Writing

Writing

The rest of the paper…

Methods

• Emulate

• Err on the side of too much detail

• Why not start today?

Results

Results are different from data!

Data = the numbers themselves = most belong in figures and tables

Results = the meaning (or analysis) of the data = the text of the Results section

• Results pertinent to the main question(s) asked

• Summarize the data; report trends and statistics

• Cite figures or tables that present data

Results

ResultsOverall, the mortality rate of organisms experiencing experimental stressors was approximately six times higher than for organisms under control conditions (Fig. 1). This effect is significant (i.e. the 95% CI does not overlap zero; Fig. 1). There was significant heterogeneity in overall mortality among experiments (QT = 951.57, df = 335, P < 0.0001).

The total suicide rate for Australian men and women did not change between 1991 and 2000 because marked decreases in older men and women were offset by increases in younger adults, especially younger men (Fig. 3).

Results: tips

• Use subheadings – relate to questions

• Include negative and control results

• Show magnitude of response/ effect (e.g., as percentage)

• Reserve the term “significant” for statistically significant

• Do not discuss rationale for statistical analyses (?)

Results: tense

Use past tense, except to talk about how data are presented in the paper

e.g.:Women were more likely to…Frog numbers declined in the South…

but:Figure 1 shows…Table 1 displays…The data suggest

Remember to use the active voice as much as possible

Information was available for 7766 current cigarette smokers. Of these, 1216 (16%) were classified as hardcore smokers. Table 1 gives characteristics of all the smokers. The most striking difference was that hardcore smokers were about 10 years older on average and tended to be more dependent on tobacco.

Significantly more hardcore smokers had manual occupations,

lived in rented accommodation, and had completed their full time education by the age of 16 years. There was no difference by sex.

ResultsJarvis et al. 2003. Prevalence of hardcore smoking in England, and associated attitudes and beliefs: cross sectional study BMJ  326:1061 

Results: reporting stats

Goal: Comparison of numbers of birds in natural and exotic forests

Stats: t = 2.51, df = 39, P = 0.016

Bad:

The observed t value (2.51) was greater than the critical t value (2.02) for 39 degrees of freedom, indicating that we can reject the null hypothesis of no difference between the treatments with a confidence level of 95%.

Results: reporting stats

Acceptable (only just)

There was a significant difference in the number of birds between natural and exotic forests (t = 2.51, df = 39, p = 0.016).

Best

There were significantly more birds in natural than in exotic forests (t = 2.51, df = 39, p = 0.016).

Figures and tables

General rules• One table or figure per page – at end of manuscript

• Table or figure + its caption make a stand-alone story

• Data presented in one format only

• Do not present raw data (?)

• Number consecutively in the order in which they appear in the text

• Separate numbering for figures and tables

TablesCaption

• Goes at top of table

• Identifies briefly the specific topic or point of the table

• Uses the same key terms as in the column headings and the text of the paper

Table 2. Effects of QTL genotype on placental and embryonic gene expression and weight, birth weight and litter size. mRNA and protein levels are expressed relative to reference samples run in each assay. Values are least squares means ± standard error.

TablesCaption

• Goes at top of table

• Identifies briefly the specific topic or point of the table

• Uses the same key terms as in the column headings and the text of the paper

Format

• Follow journal guidelines

• Only a few horizontal lines

• May use short horizontal line to group subheadings

•No vertical lines!

Tables: example

Stoving. 1999 J Clin Endocrinol Metab 84: 2056-2063

Caption at top

Same terms in headings and caption

3 lines only

FiguresCaption

• Goes at foot of figure

• Identifies briefly experimental details

• Gives definition of symbols, shading, and statsFigure 1. Survival of Galleria mellonella larvae infected with strains of Aspergillus fumigatus. (A) Mean number of larvae alive at each day after injection with each clinical strain. (B) Mean number of larvae alive at each day after injection with each environmental strain. Means are of six replicates for all of the strains, except UAMH 3762, for which 4 replicates were performed. Mating type is indicated by line type: solid lines for MAT1-1 strains and dashed lines for MAT1-2 strains. Strain names are included next to each line. (C) Means of all strains within a group. Circles indicate MAT1-1 strains, triangles indicate MAT1-2 strains, filled symbols indicate strains of clinical origin, open symbols indicate strains of environmental origin, and error bars indicate standard errors. Because of the large number of strains, data are not presented as Kaplan–Meier curves to improve clarity.

FiguresCaption

• Goes at foot of figure

• Identifies briefly experimental details

• Gives definition of symbols, shading, and stats

Format

• Follow journal guidelines

• Variable

• Primary evidence, e.g. electron micrographs, gels, etc.

• Graphs, e.g. scatter, bar, boxplots, etc.

• Diagrams or drawings, e.g. model, experimental set-up, etc.

