How to Report Statistical Results in Medical Research Papers Correctly

Q: How do you report p-values correctly in medical research?

Always report the exact p-value to three decimal places: p = 0.034, not p < 0.05. For very small p-values that round to zero at three decimal places, use p 0.999.

Q: How do you report a confidence interval in a medical research paper?

Report the 95% confidence interval alongside every point estimate (mean, OR, RR, regression coefficient, Cohen's d). The two most accepted formats are: (1) 95% CI [lower, upper] — APA 7th edition square bracket format; (2) (95% CI: lower–upper) — common in medical journals. Example: 'OR = 2.34 (95% CI: 1.42–3.86), p = 0.001.' Always report both the lower and upper bounds. Never omit the CI and report only the p-value. Confidence intervals are more informative than p-values because they show both the direction and precision of the estimate, not just whether it reaches a threshold.

Q: How do you report t-test results in a research paper?

The APA format for an independent samples t-test is: t(df) = value, p = value. Always include the degrees of freedom in parentheses and the exact p-value. Also report the group means, SDs, and effect size (Cohen's d with 95% CI). Full example: 'The intervention group showed significantly lower systolic blood pressure (M = 128.4, SD = 14.2) compared with controls (M = 134.6, SD = 15.8); t(118) = 2.31, p = 0.023, Cohen's d = 0.41 (95% CI 0.05–0.77).' For a paired t-test: 'Symptom scores decreased significantly from pre-intervention (M = 24.3, SD = 5.8) to post-intervention (M = 17.6, SD = 5.1); t(51) = 8.94, p < 0.001, Cohen's d = 1.24.'

Q: How do you report ANOVA results in a medical research paper?

Report ANOVA using the format: F(df_between, df_within) = value, p = value, followed by the effect size (partial eta squared or eta squared). Example: 'There was a statistically significant effect of treatment group on HbA1c, F(2, 147) = 8.42, p < 0.001, ηp² = 0.103.' Always follow a significant ANOVA with post-hoc comparisons, stating which method was used (Tukey HSD, Bonferroni, Scheffé): 'Post-hoc Tukey HSD comparisons revealed that Group A differed significantly from Group C (mean difference = 0.8%, 95% CI 0.3–1.3%, p = 0.002) but not from Group B (p = 0.14).'

Q: How do you report chi-square test results in medical research?

The correct format for chi-square is: χ²(df, N = total n) = value, p = value, plus an effect size (phi for 2×2 tables, Cramér's V for larger tables). Example: 'There was a statistically significant association between smoking status and myocardial infarction, χ²(1, N = 240) = 12.46, p < 0.001, φ = 0.23.' For a 3×2 table: 'χ²(2, N = 360) = 18.42, p < 0.001, V = 0.23.' Always include the total N in parentheses alongside the degrees of freedom. Report observed frequencies, expected frequencies, and percentages in the associated table.

Q: How do you report odds ratios in a research paper?

Report OR with the 95% confidence interval and p-value: 'Smoking was associated with significantly increased odds of myocardial infarction (OR = 3.21, 95% CI: 1.84–5.60, p < 0.001).' For adjusted OR from logistic regression, use aOR or OR_adjusted and specify the confounders adjusted for: 'After adjusting for age, sex, and diabetes, smoking remained independently associated with MI (aOR = 2.74, 95% CI: 1.52–4.94, p = 0.001).' Always provide the full logistic regression table with B, SE, Wald statistic, df, p-value, and OR with 95% CI for each predictor.

Q: How do you report linear regression results?

For the overall model: F(df_regression, df_residual) = value, p = value, R² = value. For each predictor: B = unstandardised coefficient, SE = standard error, β = standardised coefficient, t(df) = value, p = value. Full example: 'The regression model significantly predicted systolic blood pressure, F(3, 116) = 14.82, p < 0.001, R² = 0.277, indicating that the three predictors explained 27.7% of variance in SBP. Age was a significant predictor (B = 0.84, SE = 0.21, β = 0.33, t(116) = 4.00, p < 0.001) while BMI showed a trend toward significance (B = 1.12, SE = 0.60, β = 0.16, t(116) = 1.87, p = 0.064).'

Q: What is the correct way to report non-parametric test results?

For Mann-Whitney U test: 'Median ICU length of stay was significantly lower in the intervention group (Mdn = 4.5 days, IQR 2.0–8.5) compared with controls (Mdn = 6.0 days, IQR 3.5–11.0); U = 1842, p = 0.041, r = 0.21 (small effect).' For Wilcoxon signed-rank: 'There was a statistically significant decrease in pain score from pre-intervention (Mdn = 7.0) to post-intervention (Mdn = 4.0), z = −4.82, p < 0.001, r = 0.68 (large effect).' For Kruskal-Wallis: 'There was a statistically significant difference in CRP across treatment groups, H(2) = 18.34, p < 0.001.' Always report medians and IQRs with non-parametric tests, and always include an effect size measure.

Q: How do you report Pearson and Spearman correlation results?

