Medical dissertations

From master chart to results chapter: building thesis tables that examiners can follow

Q: Should the master chart be included in the thesis?

Follow your university's rule. If it is bound in as an annexure, include only de-identified data, with a key to the codes taken from the data dictionary.

Q: Can I enter data directly into statistical software instead of a spreadsheet?

Yes. The same rules apply: one row per participant, defined codes, a data dictionary and an untouched original file. What matters is the structure, not the program.

Q: How many tables should a results chapter have?

One for each objective, plus a participant flow figure and a characteristics table. The ICMJE guidance is to restrict tables and figures to those needed to explain the argument and assess the supporting data.

Q: Where do missing values go in a table?

Report the number missing for each variable in the table itself or its footnote, and state in the methods how missing data were handled in the analysis.

How to lay out a thesis master chart, write the coding sheet and data dictionary, draft table shells before analysis, and turn one table into results text.

Editverse Medical editorial deskPublished 10 November 2026Guideline versions checked November 2026

Key points

A master chart is one rectangular sheet: one row per participant, one column per variable, one value per cell, with no merged cells, colours or totals.
Write the coding sheet and data dictionary before entering data. They fix variable names, units, codes, allowed ranges and the missing-value rule.
Draft empty table shells from the objectives before any analysis. The shells decide what is analysed, not the other way round.
Every table gives counts with percentages, a stated summary measure with its spread, and estimates with confidence intervals.
The results text follows the tables in a fixed order: participants, descriptive data, outcomes, main results, other analyses.

What the master chart is, and what it is not

The master chart is the complete participant-level dataset of the thesis, usually kept as a spreadsheet and, where the university asks for it, bound into the thesis as an annexure. It is the raw material for every table. It is not a place for calculations, summaries or formatting. Guidance on organising data in spreadsheets, published in The American Statistician in 2018, sets out working rules for laying out a data sheet so that it is less error-prone and easier to analyse.¹ Applied to a thesis they are:

Make the sheet a single rectangle. One header row holds the variable names. Each later row is one participant and each column is one variable.
Put one piece of information in each cell and do not merge cells. Blood pressure of 130/80 is two columns, not one. Units belong in the variable name or the data dictionary, not in the cell.
Be consistent: one code for each category, one missing-value code, one form for each participant identifier and variable name, one date format throughout.
Write dates as YYYY-MM-DD, the ISO 8601 form, and check that the spreadsheet program has not altered them.
Leave no cell empty. Enter a defined missing code so that "not recorded" can be told apart from "not yet entered".
Do not use highlighting or font colour to carry information. If a flag or a note matters, give it a column.
Keep calculations and graphs out of the raw sheet. Derive body mass index, scores and age groups in a separate analysis file or a copy.
Use short but meaningful variable names without spaces or special characters, and keep the full description in the data dictionary.
Write-protect the original file, work on a copy, and keep regular backups in more than one place.

Replace names and hospital numbers with a study identifier in the master chart. Keep the key that links identifier to patient in a separate, access-controlled file, as your ethics approval describes.

The coding sheet and data dictionary

The coding sheet is written from the proforma before the first patient is entered. One row describes one column of the master chart. The example is illustrative, for an invented surgical thesis comparing two wound dressings.

Illustrative data dictionary for a fictional thesis dataset
Variable name	Description	Type	Units or codes	Allowed range	Missing code
pid	Study identifier	Text	D001 to D194	Unique	None allowed
group	Dressing received	Categorical	1 = new, 2 = standard	1 or 2	None allowed
age_y	Age at operation	Numeric	Completed years	18 to 95	NA
sex	Sex recorded in the case file	Categorical	1 = female, 2 = male	1 or 2	NA
bmi	Body mass index at admission	Numeric	kg/m², one decimal	14.0 to 60.0	NA
dm	Diabetes mellitus on treatment	Categorical	0 = no, 1 = yes	0 or 1	NA
asa	ASA physical status grade	Ordinal	1 to 4	1 to 4	NA
op_date	Date of operation	Date	YYYY-MM-DD	Within study period	None allowed
los_d	Postoperative length of stay	Numeric	Days	1 to 90	NA
ssi_30	Surgical site infection within 30 days, by the protocol definition	Categorical	0 = no, 1 = yes	0 or 1	NA

Three decisions belong in this sheet and nowhere else: the direction of every code (is 1 "yes" or "female"?), the unit and number of decimals for every measurement, and the single code that means missing. Spreadsheet data validation can then reject values outside the allowed range at the point of entry.

Cleaning before analysis

Run these checks and keep a dated log of each correction, with the source document that justified it.

Count rows. The number must equal the participants enrolled, with no duplicate identifiers.
List the minimum and maximum of every numeric variable and compare with the allowed range.
Tabulate every categorical variable. Any code not in the dictionary is an entry error.
Check date logic: admission before operation, operation before discharge, follow-up after operation.
Count missing values for every variable. STROBE item 14(b) asks for the number of participants with missing data for each variable of interest, so this count goes into the thesis.⁴
Check cross-field logic, such as pregnancy recorded only for female participants.

