When interpreting survey results, a conclusion connects a research question to what the responses actually establish. State the finding, identify its numerical basis, and define whom it concerns and what was measured. A percentage supplies evidence; it does not answer every thesis question on its own.
The worked example follows one dataset: 80 students rating instructions for booking consultations. Using the same data, we move from single responses and a rating item through missing answers and multiple responses to a complete conclusions paragraph and proportionate recommendations. Every participant, answer and number is fictional. This is a teaching example, not a client project or an actual university study.
Results, interpretation, conclusions and discussion: give each a job
A result
Reports an observation: 48 of 80 people selected an agreement response. Identify the question, category and respondents involved.
Interpretation and a conclusion
Explain what the result means for the research question: positive ratings of the instructions’ clarity predominated in this group. That remains a self-report, not a test of booking performance.
Discussion
Relates the findings to other research, possible explanations and limitations. Attaching an author’s name to a similar percentage is insufficient: compare questions, samples and measurement.
The whole-thesis conclusion
Closes the thesis aim and brings together its main findings. It may contain conclusions, but is not automatically a copy of the last results subsection. Follow the structure required for your own programme.
The 2024 Faculty of Health Sciences guidance from the Medical University of Lublin distinguishes results, discussion and conclusions arising from the study. It illustrates distinct functions; it is not a universal chapter plan for every degree. This guide focuses on moving from questionnaire responses to a defensible conclusion.
For comparing your findings with published research, use the separate discussion guide.
If you are closing the whole thesis, check the guide to its final chapter.
The example: three research questions and 80 voluntary responses
The fictional project has a specific aim: find out how students on one programme rate online consultation-booking instructions and what difficulties they report. The link was shared in student groups. Participants were not randomly sampled, and the number who saw the invitation is unknown. There are 80 questionnaires: 40 from bachelor’s students and 40 from master’s students.
- Rating question: how do participants assess the instructions’ clarity? One statement with five agreement categories measures this.
- Difficulty question: what problems did participants encounter when booking? They could select multiple problems or only “I encountered no difficulty”.
- Comparison question: do ratings differ between previous users and first-time users? This requires linked responses from the same individuals.
An additional optional item asked: “Do you need a further explanation of the instructions?” Sixty people answered it. Each dataset row represents one participant; each multiple-response difficulty has its own column. The Poznań statistical office’s teaching guide also illustrates separate columns for multiple responses and an explicit missing-answer category.
Single responses: move from counts to an answer to the question
All 80 participants rated “The booking instructions are clear.” Four strongly disagreed, eight disagreed, 20 selected “neither agree nor disagree”, 32 agreed and 16 strongly agreed. Combining the agreement categories gives 48 / 80 × 100 = 60%. Disagreement accounts for 12 / 80 = 15%; the neutral category accounts for 25%.
Three response groups for one statement
Fictional data. Every bar uses the same 80 responses.
This shows grouped agreement categories, not measured booking performance. The three groups sum to 80 people and 100%.
A clarity rating versus actual booking performance
Before
“The survey proved that the instructions are effective and all students can book without difficulty.”
After
“Positive ratings of the instructions’ clarity predominated among the 80 participants: 48 people (60%) selected one of the two agreement categories. This finding concerns perceived clarity. The project did not measure booking accuracy or completion time.”
Why does the edit matter? The first version turns an opinion into demonstrated behaviour and extends the finding to all students. The second answers the actual rating question. The denominator makes the 60% auditable; naming the measure explains what that percentage represents.
Missing answers: 70% of 60 is not 70% of the whole sample
For the optional request for further explanation, 42 participants answered yes, 18 no, and 20 gave no answer. The yes percentage among item respondents uses 60 as its denominator: 42 / 60 = 70%. Against all 80 questionnaires, the same 42 yes answers are 52.5%, no answers are 22.5%, and missing answers are 25%. These are different descriptions of the same data.
A usable statement that includes missingness
Base: 60 valid responses, 20 missing answers in a group of 80.