For Pearson correlation: r(df) = value, p = value, where df = n − 2. Example: 'There was a significant positive correlation between age and systolic blood pressure, r(118) = 0.42, p < 0.001.' For Spearman correlation: rs(n) = value, p = value (convention is to report n rather than df for Spearman). Example: 'Serum CRP was significantly positively correlated with clinical severity score, rs(95) = 0.54, p < 0.001.' Always include: the correlation coefficient, its sign (positive/negative), the df or n, and the exact p-value. Add a verbal descriptor of strength: 0.1 = small, 0.3 = moderate, 0.5 = large.

Q: What are the most common mistakes when reporting statistical results in medical papers?

The ten most common errors: (1) Writing p = 0.000 instead of p < 0.001. (2) Writing 'p = ns' instead of the exact p-value. (3) Omitting confidence intervals from OR, RR, HR, and regression coefficients. (4) Reporting SEM instead of SD in descriptive statistics. (5) Omitting degrees of freedom from t, F, and chi-square statistics. (6) Not reporting effect size alongside p-values. (7) Reporting mean ± SD for skewed data that requires median (IQR). (8) Omitting post-hoc test results after a significant ANOVA. (9) Reporting 'significant difference' without stating which groups differed. (10) Using inconsistent decimal places within the same paper.

The Results section is where your research lives or dies. A study with impeccable design, rigorous data collection, and careful analysis can be rendered useless by a Results section that reports statistics incorrectly — failing peer review, misleading readers, or making results impossible to reproduce. Despite this, statistical reporting errors remain among the most common reasons for manuscript rejection and post-publication correction in clinical journals. Reporting "p = 0.000," omitting confidence intervals from odds ratios, pairing the median with a standard deviation, or writing "there was a significant difference" without specifying which groups differed — these are not trivial stylistic choices. They are substantive errors that undermine the scientific value of your work. This guide covers the correct format for every major statistical test used in medical research, with explicit correct-versus-incorrect comparisons, complete example sentences ready for thesis or journal submission, and the ten mistakes that most reliably trigger reviewer criticism.

Why Correct Statistical Reporting Matters

Correct statistical reporting is not about satisfying journal style guides. It serves three foundational functions that determine whether your research is usable by the scientific community:

Reproducibility: Readers must be able to verify your analysis from what you report. If you write "the difference was significant" without providing the test statistic, degrees of freedom, and exact p-value, no reader can check your calculation, and your result cannot be confirmed by independent re-analysis. Meta-analysts cannot extract your data. Systematic reviewers cannot include your study.
Clinical decision-making: Clinicians, guideline writers, and health technology assessment bodies rely on the numbers you report — not just their direction. An odds ratio of 2.34 (95% CI: 1.42–3.86) communicates fundamentally more than "the treatment group had significantly higher odds." The confidence interval tells the reader the plausible range of the true effect. Without it, clinicians cannot judge whether the effect is precise enough to act on.
Peer review and publication: Major journals including The Lancet, NEJM, JAMA, BMJ, and all journals following CONSORT, STROBE, or PRISMA guidelines now explicitly require effect sizes with confidence intervals and exact p-values for all inferential tests. Submissions that omit these elements are returned to authors before peer review even begins.

Governing Standards for Statistical Reporting in Medical Research APA 7th Edition (2020) requires effect sizes and exact p-values for all inferential tests. CONSORT 2010 requires effect sizes with 95% CIs for all primary and secondary outcomes in RCTs. STROBE requires complete regression outputs for observational studies. PRISMA requires forest plots with pooled effect estimates and heterogeneity statistics. ICMJE recommends exact p-values rather than threshold statements. When in doubt, report more, not less.

How to Report P-Values

The p-value is the most reported — and most misreported — statistic in medical research. These rules apply universally:

P-Value Reporting Rules (APA 7th & ICMJE)

p = 0.034    ✓ Report exact value to 3 decimal places
p < 0.001    ✓ Use when p rounds to 0.000 at 3 d.p.
p = 0.052    ✓ Report even when not significant
p > 0.999    ✓ Use when p rounds to 1.000

Lowercase, italic, no leading zero before decimal point is the APA convention (though most medical journals accept a leading zero: p = 0.034 is universally understood). The critical rules: exact value, three decimal places, never "ns," never "= 0.000," never just "< 0.05."

✗ INCORRECT

✓ CORRECT

p = 0.000

p < 0.001

p = ns

p = 0.214

p < 0.05

p = 0.034

p = 0.0001

p < 0.001

P = 0.23 (capital P)

p = 0.230

the result was significant (p<0.05)

the result was significant, p = 0.034

p = 0.000 is mathematically impossible — a p-value of exactly zero would require infinite sample size. "ns" is uninformative and rejected by all major journals. Always report the exact p-value so readers can judge effect precision independently of your significance threshold.

How to Report Confidence Intervals

The 95% confidence interval (95% CI) is not optional. Every point estimate — mean, proportion, OR, RR, HR, regression coefficient, Cohen's d, AUC — must be accompanied by its 95% CI. The CI communicates what the p-value cannot: the direction, magnitude, and precision of the estimated effect.

