Show Summary Details
Page of

Writing up a research study 

Writing up a research study
Writing up a research study

Janet L. Peacock

, Sally M. Kerry

, and Raymond R. Balise

Page of

PRINTED FROM OXFORD MEDICINE ONLINE ( © Oxford University Press, 2016. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford Medicine Online for personal use (for details see Privacy Policy and Legal Notice).

date: 26 April 2018

  1. 4.1 Introduction [link]

  2. 4.2 Contents of each section of the report [link]

  3. 4.3 Special circumstances [link]

4.1 Introduction

A research study report may be written for any one of several different purposes, each requiring a specific format. Examples include an internal report, a report for a funding body or committee, a paper for a journal, a dissertation for an undergraduate or postgraduate degree, one of the UK Royal Medical College examinations or a thesis (e.g. PhD). The general structure is similar for all of these and follows the usual pattern for writing up the results of a scientific experiment—introduction, methods, results, and discussion (box 4.1). Exceptions are short reports and research letters, which are discussed separately in section 4.3.2. The length and relative balance of each of these sections varies according to the purpose and specified format. For example, a paper for a journal will tend to be fairly short, often less than 3000 words, whereas a dissertation may be up to 20,000 words, and a thesis considerably longer.

We outline in section 4.2 the contents of each of the sections of a research study report, with particular reference to the statistical aspects.

4.2 Contents of each section of the report

4.2.1 Abstract

The main requirement for an abstract is that it should be brief and yet be a stand-alone document. It is the first thing that most people will read and, more importantly, may be all that is read if the reader has limited time or restricted access to the full document, such as when obtaining abstracts through online journals or databases.

The abstract should state the purpose of the study and briefly describe the study design, study subjects, and the variables measured. The results section should summarize the key findings on the main variables of interest and should provide estimates of sizes of effects with confidence intervals wherever possible, as well as P values. Valid conclusions should be drawn without overstating or understating the interpretation of the findings. Common problems include interpreting simple associations as causal, assuming that statistical significance implies clinical significance and conversely assuming that ‘not statistically significant’ means that there is no effect or difference. Understating conclusions is less common but some abstracts end with a statement along the lines of, ‘The risk factor may be related to the disease’, which probably could have been said without doing the study.

A structured abstract may be required with a specified work limit. Examples of specifications are given in box 4.2, and box 4.3 gives an example of a structured abstract.

Reproduced from Emergency Medicine Journal, Peacock PJ, Peacock JL, Victor CR, Chazot C, Changes in the emergency workload of the London Ambulance Service between 1989 and 1999, 22, 1, pp. 56–59, copyright 2005, with permission from BMJ Publishing Group Ltd.

4.2.2 Introduction

This sets the scene and describes the background to the study—what is already known about the topic, what are the gaps in knowledge, and how the proposed study will add to this. In a journal article this section is likely to be short but will be much more detailed in a dissertation, report, or thesis.

4.2.3 Methods

The purpose of the methods section is to describe the conducting of the study in sufficient detail for another researcher to be able to repeat the study. However, in a journal article as opposed to a dissertation or report, space is usually too limited for this to be possible. The use of online web-based appendices has helped to remedy this problem by allowing the brief details in a paper to be supplemented by fuller details online.

The methods section should include details of the setting or area where the study was conducted, the subjects included, the study design, technical details of any measurements made, the rationale for the chosen sample size, and the statistical methods used to analyse the data. Some studies use routine data and so the description of the subjects may only need to state the time period (see box 4.3).

In other situations, subjects may be selected according to set criteria or diagnoses and these should be stated. Alternatively, the subjects may be an unselected series of available patients in a time period. However the data or subjects were selected, it is important to be able to demonstrate that the selection was done in a systematic way that will enable the study’s findings to be generalized. Box 4.4 gives an example of sample description taken from the UKOS study.

From The New England Journal of Medicine, Johnson A et al., High frequency oscillatory ventilation for the prevention of chronic lung disease of prematurity, 347, 9, p. 633. Copyright © 2002 Massachusetts Medical Society. Reprinted with permission from Massachusetts Medical Society.

The description of the study design should include the type of study; for example, cross-sectional survey, case-control, cohort, randomized controlled trial, etc. For case-control studies, the method of choosing controls and the definition of cases should be described. If cases are matched to controls then the method of matching should be described in sufficient detail to explain exactly how controls were chosen. For example, ‘cases were one-to-one matched to within 2 years using the General Practice age/sex register’.

Where formal sample size calculations have been done, these should be reported (see box 4.5) and where they are inappropriate or not possible, then it is helpful to say so. Sometimes, the original sample size estimates proved to be unachievable and therefore modified estimates are made. Box 4.5 gives examples of different scenarios.

For further information see Bruce M, Peacock JL, Iverson A, Wolfe C. Hepatitis B and HIV antenatal screening 2: user survey. British Journal of Midwifery 2001; 9:640–645; Johnson AH, Peacock JL, Greenough A, Marlow N, Limb ES, Marston L et al. High-frequency oscillatory ventilation for the prevention of chronic lung disease of prematurity. N Engl J Med 2002; 347:633–642; and Thomas MR, Rafferty GF, Limb ES, Peacock JL, Calvert SA, Marlow N et al. Pulmonary function at follow-up of very preterm infants from the United Kingdom oscillation study. Am J Respir Crit Care Med 2004; 169:868–872.

