Your health comes first. This educational article provides general information. Discuss your individual circumstances, pregnancy, illness or symptoms with your clinician; the article cannot replace an individual assessment.

A headline features thousands of marathon runners, and the conclusion offers a memorable training pattern. But who entered the analysis, and whose story stayed outside? Finishers’ data can tell us a great deal about performance, while other questions need other samples. Our own numerical example and three original studies help explain how to read research findings without mistaking them for a ready-made personal plan.

A large sample is not the same as the whole running community

There are several doors into a research database. First someone participates in an event or uses an app, then consents to data use or answers a questionnaire, and finally meets the analysis criteria. Who remains can change at every step. A final sample numbering thousands or hundreds of thousands does not itself reveal which running stories are missing.

The question is not whether every filter is bad. Data-quality rules may be needed to prevent incorrect files and inadequately identified events from misleading the result. The point is to make the consequences of filtering visible. All starters, all finishers and finishers with complete, high-quality heart-rate training diaries may describe different groups.

Your own reading question determines which sample is useful. If you are interested in finishers’ pacing, analysing those who completed may be a natural starting point. But if you ask how many people following a preparation approach could reach the start and finish at all, data containing only finishers no longer gives a complete answer. Sample suitability must always be judged in relation to the question.

Two fictional groups reveal the missing column

The next example is entirely our own illustration, not real research. In each of two groups, one hundred people prepare for an event. In group A, twenty do not complete the programme because of injury and eighty finish; sixty of those achieve the fast time threshold specified beforehand in the example. In group B, five drop out because of injury, ninety-five finish and fifty run under the threshold.

Looking only at finishers, group A’s proportion of fast runners is sixty divided by eighty, or seventy-five per cent. In B it is fifty divided by ninety-five, approximately 52.6 per cent. This correctly describes the finishers. Yet an important story is absent: injury-related dropout among the original participants was twenty in A and five in B. The two questions need different denominators and different outcomes.

The example does not prove programme A caused more injuries. We assigned no groups randomly and did not specify their histories. It therefore cannot rank training systems. The illustration helps reveal that a more favourable performance proportion among finishers is not the same as a better whole process for every starter. The missing column changes the question that can be asked.

Our own fictional example of different denominators; not an actual research finding.
Outcome Group A Group B
Initial participants 100 100
Injury-related dropout in the example 20 5
Finishers 80 95
Fast finishes 60 50
Fast runners among finishers 75% 52.6%
A woman runner reviews her notebook beside a café window, looking for patterns in her running diary.Enlarge image
Editorial illustration created with an AI tool, showing a review of a running diary; not a documentary photograph.

With app data, device use is an entry condition too

The marathon analysis based on Runtastic data retained fewer than half its initially identified finishers after data-quality filtering: 6,771 from 14,773. The authors explicitly describe and justify this. The numbers are not a concealed error, but important methodological information about the data underlying the conclusion.

A person using the app, recording runs and uploading heart rate data is not automatically identical to the entire running population. Access to devices, recording habits and what someone regards as a workout worth uploading may differ. These are possible differences, not characteristics proved by the study for every participant. A precise conclusion names whose diaries produced the finding.

The large database retains its advantage: many repeated sessions and patterns recorded in real environments can be analysed. There is no need to choose between the extremes of all data being perfect and nothing being useful. The question is which part of the sample you can discuss with justification. Knowing the selection conditions makes app data’s value more precisely usable.

Questionnaire research and performance analysis can have different samples

The questionnaire study following half-marathon and marathon preparation also collected information on injuries, weekly volume and the longest run. An important distinction is that studying finishing time naturally requires a completed race result, while other outcomes can use a different set of participants. It is incorrect to assume every analysis uses one unchanged sample.

The study’s tables or methodological notes reveal who entered each comparison. Total enrolment, questionnaire respondents and people with results may therefore differ. This is not contradictory in itself. What matters is knowing which number belongs to which question, without an article’s summary using one analysis’s denominator to present another’s result.