Confidence Interval Formats (Both Accepted in Medical Journals)

OR = 2.34 (95% CI: 1.42–3.86)    ✓ Medical journal format
OR = 2.34, 95% CI [1.42, 3.86]     ✓ APA 7th edition format
OR = 2.34 (1.42 to 3.86)            ✓ BMJ / Lancet inline format

Choose one format and apply it consistently throughout your paper. The en-dash (–) or the word "to" between bounds is preferred over a hyphen. Always label the confidence level (95%) explicitly; do not assume readers know it is 95%.

What the Confidence Interval Tells You That the P-Value Cannot A 95% CI for OR of (1.02–4.88) and a 95% CI of (2.10–2.64) might both have p < 0.05. But the first is so wide that it is consistent with anything from a near-null effect to a very large one — the study was underpowered and the result is imprecise. The second is narrow and clinically interpretable — the true OR is likely close to 2.37. CI width encodes the precision of your estimate. Report it every time.

How to Report Descriptive Statistics

Descriptive statistics are the foundation of every Results section. Correct formatting of means, medians, and their associated spread measures prevents the most basic reporting errors.

Descriptive Statistics Formats

M = 142.3, SD = 16.8        ✓ Mean + SD (normal data)
Mdn = 4.5, IQR [2.0, 8.5]    ✓ Median + IQR (skewed data)
n = 120 (54.5%)              ✓ Count + percentage (categorical)

Never report: Mean ± SEM for descriptive statistics (SEM is for inference, not description). Never pair Median with SD. Never omit the spread measure entirely.

✗ INCORRECT

✓ CORRECT

Mean SBP was 142.3 ± 1.5 mmHg

Mean SBP was 142.3 mmHg (SD 16.8 mmHg)

Median CRP was 54 mg/L (SD 38 mg/L)

Median CRP was 54.2 mg/L (IQR 28.0–108.0 mg/L)

Average age was 58 years

Mean age was 58.4 years (SD 12.1 years)

± 1.5 most likely represents SEM (SD/√120 = 1.5), not SD. Reporting SEM as the spread measure in a descriptive table understates variability by a factor of √n, falsely implying the data is far less variable than it is.

How to Report t-Test Results

Independent & Paired Samples t-Test

Format: t(df) = value, p = value, Cohen's d = value (95% CI)

Report the t statistic, degrees of freedom in parentheses, and exact p-value
Report means and SDs for both groups (or pre/post for paired)
Include Cohen's d with 95% CI as the effect size
For paired t-test, specify it is paired: t_paired or describe within-subject design in the sentence
Report the mean difference and its 95% CI for clinical interpretation

// Independent samples t-test — full correct reporting The intervention group demonstrated significantly lower systolic blood pressure
(M = 128.4 mmHg, SD = 14.2) compared with the control group (M = 134.6 mmHg,
SD = 15.8); t(118) = 2.31, p = 0.023, mean difference = −6.2 mmHg
(95% CI: −11.5 to −0.9), Cohen's d = 0.41 (95% CI 0.05–0.77),
indicating a small-to-medium effect.

// Paired t-test — pre-post design Symptom severity scores decreased significantly from pre-intervention
(M = 24.3, SD = 5.8) to post-intervention (M = 17.6, SD = 5.1);
t(51) = 8.94, p < 0.001, mean reduction = 6.7 points
(95% CI: 5.2–8.2), Cohen's d = 1.24 (95% CI 0.87–1.61), large effect.

✗ INCORRECT

✓ CORRECT

SBP was significantly lower in the intervention group (p < 0.05).

Mean SBP was lower in the intervention group (M = 128.4, SD = 14.2) vs controls (M = 134.6, SD = 15.8); t(118) = 2.31, p = 0.023, d = 0.41.

The incorrect version reports no means, no SDs, no test statistic, no df, no exact p-value, and no effect size. It provides no information a reader or meta-analyst can use.

How to Report ANOVA Results

One-Way and Factorial ANOVA

Format: F(df₁, df₂) = value, p = value, ηp² = value

Report F statistic with two sets of degrees of freedom: between-groups (numerator) and within-groups (denominator)
Include partial eta squared (ηp²) as the effect size — SPSS reports this automatically under Options
Always report which post-hoc test was used (Tukey HSD, Bonferroni, Scheffé) and its results for each significant pair
For repeated-measures ANOVA, report whether Mauchly's sphericity test was satisfied; if not, state whether Greenhouse-Geisser or Huynh-Feldt correction was applied
Report means and SDs for each group in the associated table or in the text

// One-way ANOVA — complete reporting There was a statistically significant effect of analgesic regimen on
post-operative pain score at 4 hours, F(2, 177) = 9.24, p < 0.001,
ηp² = 0.094 (medium effect). Post-hoc Tukey HSD comparisons revealed
that the combined regimen (M = 3.8, SD = 2.1) produced significantly
lower pain scores than NSAID alone (M = 5.9, SD = 2.2; p < 0.001,
d = 0.97) and opioid alone (M = 5.2, SD = 2.0; p = 0.003, d = 0.69).
NSAID and opioid regimens did not differ significantly (p = 0.282, d = 0.33).

