When survey and interview results differ. How to write the discussion in a sociology master's thesis

Most people in the survey rated signing in to the municipal portal as easy. In interviews, some residents described getting help with preparing an attachment or submitting the application. Do these findings contradict each other? First check which findings concern the same action, people and period. An easy sign-in can be followed by a difficult file upload. Suppose, however, that a person said in the survey that they had sent a particular application without help, then told you in an interview that someone else filled in and submitted the form for them. You now have two different accounts of the same event. Each comparison calls for a different sentence in your discussion.

If you are studying Sociology in Poland and have finished collecting data, start with the pair of findings that is hardest to bring together. Retrieve the exact wording of the questionnaire item and the relevant interview question. Record how participants were recruited, when and under what conditions the data were collected, and where each finding comes from. Then decide whether one finding complements the other, whether the mismatch comes from how you have compared them, or whether two comparable accounts remain in tension. The worked example below takes you from this first comparison to four discussion paragraphs.

Start with a claim you can trace in both sets of data

“The survey shows that the portal works and the interviews show that it fails” is too broad. Turn it into two specific statements about what someone did or reported. “Most people who answered the question said that signing in was easy” can be checked against the response file. “Some interviewees described getting help with an attachment” can be checked against the recordings or transcripts. With the findings written out, you can see that they address different stages of using the portal. In the discussion you can show how both findings inform a question about access to the service.

How do you identify a common unit of comparison?

Take the research question from your methods chapter and mark who it concerns, the action or experience you are interested in, and the time and place in which it occurs. “How do residents in the municipality use digital services to handle administrative matters?” gives you a starting point with several possible units of comparison. You could compare stages of one procedure, residents with different conditions of access, or one person's accounts of a single application. Choose a level that fits the evidence you collected. If a survey respondent submitted two applications and discusses only the second one in the interview, matching the participant alone does not identify the same event.

Check whether you are comparing an individual answer, a group result, or a theme developed from several interviews. A percentage describes the distribution of answers to a particular item with a specified denominator. An interview theme can bring together experiences in which help took different forms but served a similar purpose. “Help from family and friends in completing the application” could cover reading instructions together, lending a scanner or sending a file on someone else's behalf. Check which actions fit the definition of help given in the questionnaire. This tells you which cases can be compared directly.

If you need to revisit how the methods were chosen or how to analyse each separately, see the guide to research questions, interviews and surveys in sociology. At the discussion stage, use the methods you actually applied to set the boundaries of your comparison. If a question emerged when you saw the results, record it as an interpretive question or a direction for additional analysis so the reader can distinguish the original aim from an issue discovered while working with the data.

Where does reporting results end and discussion begin?

The results show what you observed. Readers should be able to find the number of responses, how an indicator was calculated and what supports a qualitative code. In the discussion, explain what the findings together contribute to the research question. This is where you assess agreement, connect the findings with theory, consider alternative readings and explain limitations that change what you can conclude. If your department allows a combined results and discussion chapter, you can keep the distinction clear within paragraphs by giving the evidence first and explaining its meaning next.

For an example from a Polish university, the Faculty of Sociology at the University of Białystok sets out requirements for its second-cycle sociology programme. For a thesis based on a research project, the faculty expects students to interpret their own findings independently in the light of existing research and to formulate their own arguments and conclusions. This guidance applies to that faculty. Check your own department's rules and the guidance for your thesis seminar, particularly on chapter structure, references to your own data and the presentation of interview excerpts.

Tell apart complementary findings, apparent contradictions and unresolved discordance

Fetters, Curry and Creswell (2013) describe how findings from different parts of a study can fit together as confirmation, expansion or discordance. In their discussion of triangulation in the social sciences, Hornowska and colleagues (2012) point out that different methods can deepen our understanding of a phenomenon even when they yield different results. You can apply that logic to each question or set of findings separately. Check comparability first. When groups, definitions or periods differ, you may be unable to say whether two methods gave different answers to the same question. This check helps you distinguish a deliberate look at different aspects of a phenomenon from an unsupported assumption that both datasets capture the same thing.