Figures: Primary evidence

Figure 1. Transcription of antisense RNA leading to gene silencing and methylation as a novel cause of human genetic disease

Cristina Tufarelli et al.Nature Genetics (2003)

Figures: Primary evidence

Zucca et al. 1998. NEJM 338: 804

Figure 1. Histologic patterns in the evolution from chronic gastritis to gastric lymphoma.

Figures: Graphs

Figure 3. Hypertension Prevalences in 6 European and 2 North American Countries, Men and Women Combined, by Age Group

JAMA Vol. 289 No. 18, May 14, 2003

Figure 3. Hypertension prevalences in six European and two North American countries for men and women combined, by age group.

Line graphs

• Used to show changes (e.g. over time)

Figure 1. The relationship between the percentage of body fat and the serum leptin concentration in 136 normal-weight and 139 obese subjects.

Considine et al. 1996. NEJM 334: 292

Figures: Graphs

Scatter plots

• Used to show relationships between two variables

Figures: graphs

Figure 2- Relationship between BMC of the forearm/heel and time since menarche. *Significantly different than forearm BMC of group 1 (< 1 yr since menarche); BMCA: forearm BMC; BMCH: heel BMC.

Medicine & Science in Sports & Exercise 2003; 35(5):720-729

Bar graphs

• Used to compare groups

Figures: Diagrams

Talan et al. 1999. NEJM 340: 85

Figure 1. Location of wound infections in 50 patients bitten by dogs and 57 patients bitten by cats.

Figures: Diagrams

Figure 2: Proposed pathways among disordered eating, menstrual irregularity, and low BMD. Solid lines represent associations suggested by the current study; dashed lines represent associations suggested by previous studies.

Figures: Diagrams

Fig. 1. Location of the study beaches around the island of Barbados.

Fish et al. 2008Ocean and Coastal Management

What about pie diagrams?

Pie charts are for children and politicians!

Statistics and power analysis

Traditional statistics

• Null hypothesis (H0): – no difference/ relationship

• Alternative hypothesis: – there IS a difference/ relationship

• Can the null hypothesis be rejected?

Bayesian statistics

Statistics and power analysis

Traditional statisticsReality:

Null hypothesis is true Null hypothesis is false

What you do:

Accept null hypothesis

Reject null hypothesisError!

α=0.05P-value (sort of)

Error!β=???

1-β = statistical powerIf the null hypothesis is false,

what is the probability that you will reject it?

Statistics and power analysis

Traditional statisticsReality:

Null hypothesis is true Null hypothesis is false

What you do:

Accept null hypothesis

Reject null hypothesisError!=0.05

P-value (sort of)

Error!β=???

1-β = statistical powerIf there really is an effect,

what is the probability that you will detect it?

What affects power?

• Sample size

• Variability

• Effect size-level

Sampling distributions and power

Mean of sample

Frequency

= 0.05

Null hypothesis: μ = 0

Sampling distributions and power

Frequency

= 0.05

Null hypothesis: μ = 0

Mean of sample

Frequency1 – β = power

Reality: μ = 2

Effect of sample size

Frequency

= 0.05

Null hypothesis: μ = 0

Mean of sample

Frequency

Reality: μ = 2

Effect of variability

Frequency

= 0.05

Null hypothesis: μ = 0

Mean of sample

Frequency

Reality: μ = 2

Original example

Frequency

= 0.05

Null hypothesis: μ = 0

Mean of sample

Frequency

Reality: μ = 2

Effect size

Frequency

= 0.05

Null hypothesis: μ = 0

Mean of sample

Frequency

Reality: μ = 4

-level

Frequency

= 0.10

Null hypothesis: μ = 0

Mean of sample

Frequency

Reality: μ = 4

When should you do power analysis?

• Before your experiment

• After your experiment

Power analysis after your experiment

• E.g., with sample size X, you observed effect size Y, and variability Z

• Silly power analysis:– What is the power to detect a difference of Y

with sample size X and variability Z?

• Useful power analysis:– What is the power to detect a biologically

important effect with sample size X and variability Z?

How to present “negative” results

• Power analysis

• Confidence intervals– The confidence interval for the difference

between group A and group B is …– The confidence interval for the strength of the

relationship between A and B is …

Discussion

• Not a detective story

Introduction structureBIG picture

General background, i.e., what is known

The gap: what is not known

Your question/ goalsYour

approach

Discussion structure

Your results

Address your question

Put results in context:How does it fill a gap?

BIG picture

Things you can do(to improve your writing)

• Read, pay attention, and imitate.• Let go of “academic” writing habits • Talk about your research before trying to write about it.• Search for the right word rather than settling for any old

word.• Try not to bore your audience.• Stop waiting for “inspiration.”• Accept that writing is hard for everyone.• Revise. • Learn how to cut ruthlessly. Never become attached.• Find a good editor!