✗ INCORRECT

✓ CORRECT

ANOVA showed a significant difference between groups (p = 0.001).

F(2, 177) = 9.24, p < 0.001, ηp² = 0.094. Post-hoc Tukey HSD: combined vs NSAID p < 0.001; combined vs opioid p = 0.003; NSAID vs opioid p = 0.282.

The incorrect version omits: the F statistic, degrees of freedom, effect size, which groups actually differed, and the post-hoc method. "Significant difference between groups" tells readers nothing clinically actionable.

How to Report Chi-Square Results

χ²

Chi-Square Test of Association

Format: χ²(df, N = n) = value, p = value, φ or V = value

Report the chi-square statistic, degrees of freedom, and total N in one parenthetical block
Include phi (φ) for 2×2 tables or Cramér's V for larger tables as the effect size
Report observed frequencies and percentages for each cell, not raw counts alone
If any expected cell count falls below 5, state that Fisher's Exact Test was used instead
Specify the comparison being described — do not just say "there was a significant association"

// Chi-square — 2×2 association Smoking was significantly associated with myocardial infarction at 10-year
follow-up. Among smokers, 42 of 120 (35.0%) developed MI compared with
18 of 120 (15.0%) non-smokers;
χ²(1, N = 240) = 12.46, p < 0.001, φ = 0.23 (small-to-medium effect).

// Chi-square — 3-group comparison (Cramér's V) Distribution of AKI severity stage differed significantly across treatment groups;
χ²(4, N = 360) = 18.42, p = 0.001, V = 0.16 (small effect).
[Always report the contingency table separately as Table X]

✗ INCORRECT

✓ CORRECT

There was a significant association between smoking and MI (chi-square, p = 0.001).

Smokers had significantly higher MI incidence (35.0%) than non-smokers (15.0%); χ²(1, N = 240) = 12.46, p < 0.001, φ = 0.23.

The incorrect version omits: the chi-square value, degrees of freedom, total N, observed proportions, and effect size. Writing "chi-square" as the test name without its statistic is insufficient for any peer-reviewed journal.

How to Report Odds Ratios and Relative Risk

Odds Ratio, Relative Risk, Hazard Ratio

Format: OR = value (95% CI: lower–upper), p = value

Every OR, RR, or HR must include a 95% confidence interval and p-value — the point estimate alone is never sufficient
Distinguish crude (unadjusted) OR from adjusted OR (aOR) and state what confounders were adjusted for
Report the reference category for all categorical exposures: "relative to non-smokers" or "referent: no diabetes"
For logistic regression, report the model fit statistics: −2 log likelihood, Cox & Snell R², Nagelkerke R², Hosmer-Lemeshow goodness-of-fit test
For Kaplan-Meier and Cox regression, the HR plays the same role as OR — same reporting rules apply

// Crude OR — cross-tabulation output Smoking was associated with significantly increased odds of myocardial infarction
(crude OR = 3.21, 95% CI: 1.84–5.60, p < 0.001). // Adjusted OR — logistic regression After adjusting for age, sex, diabetes, and hypertension, smoking remained
independently associated with MI (aOR = 2.74, 95% CI: 1.52–4.94, p = 0.001). // Relative risk — cohort or RCT The 5-year risk of cardiovascular events was 18.4% in the treatment group
vs 24.2% in the control group; RR = 0.76 (95% CI: 0.61–0.95), p = 0.014,
ARR = 5.8% (95% CI: 1.2–10.4%), NNT = 17 (95% CI: 10–83).

✗ INCORRECT

✓ CORRECT

Smokers had three times higher odds of MI (p = 0.001).

Smoking was associated with significantly increased odds of MI (aOR = 2.74, 95% CI: 1.52–4.94, p = 0.001), adjusted for age, sex, diabetes, and hypertension.

OR = 3.21, p < 0.05.

Crude OR = 3.21 (95% CI: 1.84–5.60), p < 0.001.

An OR without its CI is unpublishable. The confidence interval tells readers the precision of the estimate and is required for meta-analysis inclusion. "Three times higher" is also imprecise — always state the exact OR.

How to Report Regression Results

Linear Regression

Linear regression reporting requires two levels: the overall model and individual predictors. Both must be included in a complete results write-up.

Linear Regression — Complete Format

Model: F(df₁, df₂) = value, p = value, R² = value
Each predictor: B, SE, β, t(df), p

Report the unstandardised coefficient B (in outcome units), its standard error SE, the standardised coefficient β (beta, for comparing predictor strength), and t(df) with p-value.

// Linear regression — full example The regression model significantly predicted systolic blood pressure,
F(3, 116) = 14.82, p < 0.001, R² = 0.277, indicating that age, BMI,
and diabetes status collectively explained 27.7% of variance in SBP.

Age was a significant positive predictor
(B = 0.84, SE = 0.21, β = 0.33, t(116) = 4.00, p < 0.001).
BMI showed a trend toward significance
(B = 1.12, SE = 0.60, β = 0.16, t(116) = 1.87, p = 0.064).
Diabetes was not a significant predictor
(B = 2.84, SE = 3.10, β = 0.08, t(116) = 0.92, p = 0.361).