Confirmation and complementary findings

Confirmation concerns a specific finding. If survey respondents report regularly getting help from people close to them during submission, and interviewees from a comparable group describe actually receiving help at that stage, both sources point to the presence of help. Each provides evidence collected in a different way. At the individual level, two consistent accounts by the same person can support the coherence of the description. Both are self-reports, so their agreement alone provides no independent verification of what happened.

Findings are complementary when they let you describe different, connected parts of a process. The survey asks how easy it was to sign in. The interview takes participants from opening the portal to receiving a confirmation. The interviews show that a difficulty could occur after sign-in. Bringing the evidence together helps you describe the stages of access, explain the role of help and identify the stage that the survey percentage refers to. A general label such as “easy to use” would otherwise cover steps that the item did not measure.

Look for cases where the additional mechanism is absent. If some interviewees sent the documents independently because they had experience with similar forms, include those accounts alongside cases involving help. You can then say that support enabled some people to get through a particular step, while others completed it independently. Keeping that variation visible gives you grounds for further questions about resources and experience.

Apparent contradictions created by the comparison

Your spreadsheet shows 70% positive answers. You also find complaints throughout the transcripts and are tempted to say that respondents changed their minds. Check recruitment first. If the survey went to active portal users and the interviewees were recruited at a service desk where people sought assistance, both sources may have different participant profiles even if they refer to a population by the same name. Describe how each group was recruited. If you can identify people who took part in both and the study and consent arrangements allow you to link the data, check the comparable cases. If one source lacks an identifier that can link the accounts, compare at the level of the group or theme.

Timing can also create an apparent contradiction. “In the last 30 days” refers to a particular period in each participant's life. The interview question “When did you first submit an application?” may open an account of something that happened years earlier. If the way people sign in changed in the meantime, you can present the newer result and earlier account as evidence from different periods. Without a relevant comparison, you cannot attribute the difference to the change alone. Give the date or period, say what you know about the change and identify the explanations that your study cannot distinguish.

Look at wording and analysis procedures as well. People may interpret “Did you use the portal independently?” as asking whether they signed in on their own. An interview code may count help with any step. A rating of “fairly easy” may reflect someone's assessment after they got help. Silence in an interview may mean the question was never asked. Record the exact feature that limits each comparison. If you discover a calculation error or a missed coding rule, document the correction in the source analysis and update the affected findings.

Discordance that remains once comparability is checked

Discordance describes findings that concern the same phenomenon within the same scope yet still differ in content. Imagine that a resident reported preparing and submitting a particular application without help in April. In an interview about the same application and period, the resident said their daughter prepared a required file and sent the form on their behalf. Once you have checked consent, identified the event and compared definitions of help, you can describe a specific conflict between the accounts. That comparison alone cannot establish which account better represents what happened.

Discordance can also occur between findings for comparable groups. Consider the strength of the quantitative result and the evidence for the qualitative theme. Very small subgroups, substantial missing data or an imprecise estimate limit the quantitative evidence. One striking interview excerpt can document a possible exception. A claim about a pattern across several cases needs their context and evidence from any accounts that point in a different direction.

Report what your checks show even if you cannot identify a single cause. You can have evidence of disagreement and only hypotheses about its origins. The interview may have prompted people to reconstruct individual steps. Someone saying “I did it myself” in everyday conversation may mean that they made the decision independently even when they received technical assistance. Explain what evidence supports each possibility and what would have been needed to distinguish them. Different accounts alone give you little basis for attributing deliberate concealment to a participant.

