Results
Text, Tables & Figures
Dr. Arun Mitra
Three Day workshop on Manuscript Preparation & Research Communication
3 - 5 June, 2026
Contents
- The Methods–Results mirror
- Where data live: text vs table vs figure
- Reporting a result fully
- Tables & figures by design
- Activities · Checklist
“Methods was the recipe.
Results is the dish.”
Yesterday’s recipe, today’s dish
- M — what you did ← Day 1
- R — what you found ← this session
Tip
Results is where we either over-cook (data dump) or under-plate (interpretation).
Results must mirror Methods
Methods
- Study design
- Variables & measurements
- Primary analysis
- Subgroups / sensitivity
Results
- → Participant flow
- → Table 1
- → Primary outcome
- → Subgroups / sensitivity
Lay the two sections side by side — they should reflect. Anything in Results not set up in Methods is a red flag.
Steps
- Structure Results to mirror Methods
- Decide whether as text vs table vs figure
- Report stats fully: effect + CI + exact P
- Match figure type to study design
- Step 3 is the commonest reviewer complaint on Indian submissions
- Step 4 separates a generic paper from an intentionally designed one.
Structuring Results
The first Results paragraph is always participant flow
- Paragraph 1 = who is in the analysis
- Screened → Eligible → Enrolled → Analysed
- Account for every participant
- Losses & exclusions, with reasons
State Every Finding Only Once
TEXT
headlines, ≤ 3 numbers
TABLE
many exact numbers
FIGURE
patterns, trends, time
Then cross-reference — never repeat. “Anaemia was the commonest morbidity (Table 2).”
The same numbers in a sentence, a table, and a chart inflates length and signals inexperience.
Decision Time?
- Read precisely → table
- See instantly → figure
- Headline → text
Where to place??
| Data scenario | Where |
|---|---|
| One key number (a single prevalence / effect) | Text |
| Baseline characteristics, many variables × groups | Table |
| Participant flow (screened → analysed) | Figure |
| Spread of a continuous variable | Figure |
| Many exact estimates readers may re-use | Table |
A complete result has direction, magnitude, precision, and exact P
RR 0.66 (95% CI 0.48–0.91); P = 0.01
- Direction - which way
- Magnitude - how much
- Precision - the 95% CI
- P value - 0.01, not “< 0.05”
P is the adjunct; the CI is the message.
Never “P = 0.000” → report P < 0.001
Three sentences a reviewer will flag
“The mean age was 42.3 years, 58% male, 31% smokers, 22% diabetic, 40% hypertensive, 18% CKD, BMI 26.4, HbA1c 7.8%…”*
\[ \downarrow \]
“Participants had a mean age of 42 years and were predominantly male (58%); cardiometabolic comorbidity was common (Table 1).”
“Mortality was higher in the control arm because dexamethasone suppresses the harmful inflammatory cascade.”
\[ \downarrow \]
“28-day mortality was lower in the dexamethasone arm (22.9% vs 25.7%).”
Mechanism is Discussion.
“The intervention significantly reduced infection (p < 0.05).”
\[ \downarrow \]
“…reduced surgical-site infection by 6.2 points (12.1% vs 18.3%; RR 0.66, 95% CI 0.48–0.91; P = 0.01).”
⚡ Question Time
- A single overall prevalence
- Baseline characteristics, 12 variables × 2 groups
- Survival over 24 months by arm
- Participant flow
- Distribution of CRP in 30 patients
- Three subgroup odds ratios (deliberately ambiguous)
Answers
| # | Scenario | Decision |
|---|---|---|
| 1 | Single prevalence | Text |
| 2 | Baseline characteristics | Table |
| 3 | Survival by arm | Figure |
| 4 | Participant flow | Figure |
| 5 | CRP in 30 patients | Figure |
| 6 | Three subgroup ORs | Table or forest plot |
Tables
- Continuous, symmetric → mean ± SD
- Continuous, skewed → median (IQR)
- Categorical → n (%) with denominator
- Title above · footnotes below
- Abbreviations defined · no vertical rules
| Characteristic | Arm A (n=105) | Arm B (n=108) |
|---|---|---|
| Age, yr : mean ± SD | 54 ± 11 | 55 ± 10 |
| LOS, d : median (IQR) | 8 (5–12) | 9 (6–13) |
| Diabetes : n (%) | 31 (29.5) | 29 (26.9) |
Don’t put mean ± SD on length-of-stay. It is always right-skewed.
P-values in Table 1
- NO baseline p-values
- Randomisation handles balance by design
- Any baseline difference is a chance difference; testing it answers a question no one asked
- CONSORT discourages it
- Groups self-select → show comparability
- Use a standardised mean difference (SMD) or a group comparison
- Interpret as balance, not hypothesis-testing
The single most common Table 1 error in resident RCT manuscripts.
Match the figure type to the study design =
| Design | Key figures |
|---|---|
| All | Participant flow diagram |
| Cross-sectional | Ordered bar · prevalence map · forest (aOR) |
| Case-control | Forest (OR) · exposure distribution |
| Cohort | K–M · incidence · forest (subgroups) |
| Diagnostic | STARD flow · ROC · dot plot by status |
| RCT | CONSORT flow · K–M · forest · NNT |
| Syst. review | PRISMA flow · forest · funnel |
Maximise data-ink ratio
Remove
- 3-D, exploding slices
- gridlines, heavy borders
- redundant legends
Keep
- the data
- direct labels
- the one key finding
Generate your figures to be reproducible by design
Observational Results: prevalence, odds, person-time
- Prevalence + 95% CI; adjusted ORs (aOR)
- Weights / design effect if clustered (NFHS-style)
- Say “associated with” ; never “caused”
- Odds ratios, not risk ratios
- Matched design → conditional logistic (matched OR)
- Report exposure distribution by case/control
- Person-time, incidence rates
- Follow-up completeness & attrition before the HR
- HR + CI · K-M · acknowledge competing risks
Diagnostic Results: the 2×2, accuracy with CIs, and the ROC
| Disease + | Disease − | |
|---|---|---|
| Test + | TP | FP |
| Test − | FN | TN |
- Sens, Spec, PPV, NPV, LR± ; all with 95% CI
- ROC + AUC ; the sens/spec trade-off
- Report indeterminate / failed tests
- STARD flow accounts for everyone
Give the 2×2 explicitly so the reader can recompute anything. Accuracy varies with disease spectrum — report across it.
Take Home Message
Report what you found: direction, magnitude, precision and let the Discussion do the rest.
Text vs Table vs Figure