Logistic Regression

Logistic regression produces odds ratios (Exp(B) in SPSS output) for each predictor. Report the complete model as a table with B, SE, Wald, df, p, OR, and 95% CI for OR.

// Logistic regression model reporting Binary logistic regression was performed to examine predictors of 30-day
hospital readmission. The model was statistically significant,
χ²(5) = 28.42, p < 0.001, Nagelkerke R² = 0.184.
Hosmer-Lemeshow goodness-of-fit: χ²(8) = 6.14, p = 0.632 (adequate fit).

Age was a significant predictor of readmission
(B = 0.062, SE = 0.018, Wald = 11.88, df = 1, p = 0.001,
OR = 1.064, 95% CI: 1.027–1.102).
Prior admission within 30 days was the strongest predictor
(OR = 4.12, 95% CI: 2.18–7.78, p < 0.001).
[Present all predictors in a structured table — see Table X]

How to Report Non-Parametric Tests

Test	Correct Format	Effect Size
Mann-Whitney U 2 independent groups, non-normal	U = 1842, p = 0.041 Report medians and IQRs for each group	r = Z/√N 0.1/0.3/0.5 = sm/med/lg
Wilcoxon Signed-Rank Paired / pre-post, non-normal	z = −4.82, p < 0.001 Report medians and IQRs at each time point	r = Z/√N Same benchmarks as above
Kruskal-Wallis ≥3 independent groups, non-normal	H(2) = 18.34, p < 0.001 H = K-W statistic, df in parentheses	ε² or η² from H statistic
Spearman Correlation Non-normal bivariate relationship	rₛ(95) = 0.54, p < 0.001 Report n (not df) for Spearman	r itself is the effect size: 0.1/0.3/0.5
Fisher's Exact Small expected cell counts (<5)	State Fisher's Exact Test was used and report p-value only (no χ² statistic)	Phi (φ) or OR with 95% CI

// Mann-Whitney U — complete example Median ICU length of stay was significantly lower in the intervention group
(Mdn = 4.5 days, IQR 2.0–8.5) than in controls (Mdn = 6.0 days, IQR 3.5–11.0);
U = 1842, p = 0.041, r = 0.21 (small effect).

// Wilcoxon signed-rank — pre/post Pain scores decreased significantly from pre-intervention
(Mdn = 7.0, IQR 5.0–8.0) to post-intervention (Mdn = 4.0, IQR 3.0–6.0);
z = −4.82, p < 0.001, r = 0.68 (large effect).

How to Report Correlation Results

Pearson and Spearman Correlation Formats

Pearson: r(df) = value, p = value [df = n − 2]
Spearman: rₛ(n) = value, p = value [report n, not df]

Always include the sign of the correlation (positive/negative), the coefficient, and the exact p-value. Add a verbal descriptor: r = 0.42 = moderate positive correlation. Report 95% CI for r where possible (computed via Fisher's z transformation).

// Pearson correlation There was a statistically significant moderate positive correlation between age
and systolic blood pressure, r(118) = 0.42, p < 0.001
(95% CI: 0.26–0.56), accounting for 17.6% of variance in SBP (r² = 0.176).

// Spearman correlation (non-normal data) Serum CRP was significantly positively correlated with clinical severity score,
rₛ(95) = 0.54, p < 0.001, indicating a strong positive relationship.

Master Reference: Correct Reporting Formats at a Glance

Test	Correct Statistical Notation	Required Elements
P-value	p = 0.034 p < 0.001	Exact value to 3 d.p.; never "ns" or "0.000"
Confidence Interval	95% CI: 1.42–3.86	Both bounds, CI level, alongside every point estimate
Mean	M = 142.3, SD = 16.8	Mean + SD (not SEM); normal data only
Median	Mdn = 4.5, IQR [2.0, 8.5]	Median + IQR; skewed or non-normal data
Independent t-test	t(118) = 2.31, p = 0.023	t, df, p, means, SDs, Cohen's d with CI
Paired t-test	t(51) = 8.94, p < 0.001	t, df, p, pre/post means, SDs, Cohen's d_paired
One-way ANOVA	F(2, 177) = 9.24, p < 0.001	F, both df, p, ηp², post-hoc method and results
Chi-square	χ²(1, N=240) = 12.46, p < 0.001	χ², df, total N, p, phi or Cramér's V, counts/%, expected
Odds Ratio	OR = 2.74 (95% CI: 1.52–4.94)	OR, 95% CI, p, crude vs adjusted, reference category
Relative Risk	RR = 0.76 (95% CI: 0.61–0.95)	RR, 95% CI, p, ARR, NNT
Linear Regression	F(3, 116) = 14.82, p < 0.001, R²	Model F, df, p, R², each predictor: B, SE, β, t, p
Logistic Regression	χ²(5) = 28.42, p < 0.001, Nag. R²	Model fit, Nagelkerke R², each predictor OR with CI
Mann-Whitney U	U = 1842, p = 0.041, r = 0.21	U, p, r effect size, medians and IQRs
Pearson r	r(118) = 0.42, p < 0.001	r, df (= n−2), p, 95% CI for r, verbal descriptor
Spearman rₛ	rₛ(95) = 0.54, p < 0.001	rₛ, n (not df), p, verbal descriptor

Thesis Writing Recommendations

Theses in medicine and health sciences face different reporting expectations from journal articles — they must be self-contained, methodologically transparent, and demonstrate that the candidate understands what every statistic means. These conventions guide thesis statistical reporting:

🎓 Thesis-Specific Statistical Reporting Guidance

Declare your significance threshold in Methods: "Statistical significance was set at α = 0.05 for all analyses." Then report exact p-values throughout Results — never just "< 0.05." Your threshold is declared once; exact values are reported every time.