“Further explanation was requested by 42 of the 60 people who answered the item (70%). The answer was missing in 20 of 80 questionnaires. The available data do not establish the needs of those who skipped it.”
The wording identifies both counts without inventing an answer for the missing cases.
Establish what a missing entry means: the person did not see a branched question, declined to answer, left a field blank, or selected not applicable. These situations need not mean the same thing. Do not automatically recode them as no, or exclude a participant from every analysis because one variable is missing. Define the base for each specific comparison.
Multiple responses: people, selections and totals above 100%
All 80 participants answered the difficulties item. Forty-eight selected lack of a suitable slot, 32 difficulty identifying the right person, and 24 changing or cancelling a booking. Sixteen selected only no difficulty. Thus 64 people (80%) reported at least one problem. The same individual could report several problems: there are 104 difficulty selections, not 104 respondents.
Two denominators answer two different questions
Percentage of people reporting the problem
Use 80 as the denominator. Lack of a suitable slot is 48 / 80 = 60%, identifying the right person is 40%, and changing or cancelling is 30%. The total is 130% because multiple selections were allowed. That total is not itself an error.
Share of the difficulty selections
Use the 104 difficulty selections as the denominator. Slot availability accounts for 48 / 104 × 100, approximately 46.2% of those selections. That does not mean 46.2% of students experienced it. No-difficulty answers are not counted as difficulty selections.
A conclusion answering the first question is: “The most frequently reported difficulty was the lack of a suitable slot, selected by 48 of 80 participants. This describes reports in this group, rather than the actual number of slots available.” Assessing availability would need other material, such as booking records for a defined period.
For dataset preparation, percentages and reporting in Excel, use the survey-results guide.
A 1-to-5 rating: what does the mean say, and what gets hidden?
The most common response is 4, agree. The median is also 4: the 40th and 41st ordered responses both fall in that category. Coding the responses from 1 to 5 gives an arithmetic mean of 3.60. That is a calculation on the selected codes, not “72% understanding”. The item has no defined zero or measurement unit that would justify that conversion.
For a single ordinal item, show the response distribution and justify the summary you choose. Combining agree and strongly agree gives a concise agreement measure but hides intensity: only 16 people strongly agreed, not all 48. One statement is not automatically a validated psychometric scale.
Group comparisons: percentage points and linked responses
Among 40 bachelor’s students, 26 selected agreement (65%); among 40 master’s students, 22 did so (55%). The difference is 10 percentage points. Do not label subtraction of these percentages “10% more”. Nor should the difference be called statistically significant without an appropriate analysis.
Separate charts do not, by themselves, establish a relationship between previous system use and ratings. Link those variables within the same participant rows. In this fictional dataset, 30 of 40 previous users selected agreement (75%), compared with 18 of 40 first-time users (45%). We are now comparing two defined groups, rather than two unrelated percentages.
An association does not yet answer why
An association versus an effect of previous use
Before
“Using the system improves understanding of the instructions by 30%.”
After
“Previous users in this sample more often rated the instructions positively than first-time users: 75% versus 45%, a difference of 30 percentage points. The comparison is descriptive. A one-off questionnaire cannot establish whether prior use caused the difference in ratings.”
The edit corrects two errors: it uses percentage points and does not turn co-occurrence into an intervention effect. Digital experience, different instruction versions or self-selection might also matter. Marcin Sydow’s PJATK teaching materials explicitly distinguish correlation from causation; this example does not invent a correlation coefficient or p-value.
How should conclusions address a hypothesis, including an unsupported one?
Start with the actual prediction and the procedure chosen to assess it. “Previous users more often give positive clarity ratings” matches the preceding comparison, and the observed direction is consistent with it. If the project requires statistical inference, however, the direction of percentages alone does not complete the assessment: justify the method, its assumptions and the analysis results.