Table shells before analysis

A table shell is the finished table without numbers: title, row labels, column heads, units and footnotes. Drafting shells from the objectives in the approved synopsis keeps the analysis to what was planned and makes exploratory additions visible. The ICMJE Recommendations, updated January 2026, ask authors to distinguish prespecified from exploratory analyses, and shells dated before analysis are the simplest evidence of which is which.³

Standard table set for a two-group observational thesis, mapped to STROBE items
Table or figure	Content	STROBE item
Figure 1: flow of participants	Numbers potentially eligible, examined, confirmed eligible, included, completing follow-up and analysed, with reasons for non-participation at each stage.	13(a) to 13(c)
Table 1: characteristics	Demographic and clinical characteristics, exposures and potential confounders, by group, with missing data for each variable.	14(a), 14(b)
Table 2: outcomes	Numbers of outcome events or summary measures by group.	15
Table 3: main results	Unadjusted and, where applicable, confounder-adjusted estimates with 95% confidence intervals, and the confounders adjusted for.	16(a)
Table 4: other analyses	Subgroup, interaction and sensitivity analyses, labelled as prespecified or exploratory.	17

A shell for the main results table looks like this. Agree it with your guide before running a single test.

Table shell: association between [exposure] and [primary outcome]
Exposure group	Participants, n	Outcome events, n (%)	Unadjusted estimate (95% CI)	Adjusted estimate (95% CI)
Reference group	xx	xx (xx.x)	Reference	Reference
Comparison group	xx	xx (xx.x)	x.xx (x.xx to x.xx)	x.xx (x.xx to x.xx)

Footnote to draft now: the effect measure used, the variables adjusted for, the model, and the number of participants excluded from the adjusted model because of missing covariates.

One worked table

The table below uses invented data for the fictional dressing study. It shows the conventions examiners look for, drawn from the SAMPL guidelines and the ICMJE Recommendations: absolute numbers with every percentage, the denominator in the column head, mean with standard deviation for roughly symmetric data and median with interquartile range for skewed data, units in the row label, and the summary measure named in each row.^2,3

Illustrative Table 1 and primary outcome for a fictional study (invented data)
Characteristic	New dressing (n = 96)	Standard dressing (n = 98)
Age, years, mean (SD)	52.4 (13.1)	53.8 (12.6)
Female, n (%)	41 (42.7)	44 (44.9)
Diabetes mellitus, n (%)	22 (22.9)	25 (25.5)
Body mass index, kg/m², median (IQR)	25.9 (23.1 to 29.4)	26.3 (23.4 to 29.8)
ASA grade 3 or 4, n (%)	13 (13.5)	15 (15.3)
Surgical site infection within 30 days, n (%)	9 (9.4)	19 (19.4)

SD, standard deviation; IQR, interquartile range; ASA, American Society of Anesthesiologists. Body mass index missing for 3 participants (new dressing 1, standard dressing 2). Risk difference for surgical site infection, new minus standard: −10.0 percentage points (95% CI −19.8 to −0.3); chi-square test, P = 0.047.

Check the arithmetic the way an examiner will. Group sizes add to 194, the number enrolled. Each percentage can be recomputed from its count and the column denominator: 9 of 96 is 9.4%, 19 of 98 is 19.4%. Percentages are given to one decimal because the denominators are near 100. The footnote explains every abbreviation and states where data are missing.

Turning the table into text

The ICMJE guidance is not to repeat all the data from tables in the text but to emphasise the most important observations.³ For this table a results paragraph could read:

"Of 194 participants, 96 received the new dressing and 98 the standard dressing. The groups were similar in age, sex, diabetes, body mass index and ASA grade (Table 1). Surgical site infection within 30 days occurred in 9 of 96 participants (9.4%) with the new dressing and 19 of 98 (19.4%) with the standard dressing, a risk difference of −10.0 percentage points (95% CI −19.8 to −0.3; P = 0.047)."

The paragraph reports the estimate and its interval and stops there. It does not call the result "highly significant" or claim the new dressing "prevents" infection. The interval runs from a large reduction to almost none, and that uncertainty is the finding. A P value close to 0.05 is not a different kind of evidence from one just above it, and it does not measure the size or importance of an effect.⁷ Interpretation belongs in the discussion chapter.

Presentation rules that save a revision

Frequent table faults and the fix
Fault	Fix
Percentages without counts	Give n (%) and state the denominator. Report absolute numbers from which every derived figure was calculated.
Mean ± SD for skewed data	Use median with interquartile range, and name the measure in the row label.
"±" with no label	Say whether the value is a standard deviation or a standard error.
P values alone	Report the estimate with its confidence interval first. Give exact P values, not "NS" or "P < 0.05".
Too many decimals	Use the precision of the measurement. Ages to one decimal at most; percentages to one decimal for samples in the hundreds.
Same data in a table and a chart	Choose one. Use a chart only where it shows a pattern a table cannot.
Totals that change between tables	State the number analysed in every table and explain any difference in a footnote.
Unnamed test	Name each test in the methods or a footnote, and name the software with its version.

Two sets of journal guidelines written for clinical researchers give more detail on precision, P values and table design and are worth reading once before formatting a results chapter.^5,6

Where support fits

The candidate collects the data and decides, with the guide, what the thesis claims. Statistical support covers the technical steps between a clean master chart and a written chapter. For thesis statistics and results chapter work, the candidate supplies the dataset, the codebook, the objectives and hypotheses and the approved synopsis. An Editverse Medical statistician runs the analysis in SPSS, R or Stata and returns the output, tables, figures, a results narrative and the syntax or code, so the candidate can reproduce and defend every number. Candidates who have already run their analysis and want the output explained can use statistical results interpretation. The full range of thesis services is on the medical dissertations page. Work follows the university's rules and the reference style it names.

Frequently asked questions

Should the master chart be included in the thesis?

Follow your university's rule. If it is bound in as an annexure, include only de-identified data, with a key to the codes taken from the data dictionary.

Can I enter data directly into statistical software instead of a spreadsheet?