Make brief comparison notes

O’Cathain, Murphy and Nicholl (2010) discuss comparing findings from both parts of a study around shared themes and following a finding from one dataset into another. You can apply a similar approach using brief written notes. Start with a finding that matters for your main question. In the other source, look for evidence that supports it, specifies the conditions under which it applies, or calls it into question. Leave room to record that no comparison is possible. The note helps you reason through the comparison. In the thesis you mainly need the decision and its strongest justification.

What should you record when comparing a pair of findings?

• Comparison question. Keep it narrow, such as “Did another person carry out any of the work required to submit this application?”
• Survey finding. Record the item, response options, numerator and denominator. Add the reference period, eligibility filter and how you treated missing values. For a subgroup result, say how you defined it.
• Interview finding. Record the question or point in the interview, the code, the passage identifier and a brief account of what was said. Note whether the topic came up spontaneously or after a prompt. Record relevant passages that depart from your initial interpretation.
• Shared conditions. State what can be compared directly, such as the exact period, action and episode. Separate this from broad similarities such as using the same portal for different tasks.
• Decision and limits. Record how the findings relate, what the evidence supports together and what remains beyond its reach. Add a check, such as recalculating the numerator or reading a longer stretch of the transcript.

Make notes for findings that agree as well. If you cover only the discrepancies, a reader may think the entire project produced incompatible answers. If you cover only agreements, some of your hardest evidence goes unexplained. Choose notes that collectively cover the evidence needed to answer the research question. Add secondary findings where they clarify a limitation of the main conclusion or draw attention to an important exception.

How do you check the comparison within the original study?

Return to the documentation and metadata that you already have. For the survey, check the questionnaire version, filters, multiple-response items, the group included in the calculation and the number of blanks. If two denominators answer different questions, you can present both and explain them. The share of independent submissions among people who tried the procedure concerns the transition from attempt to outcome. The share among those who completed it concerns how a successful submission came about. Each denominator works within its own scope.

Return to the surrounding interview passage. Check whether a sentence about help refers to the same application or, for instance, to an introductory computing lesson years earlier. In your note, distinguish what the person said from your reading of its meaning. If the account is ambiguous, record the possible readings. When you revise a code definition, review similar passages in all relevant interviews so the same rule also applies to material that challenges your initial claim.

Test whether your conclusion is sensitive to one justified decision. You might limit the comparison to people describing events from the same period and then see whether including other accounts changes the conclusion itself or only the scope of the material. For the survey, report a result based on valid answers and show how many were missing. If people who encountered a particular problem may have been more likely to skip an item, avoid guessing their answers. Explain how that possibility limits your interpretation. Clearly label additional analyses prompted by inspecting the data as exploratory.

When can you link survey and interview records for the same person?

Use individual linkage when you have a reliable identifier, permission under the study and consent arrangements, and accounts covering the same unit. Similar ages, neighbourhoods or experiences do not establish that the records refer to the same person. With an anonymous questionnaire you can still compare themes or groups. Explain who contributed to each dataset and what that comparison can tell you about the phenomenon.

When you link cases, check how easily an interviewee might be identified. A study code helps organise your work, but a distinctive combination of roles, place, age and an incident can identify someone whose name has been removed. Include only the details necessary for the interpretation and allowed by the conditions of participation. Quote what was actually said when presenting a direct quotation. Label a summary as a paraphrase. When you put an excerpt next to a numerical result, state whether it illustrates one case or is an example of a pattern supported elsewhere in your analysis.

A worked example. Online applications in a fictional municipality

The municipality, numbers, accounts, dates and identifiers are fictional. Natalia studied how residents handled the stages of applying online for a fictional certificate and when they received help. She focused on April. The questionnaire was advertised through a municipal mailing list and notices at the library. Participation was voluntary. She selected interviewees who had used the service in different ways, including people who had not completed the procedure. Reliable identifiers and consent to link the data were available for nine cases.

What do the questionnaire and interviews show on their own?