“The instructions reduce booking time” is different. This dataset has no timing measure or comparison with other instructions, so it cannot assess that hypothesis. That differs from obtaining evidence inconsistent with a prediction. Separate support, lack of support and inability to assess. Do not retrospectively alter the hypothesis or data to obtain a convenient confirmation.
If a test was conducted, failure to reject its null hypothesis is not automatically proof that no association of any kind exists. Consider effect size and uncertainty where they were assessed. No such tests were performed in this example, so its conclusions remain descriptive.
Write a limitation that changes how a conclusion is read
Replace “the study had limitations” with a design feature, its possible consequence and the claim it limits. For this example: “Recruitment through groups on one programme and voluntary participation may have attracted people particularly interested in the issue. These findings describe the collected responses; they were not used to estimate opinions across the whole university.”
Statistics Poland explains random selection through its representative household surveys. Our fictional volunteer recruitment is different: equal numbers from two degree levels do not make it a random sample, and collecting more questionnaires does not automatically repair selection bias. The number 80 alone also cannot provide a response rate without a suitably defined invited population.
For a fuller account of connecting limitations to findings, see the separate guide.
A complete survey-conclusions example after editing
Conclusions answering the questions about instructions and booking
A complete conclusions passage
Assumptions: only the fictional group of 80 described above, voluntary participation, and no observation of actual booking performance. This passage connects findings, meaning and limits without adding an intervention effect.
“Positive clarity ratings predominated in the surveyed group: 48 of 80 participants selected agreement (60%). At the same time, 64 people reported at least one booking difficulty. Lack of a suitable slot was reported most often (48 people), concerning perceived availability rather than instruction wording alone.
Previous users rated the instructions positively more often than first-time users: 75% versus 45%. This is an observed association without an established cause. Further explanation was requested by 42 of the 60 people answering that item; the information was unavailable for 20 participants.
The findings identify two distinct issues: assessments of the information and reported availability problems. Findings concern a voluntary questionnaire on one programme and self-reported assessments. Booking performance was not tested, nor were university-wide opinions estimated.”
The passage does not repeat every category from every chart. It selects findings that answer the aim, distinguishes instruction ratings from availability, and makes the interpretation boundary explicit. In an actual thesis, add references to your own tables or figures and keep the variable names used in the methods. Do not transplant these fictional numbers into your results chapter.
Recommendations: an action, its basis and a way to assess it
“The system should be improved” is a weak recommendation. A useful one identifies what to examine and why. Because suitable-slot availability was the most frequently reported difficulty, first compare those reports with records of slots actually offered. If the instructions may be a problem, pilot a revised version using a defined task, such as independently making and changing a booking.
A recommendation clearly separated from the findings
An evidence-based next step
Assumption: a proposal for further evaluation, not the outcome of a completed experiment. The action follows the reported type of difficulty rather than the appeal of the idea itself.
“First examine whether reported difficulty finding a suitable slot corresponds to actual availability. Separately, compare the original and clarified instructions in an appropriately planned study, observing first-time users making and changing bookings. Assessing improvement requires a baseline and consideration of differences between the compared conditions. Observing only the new version would describe performance, not establish the effect of the change; the present questionnaire does not establish it either.”
The recommendation specifies a sequence and an evaluation. It does not promise that wording changes will fix slot availability or declare an untested solution effective.
A follow-up could measure successful completion, time and error types. Those indicators have yet to be collected. Present the recommendation as a proposed evaluation, not a demonstrated effect. If the thesis only aims to describe opinions, agree with your supervisor whether an extended practical-recommendations section is needed at all.
Organise the chapter around questions, so conclusions follow the evidence
Question and evidence
Start with a research question and identify the evidence needed. Decide whether it asks for a description, a comparison or an association.
Finding and basis
Present the key finding with its denominator, unit and missingness. Refer to your own display rather than narrating every cell.
Meaning and boundary
Explain what has been established for the question and identify a limitation affecting that conclusion. Separate the answer from a possible explanation, then move to the next research question.
