graph 2 Graph 3 East Rest. West Rest. School
4 0 0 154 182 125
5 0 0 166 166 140
6 0 0
172 159 136
7 1 0
148 155 160
7 1 0
8 1 0 158 143 125
8 2 0 173 177 150
8 2 0 161 156 158
9 2 0 180 172 136
9 2 0 175 168 124
9 4 0 155 165 180
10 6 0
159 145 172
10 6 0
10 7 0 183 156 124
11 9 0 174 154 118
12 10 0 165 157 156
12 12 0 147 172 145
13 16 0 164 167 136
13 20 0 167 154 145
14 20 37
177 177 154
14 22 51
15 33 66 150 184 167
16 54 76 173 158 158
Mean 10 10 10 Mean 165.05 163.3 145.4
StDev 3.23 13.01 23.16 StDev 10.80 11.54 17.63
t‐test 0.633 0.005 (WS
The mean and standard deviation were (ER vs. WR) vs. WR)
calculated using functions in Excel
(2010). To calculate the mean of a group The mean and standard deviation were
of numbers, in an empty cell below them, calculated as described previously. The t-
type “=average(…)”. The space between test was also calculated using a function
the parentheses is filled with the first and in Excel. Click on the function box (fx)
last cell in the column—for example, icon at the top of the page. In the pop-up
(A6:A29). After you type the left window, select the category “statistics”
parenthesis, you can use your mouse to and then T.Test. In the pop-up box, select
draw a box around the column of one set of numbers for Array 1, the
numbers and then type the right second for Array 2. For this kind of data,
parenthesis. To calculate the standard (independent, Gaussian distributed),
deviation, type into an empty cell select “2” for tails and “2” for “type.”
“=stdev(A6:A29).”
Additional Resources and Credits
Additional Resources
American Statistical Association: Education.
(http://www.amstat.org/education/onlineresources.cfm)
Benchpress Project. (2012). Resources to teach journalists about scientific papers and statistics
(http://www.benchpressproject.org.uk/)
Hans Rosling. (2010). The Joy of Stats (Video)
(http://www.gapminder.org/videos/the-joy-of-stats/)
Visionlearning
Scientific communication: Understanding scientific journals and articles
(http://www.visionlearning.com/library/module_viewer.php?c3=&mid=158&l=)
Scientific communication: Peer review
(http://www.visionlearning.com/library/module_viewer.php?mid=159)
Scientific communication: Utilizing the scientific literature
(http://www.visionlearning.com/library/module_viewer.php?mid=173&l=)
Scientific research: The case of the ivory-billed woodpecker
(http://www.visionlearning.com/library/module_viewer.php?mid=174&l=)
Produced by Mary Williams for the American Society of Plant Biologists (www.aspb.org).