Describe every statistical test before using it: "An independent samples t-test was used to compare mean systolic blood pressure between groups, as data were normally distributed (Shapiro-Wilk: p = 0.38 for intervention, p = 0.42 for control)." Examiners expect you to justify each test choice.

Report normality test results: For every continuous variable, state the result of the Shapiro-Wilk or Kolmogorov-Smirnov test and the decision it produced. Examiners look for this in the Data Analysis section. Link normality test results directly to your choice of mean vs median and parametric vs non-parametric test.

Present full regression tables: Unlike journal articles that summarise key predictors in the text, theses should include complete regression output tables showing all predictors, even non-significant ones. This demonstrates analytic thoroughness. Include B, SE, β (or OR), t (or Wald), p, and 95% CI for each predictor.

Define abbreviations on first use in Results: The first time you report an SD, IQR, OR, CI, define it: "mean (standard deviation, SD)." Theses are read by non-specialist examiners who may need these anchors.

Use consistent decimal places throughout: Choose two decimal places for means and proportions, three for p-values, and two for ORs and CIs. Apply these consistently across all tables and text. Inconsistency signals careless analysis.

Report both statistical and clinical significance: After every significant finding, add a sentence interpreting its practical meaning: "This Cohen's d of 0.54 exceeds the MCID of 0.40 for this scale, indicating the difference is also clinically meaningful, not only statistically detectable."

Common Mistakes Researchers Make

Mistake 1: Writing p = 0.000

This is one of the single most common errors in published medical research. p = 0.000 is mathematically impossible — it would require infinitely precise data. When SPSS or other software displays "0.000" in output, this means the p-value is smaller than the display rounds to at three decimal places — typically p < 0.0005. The correct way to report this is p < 0.001, not p = 0.000, not p = 0.0001 unless you have verified the exact value from more precise output.

✓ Fix: Replace every instance of "p = 0.000" in your paper with "p < 0.001." In SPSS, access the exact p-value by double-clicking the output table to enter editing mode, which reveals additional decimal places.

Mistake 2: Writing "p = ns" or "not significant" without the p-value

"ns" (not significant) is a legacy notation that no major journal has accepted for decades. A non-significant result with p = 0.052 and a Cohen's d = 0.70 (underpowered study, large true effect) is scientifically very different from p = 0.83 with d = 0.04 (truly null). Reporting only "ns" erases this critical distinction. ICMJE guidelines explicitly require exact p-values regardless of significance.

✓ Fix: Report the exact p-value for every statistical test, significant or not. Write "there was no statistically significant difference between groups, t(58) = 1.24, p = 0.220, d = 0.32 (95% CI −0.19 to 0.83)." The CI spanning zero and the effect size both inform the reader about whether the study was powered to detect a meaningful effect.

Mistake 3: Omitting confidence intervals from odds ratios, relative risks, and regression coefficients

Reporting "OR = 2.34, p = 0.001" without the 95% CI is insufficient for publication in any peer-reviewed journal. The CI is not supplementary information — it is the primary evidence about the precision of the estimate. Meta-analyses cannot include your study without it. Guideline writers cannot use your result. Every OR, RR, HR, regression coefficient, and effect size must be accompanied by its confidence interval.

✓ Fix: Make a personal rule: the confidence interval is always reported in the same breath as the point estimate. Write a template: "OR = __ (95% CI: __–__), p = ___" and fill it in for every OR, RR, HR, and regression coefficient in your paper.

Mistake 4: Omitting degrees of freedom from test statistics

t = 2.31 and F = 9.24 are meaningless without their degrees of freedom. The df tells the reader the sample size context and allows them to look up critical values, check calculations, or re-analyse your data. Every t, F, and chi-square must appear with its df: t(118) = 2.31, F(2, 177) = 9.24, χ²(1, N=240) = 12.46. This is a mandatory requirement in APA format and across all major medical journal styles.

✓ Fix: Immediately after every t, F, or χ² symbol in your text, add the degrees of freedom in parentheses. For ANOVA, include both numerator and denominator df: F(2, 177). For chi-square, include df and total N: χ²(1, N = 240).

Mistake 5: Reporting SEM instead of SD in descriptive statistics

"Mean ± 1.5" likely represents SEM (= SD/√n), not SD. The SEM shrinks as sample size increases — it describes precision of the mean estimate, not variability in the data. Reporting SEM as if it were SD makes your sample look far less variable than it is. For n = 120, SEM = SD/√120 = SD/10.95, so substituting SEM for SD understates variability by nearly 11-fold. The correct measure for descriptive statistics tables is always SD.