Answering the thesis aim
Finally bring the answers together around the thesis aim. Put literature discussion and any recommendations where the agreed structure requires them.
Final checklist before your survey conclusions are reviewed
- Each important claim can be traced to a question, data and an analysis actually conducted.
- Percentages have explicit denominators; people and selections are not treated as interchangeable.
- Missing, neutral and not-applicable responses are distinguished as required by the instrument.
- An association has not become a cause, and percentage differences use the correct unit.
- The sample, measure and period agree with the methods; no test or significance claim has been invented.
- Recommendations identify their evidence and a possible evaluation rather than promising the change will work.
Polish primary sources and how they were used
Sources were checked on 25 October 2026. The Poznań statistical office guide informs dataset organisation, multiple responses and explicit missingness. PJATK teaching materials support distinctions involving missing data, data limitations, correlation and causation. The Lublin document illustrates one faculty’s thesis guidance. Statistics Poland explains random selection in its own surveys. None of these sources contains or validates the fictional dataset used here.
- Magdalena Bryza, Wanda Nowara, Arleta Olbrot, Artur Owczarkowski; Urząd Statystyczny w Poznaniu (2024). Jak przygotować i przeprowadzić badanie ankietowe?. Dataset organisation, missing answers and multiple responses. Printed pp. 6-8.
- Uniwersytet Medyczny w Lublinie, Wydział Nauk o Zdrowiu (2024). Wytyczne do przygotowania pracy dyplomowej. A faculty-specific example separating results, discussion and conclusions. P. 1.
- Marcin Sydow, PJATK (n.d.). Eksploracja Danych: Dane. Missingness, data quality and the distinction between correlation and causation. PDF pp. 10-11, 32.
- Główny Urząd Statystyczny (n.d.). Badania ankietowe: Informacje ogólne. Context for random selection in Statistics Poland surveys.
- AAPOR (n.d.). Best Practices for Survey Research. Supplement: survey planning and reporting.
- AAPOR (n.d.). Disclosure Standards. Supplement: disclosure of recruitment and question wording.
- Office for National Statistics (n.d.). Uncertainty and how we measure it for our surveys. Supplement: the context of survey uncertainty.
Questions about survey interpretation and conclusions
What if the results do not support the hypothesis?
I describe this in line with the analysis. A lack of support for the hypothesis is a research finding, not a reason to change the respondents' answers.
It also does not automatically mean the aim was not achieved: the aim may have been to check a specific relationship. I explain what follows from the data and what limitations must be considered. A lack of support for the hypothesis does not by itself prove that the relationship definitely does not exist.
Does every survey or thesis require statistical tests?
No. The scope of the analysis depends on the aim of the work and the research questions. Sometimes counts, percentages, and descriptive statistics are enough. If the study concerns differences or relationships, I check which methods are appropriate for the data and whether the conditions for their application are met.
A thesis based on interpreting regulations, a literature review, qualitative analysis of interviews, or an individual case study may not require testing statistical hypotheses. In qualitative research, the analysis may focus on the meaning of the statements. In that case, the appropriate support may be methodology, material analysis, or text editing. I do not add calculations just to make the thesis look more scientific or advanced.
How can I connect a limitation to its effect on the conclusions?
Identify a specific feature of the procedure, the finding or claim it affects, and the consequence. Then explain what you actually did to address it and what remains uncertain. A generic list mentioning sample size and time pressure is not enough.
Can I order just the analysis and presentation of survey results?
Yes. The scope may include organising data, tables, charts, frequency and percentage summaries, commentary, overall summaries and hypothesis testing if the research design includes it. I link the analysis to the aim and research questions.
Send the collected responses, the questionnaire, the aim and research questions and the supervisor's requirements for the analysis.
Your results are ready. Let’s build conclusions they support.
Send your research questions, questionnaire, anonymised data or summaries, and draft text. I will review percentage bases, whether the description matches the results, and how the conclusions answer the thesis aim. If further calculations are needed, I will agree their scope before starting, based on the actual material.