Chapter 3 describes how to report sample size statements in a protocol, and the same principles apply to reporting them in a paper. Further discussion of reporting sample size for a randomized clinical trial is given in chapter 12.

The report should state all statistical methods used, including methods to address missing data and any assumptions made about the data. Although the statistical package or software employed should be specified it is not enough merely to name it, as in ‘the data were analysed using SPSS’, or to say that ‘parametric methods were used’. If the statistical method is not a standard technique, then a reference should be given. In a longer report there is room to give a full justification of the methods used; box 4.6 gives a detailed example. The report of the statistical analysis clearly states the main outcome and the predictor variables, and describes the type of data—here, categorical data in three and two categories, respectively. Then the actual analysis (chi-squared test for trend and logistic regression) is stated. Finally, the reader is told how the results will be presented (odds ratios and 95% confidence intervals) and what statistical package was used. Further examples of how to describe specific statistical methods are given throughout the book.

Sometimes the researcher may not know in advance exactly which methods will be used, as this can depend on early findings. In such circumstances it is not obvious whether to include these in the ‘methods’ section as if they were determined in advance, or to describe them in the ‘results’ section. We advise that all methods are described in the methods section if at all possible, unless the text flows better if they are included with the results.

4.2.4 Results

The results section should begin by describing the basic characteristics of the study population. This should include the total numbers of subjects or observations with a breakdown of these numbers to show the reasons for missing data, e.g. refusal, non-response, dropout, data not recorded, etc. If the study is comparing groups, as in a randomized trial, then baseline data for the groups should be given. If there is a lot of information it may be easier for the reader to assimilate if these data are presented in a table, but where there are many baseline variables the choice of which to present is a matter of judgement. Table 4.1 gives an example of summarizing a subset of data from a larger sample. The authors have previously used this as a teaching exercise to illustrate how different types of data can be incorporated into just one table. (We noticed that less experienced students and researchers tended to put each variable in a separate table, which not only wasted space but also made the results rather disjointed.) Guidelines on how to describe baseline characteristics can be found in chapter 5.

Table 4.1 Baseline characteristics of a sample


Characteristics of study group: 230 pregnant women who missed antenatal clinic appointments



Mean (SD) or %

Age (years)


26.1 (5.5)

Height (cm)


161.6 (6.3)

Weight (kg)


61.7 (10.4)

Alcohol (g/week)


17.3 (33.6)

Birthweight of baby (g)


3280 (477)

Marital status













Social class














     More than minimum






Current smoking










The main results of statistical analyses can often be summarized in tables and graphs. Missing data should be accounted for wherever possible so that the numbers ‘add up’. Often totals will vary from table to table by a small amount, either because data are missing for that variable or because subjects do not answer a particular question. In such cases it is usually sufficient to give the maximum total and then to say, ‘numbers vary slightly from table to table due to missing data’.

It is often easier to assimilate several sets of numbers when they are in table form rather than given in the text. This is certainly possible when writing a report, dissertation, or thesis, where there is usually space to include many tables. However, this may not be possible when writing a journal article, as there may be a limit to the number of tables allowed. The text itself should describe and summarize the important features in terms of the sizes of effects, the differences between groups, or the strengths of associations, etc., as appropriate. It is unnecessary to repeat information, such as a difference, its confidence interval, and P value, when that information is already presented in tables.

4.2.5 Discussion

Many aspects of the discussion will centre on interpreting the findings in the context of previous work and current knowledge. Such discussion may not be intrinsically statistical. However, there are several statistical issues which may require discussion or comment. Box 4.7 lists some of these.

It can be useful to set the sizes of effects and widths of confidence intervals in the context of current knowledge. Box 4.8 gives an example from a study of the adverse health effects of outdoor air pollution, which had, in general, found very similar effect sizes to those previously reported by others. For this particular outcome the estimate was higher than one reported previously, but the current study’s 95% confidence interval was wide. Hence, the two studies’ findings were not inconsistent with each other.

For further information see Peacock JL, Symonds P, Jackson P, Bremner SA, Scarlett JF, Strachan DP et al. Acute effects of winter air pollution on respiratory function in schoolchildren in southern England. Occup Environ Med 2003; 60:82–89.

In some studies, many hypothesis tests are performed, increasing the possibility of spurious significant results (type 1 errors). Where a single and unexpected significant result has been found it is sensible to view this cautiously and discuss the possibility that it is a false positive finding. This problem is common in exploratory studies.

Sometimes there are unavoidable limitations in the design or statistical analysis used. This can often happen with student projects where there are tight time constraints, but may also happen with other studies. In such cases any limitations and their potential implications should be clearly described and discussed. In practice, no study is perfect and all have some limitations. In a well-designed study, the limitations will be outweighed by the strengths, and the results will be robust.

4.3 Special circumstances

4.3.1 Writing abstracts for conferences

Like abstracts for reports, abstracts for oral and poster conference presentations must stand alone, because conference abstracts in particular are often published in their own right. The specification for these abstracts varies from conference to conference but may be structured and will almost certainly have a word or character limit (box 4.2).

4.3.2 Short reports and research letters

Some journals allow authors to submit short reports or research letters, which are typically 500–1000 words in length with only one table or figure allowed. They do not always fit the usual structure and sections, such as introduction and methods, can be combined. It can be difficult to write for this type of format because of these constraints but it is ideal for a small piece of work.