✓ Fix: Explicitly label every spread measure in your tables and text: "(SD 16.8 mmHg)" or "(SEM 1.5 mmHg)" never just "±." Reserve SEM for graphs showing confidence around the mean and for inferential contexts like confidence interval calculation.

Mistake 6: Reporting a significant ANOVA without post-hoc results

A significant F-statistic tells you that at least one pair of group means differs. It does not tell you which pair. Writing "there was a significant effect of treatment group on HbA1c, F(2, 147) = 8.42, p < 0.001" and stopping there is clinically useless — readers cannot determine which treatment is better. Post-hoc test results (with the method specified) are mandatory after any significant omnibus ANOVA.

✓ Fix: Immediately after reporting the ANOVA F statistic, specify the post-hoc test used (Tukey HSD is the most common choice for equal group sizes; Bonferroni for smaller numbers of planned comparisons; Games-Howell if variances are unequal) and report the pairwise results with mean differences, 95% CIs, and p-values for each comparison.

Mistake 7: Not reporting effect sizes

APA 7th edition and CONSORT 2010 both require effect sizes for all inferential results. Yet a 2023 audit of 300 clinical trials found that 42% reported no effect size beyond the raw mean difference. Cohen's d for t-tests, partial eta squared for ANOVA, phi or Cramér's V for chi-square, and r for non-parametric tests are all straightforward to calculate and are now automatically provided by SPSS when you tick the "Effect size" option before running each test.

✓ Fix: Before finalising any results section, run a check: does every inferential test result include an effect size with 95% CI? If not, add it. The SPSS path: for t-tests and ANOVA, check "Effect size" in Options before running. For chi-square, compute phi or Cramér's V from the crosstab output.

Mistake 8: Reporting "the mean was significantly different" for skewed variables

If you have already established that your variable is right-skewed (ICU LOS, CRP, creatinine) and correctly reported it as median + IQR in your Table 1, you must compare groups using a non-parametric test and report the comparison as a Mann-Whitney U result — not as a t-test on means. Using a t-test on data you have explicitly described as skewed is an internal methodological contradiction that will be caught in peer review.

✓ Fix: Ensure consistency between your Table 1 descriptive statistics and your inferential tests. For each variable: if Table 1 shows median + IQR, the comparison test must be Mann-Whitney U (or Kruskal-Wallis). If Table 1 shows mean + SD, the comparison test must be t-test (or ANOVA). Cross-check this before submission.

Mistake 9: Confusing adjusted and unadjusted results

In regression analyses, crude (unadjusted) and adjusted estimates often differ substantially. Reporting only the crude OR without the adjusted OR (or vice versa) gives an incomplete picture. Conversely, labelling an adjusted OR as "OR" without specifying it is adjusted, or listing confounders without stating they were controlled for, prevents readers from evaluating the validity of the adjustment.

✓ Fix: Clearly label crude estimates as "crude OR" or "unadjusted OR" and adjusted estimates as "adjusted OR" or "aOR," specifying which variables were controlled: "aOR = 2.74 (95% CI: 1.52–4.94), adjusted for age, sex, diabetes, and hypertension." Provide both crude and adjusted estimates in regression tables so readers can see the impact of adjustment.

Mistake 10: Inconsistent decimal places and number formatting

"p = 0.038" in one table and "p = .038" in another (missing leading zero), or "OR = 3.2" and "OR = 2.74" in the same paper (inconsistent decimal places), signals lack of attention to detail that undermines confidence in your analysis. These inconsistencies are among the easiest errors for editors and reviewers to spot, and they suggest the results section may not have been carefully proofread.

✓ Fix: Establish a number formatting policy at the start of writing and apply it everywhere: p-values to 3 decimal places; ORs, RRs, and correlation coefficients to 2 decimal places; means to 1 decimal place more than the raw measurement precision; percentages to 1 decimal place. Use Find & Replace to check consistency across the document before submission.

Frequently Asked Questions

How do you report p-values correctly in medical research? +

Always report the exact p-value to three decimal places: p = 0.034, not p < 0.05. For very small values that round to zero at three decimal places, write p < 0.001. Never write p = 0.000 (mathematically impossible), p = ns (uninformative), or p = 0.0001 unless you have verified the exact value from SPSS high-precision output. Always italicise p in APA format. Report p-values for significant and non-significant results alike — a non-significant p = 0.052 with Cohen's d = 0.70 is critically different from p = 0.852 with d = 0.03, and both deserve their exact values.

How do you report a confidence interval in a medical research paper? +

Report the 95% CI alongside every point estimate. Accepted formats: (95% CI: 1.42–3.86) for medical journals, or 95% CI [1.42, 3.86] for APA style. Include both lower and upper bounds, always label the confidence level, and use an en-dash (–) between bounds, not a hyphen. Every mean difference, OR, RR, HR, regression coefficient, Cohen's d, and AUC needs its confidence interval. The CI communicates the precision and plausible range of your estimate in a way that the p-value fundamentally cannot.