For your own question, therefore, read more than the abstract’s final sentence. Check the follow-up period, definition and participants used to examine an injury claim. An association with finishing time and occurrence of injury are two outcomes. An association found or not found in one does not automatically supply missing information in the other.

Sample size does not repair missing data

If a run never entered the app, the database does not contain the same situation as a run known not to have occurred. Using several devices or recording another activity separately can also change how a weekly total is interpreted. Such questions remain in a large sample. More rows do not alone perfect each row’s content.

Self-report has different strengths and limits. It may provide information absent from a watch file, such as why a session did not happen. At the same time, weekly volume or earlier symptoms recalled from memory may be inaccurate. There is no need to simplify the two sources into a slogan ranking them. Ask instead which fact was recorded in which way and whether that fits the conclusion.

The same discipline helps when reading your own diary. Separate a definite missed session from an unresolved upload, and the plan from what was actually completed. This does not make you a researcher, but helps reveal why an apparently overly precise weekly average may mislead. Caution with data here is not general distrust, but preservation of its specific meaning.

Faster runners’ patterns are not assigned training programmes

The research examining preparation among participants in the 2022 Boston Marathon linked training characteristics and their changes to official results. Analysis required complete responses and a finished race. The finding therefore concerns people studied under those inclusion criteria, rather than every possible path through marathon preparation.

More training observed in a faster group does not automatically establish that adding the same amount causes the same improvement for anyone. Years of previous training, present capability and other circumstances may differ together with training volume. Statistical adjustment can address certain differences, but does not retrospectively turn observation into a trial of randomly allocated programmes.

The practical interpretation should therefore be a set of questions. Which training histories did the people studied have, how was volume defined and how similar is my own situation? If several answers differ, there is still scope to learn from research, but interpretation is needed rather than copying. A group average can support a discussion without replacing planning of your own preparation.

A runner adds a separate note about an uncertain measurement while reviewing training diary data.Enlarge image
Editorial illustration created with an AI tool, showing a traceable note for suspicious sensor data; not a documentary photograph.

Five questions for a useful research summary

First identify the group studied: who was eligible and who supplied data? Second, examine attrition: how many people did not enter the final analysis and why? Third, clarify the outcome: finishing time, completion, injury, a physiological measure or a variable calculated from an app? These three steps prevent many misunderstandings even before detailed examination of the numbers.

The fourth question concerns study design. Were existing training patterns observed, or were programmes assigned? The fifth is your own purpose as a reader: what do you want to use the finding for? Describing a phenomenon, making a prediction and promising an intervention’s personal effect require different evidence. A good summary does not move unnoticed from one question to another.

You can use these questions even without reading every paper in full. Simply seeking this information in a popular article helps; if it is missing, do not automatically supply the strongest possible interpretation. Not every uncertainty can be resolved quickly. Named uncertainty nevertheless offers a more precise starting point than a certain story constructed from missing details.

How should you phrase a result precisely when passing it on?

A sound sentence includes the group, measured characteristic and nature of the association. For example, among finishers using the app studied and with suitable diaries, greater volume was associated with a faster result. This is longer than saying more kilometres guarantee greater speed, but does not change the original evidence. Precision starts at the beginning of the sentence, not in a hidden footnote.

You do not need to invent the stories of excluded people either. If you do not know why they are missing, you can say so. Not every dropout should be called an injury, just as not every missing file means laziness. Naming a methodological boundary does not authorise manufacturing new causes. The aim is to keep the available data’s meaning clear and let absence remain visibly absent.

Research on finishers can therefore teach much about the marathon when we know whose stories it contains. Entry doors, filters, missing data and study design are not dull appendices: they determine the answer’s limits. From this perspective, a scientific paper becomes an information source for more precise questions connected to your situation, rather than a training plan to copy blindly.

Sources

Helpful? Pass it on.