How do you report t-test results in a research paper? +

APA format: t(df) = value, p = value. Always include degrees of freedom in parentheses, report both group means and SDs, and include Cohen's d with 95% CI. Full example: "The intervention group had significantly lower SBP (M = 128.4, SD = 14.2) compared with controls (M = 134.6, SD = 15.8); t(118) = 2.31, p = 0.023, Cohen's d = 0.41 (95% CI 0.05–0.77), indicating a small-to-medium effect." For paired t-tests, use the same format but report pre/post means and state that it is a paired design, and use Cohen's d_paired (M_diff / SD_diff).

How do you report ANOVA results in a medical research paper? +

Format: F(df_between, df_within) = value, p = value, ηp² = value. Always include two degrees of freedom (between-groups and within-groups), report the partial eta squared effect size, and follow every significant ANOVA with named post-hoc comparisons. Example: "F(2, 147) = 8.42, p < 0.001, ηp² = 0.103. Post-hoc Tukey HSD showed Group A differed significantly from Group C (mean difference = 0.8%, 95% CI 0.3–1.3%, p = 0.002) but not from Group B (p = 0.144)." If variances are unequal (Levene's test p < 0.05), use Welch's ANOVA and Games-Howell post-hoc instead.

How do you report chi-square test results in medical research? +

Format: χ²(df, N = total n) = value, p = value, φ = value (or Cramér's V for non-2×2 tables). Always include the total N alongside df. Report observed frequencies and percentages for each cell. Example: "Smokers had significantly higher 10-year MI incidence (35.0%) than non-smokers (15.0%); χ²(1, N = 240) = 12.46, p < 0.001, φ = 0.23." If any expected cell count is less than 5, use Fisher's Exact Test and report only the p-value (no chi-square statistic) along with an appropriate effect size.

How do you report odds ratios in a research paper? +

Format: OR = value (95% CI: lower–upper), p = value. Distinguish crude from adjusted: label them "crude OR" and "aOR" and specify the confounders adjusted for. Always state the reference category. Example: "After adjusting for age, sex, diabetes, and hypertension, smoking remained independently associated with MI (aOR = 2.74, 95% CI: 1.52–4.94, p = 0.001, reference: never-smokers)." For logistic regression, provide a complete table with B, SE, Wald, df, p, and OR (95% CI) for every predictor — not just the significant ones.

How do you report linear regression results? +

Two levels are required. Model level: F(df_regression, df_residual) = value, p = value, R² = value. Predictor level: B (unstandardised), SE, β (standardised, for comparing predictor importance), t(df), p. Example model: "F(3, 116) = 14.82, p < 0.001, R² = 0.277 (27.7% of variance explained)." Example predictor: "Age: B = 0.84, SE = 0.21, β = 0.33, t(116) = 4.00, p < 0.001." In theses, include the complete regression table with all predictors. Adjusted R² is preferred over R² when comparing models with different numbers of predictors.

What is the correct way to report non-parametric test results? +

Always report medians and IQRs (never means and SDs) with non-parametric tests. Mann-Whitney U: "U = 1842, p = 0.041, r = 0.21." Wilcoxon signed-rank: "z = −4.82, p < 0.001, r = 0.68." Kruskal-Wallis: "H(2) = 18.34, p < 0.001." Effect size for Mann-Whitney and Wilcoxon: r = Z/√N (benchmarks: 0.1 = small, 0.3 = medium, 0.5 = large). Always state that the test was non-parametric because of non-normal distribution, and specify which normality test confirmed this.

How do you report Pearson and Spearman correlation results? +

Pearson: r(df) = value, p = value, where df = n − 2. "There was a moderate positive correlation between age and SBP, r(118) = 0.42, p < 0.001." Spearman: rₛ(n) = value, p = value, reporting n (not df). "CRP was positively correlated with severity score, rₛ(95) = 0.54, p < 0.001." For both: include the correlation's sign (direction), add a verbal descriptor of strength (small/moderate/large), and where possible include the 95% CI for the coefficient (calculated via Fisher's z transformation). Report the coefficient of determination (r²) to describe variance explained.

What are the most common mistakes when reporting statistical results in medical papers? +

The ten most frequent errors: (1) p = 0.000 instead of p < 0.001. (2) "p = ns" instead of the exact value. (3) Missing confidence intervals from OR/RR/HR/regression coefficients. (4) Reporting SEM instead of SD in descriptive statistics. (5) Missing degrees of freedom from t, F, and chi-square. (6) No effect size alongside p-values. (7) Mean ± SD for skewed data requiring median (IQR). (8) No post-hoc results after a significant ANOVA. (9) "Significant difference" without specifying which groups or directions. (10) Inconsistent decimal places within the same paper. Addressing these ten errors will resolve the majority of statistical reporting reviewer comments across all medical journals.

Get your statistical results right the first time

StatClinic's AI assistant analyses your data, runs the correct tests, and formats results in journal-ready language — p-values, CIs, effect sizes, and all.

Open StatClinic →