Natalia received 132 questionnaires that met the analysis criteria. For the item “How easy was it for you to sign in to the municipal portal in April?”, 120 respondents gave a valid rating and 12 did not give one. Of those who gave a rating, 84, or 70%, selected “very easy” or “fairly easy”. The result concerns signing in to the portal. Some people may have signed in to handle matters other than the application studied.

Of the 132 respondents, 100 reported attempting the target application in April. In a separate closed question Natalia distinguished submitting independently, submitting with help and having made no submission by the end of April. The instructions counted another person doing part of the required work, such as preparing an attachment or operating the form, as help. Eight people skipped the item. Of the 92 valid responses, 58 reported submitting without such help, 25 submitting with help and 9 having made no submission. Thus 58/92, or about 63%, is the share of independent submissions among attempts with a reported outcome. If you consider only those who eventually submitted, the denominator is 83 and the share is about 70%. The resemblance between that percentage and the sign-in rating is coincidental. They come from different items and denominators.

In 12 interviews Natalia asked about a particular attempt in April. She prompted interviewees to reconstruct the steps and say who did what. They described previous experience using forms independently, help preparing files, reading instructions together and stopping before submission. If someone mentioned a separate procedure or an event before April, Natalia recorded it as context. She used the April episode in the direct comparison where it matched the questionnaire's scope.

What changes when the action and people are matched?

The first comparison concerns stages. Residents could find it easy to sign in and arrange help afterwards. In three interviews, participants specifically said that they had signed in independently but a friend or family member prepared the attachment. Natalia kept these passages as examples of a transition from reaching the service to a step that required support. Each passage describes one attempt. Together they identify a role for help that the sign-in item did not capture. No survey percentage measured how often this exact sequence occurred.

The second comparison concerns independent submission. Of the nine linked cases, six reported submitting without help in the survey, two reported submitting with help and one reported having made no submission. Four of the first six described completing all the required steps without another person's involvement. Two described help that fell within the questionnaire's instructions. In one case someone else prepared the required file. In the other someone else filled in part of the form and sent it. Natalia checked that both interviews referred to the same April application and applied the definition from the questionnaire when coding those passages.

One of the two respondents who reported submitting with help confirmed the situation in the interview. The other interview contained too little information about another person's involvement. The ninth case contained consistent accounts of no submission by the end of the month. In all, six pairs of accounts agreed, two pairs were discordant and one pair contained too little information to assess agreement. Natalia had no linked questionnaire for the remaining three interviews. She included them in her thematic analysis without assigning them individual survey answers.

What remains uncertain?

Two discrepancies remain after checking the action and period. In one interview, the resident explained that they felt they had handled the matter independently because they had chosen the procedure and made the decision themselves, although someone else had prepared a file for them. The second account described giving part of the work to a person close to them. It contained no explanation of what “independently” meant to the participant. Distinguishing responsibility for the matter from technical execution helps interpret the first case. In the second, that explanation remains open alongside differences in recall or how the interview question was asked.

Recruitment limits the conclusions about residents as a population. People who subscribed to the mailing list or visited the library may have had different contact with the service from other residents. Interviewees were deliberately selected to cover different experiences, and the linked cases formed a subset of the material. The fraction 2/9 describes two discrepancies among nine cases examined in pairs. Natalia kept 58/92 as the survey result under the original analysis rule. The study had no independent verification of how the application was handled and no equivalent data for every survey response. She added that the comparative analysis found two discordant accounts of independent submission.

The study also lacks a complete estimate of how often people received help after signing in independently. The questionnaire did not follow every respondent through the same sequence of events and the two results use different denominators. The interviews show how the sequence could unfold and where another person's help could enter it. A defensible conclusion describes several documented paths through the procedure and points to cases where the sources captured the same event differently.

Turn the comparison into discussion paragraphs

Open with the question you are addressing or with the answer you will support. Bring in the most relevant survey result and interview finding. Describe the relationship between them before drawing on theory or previous research. If you give a possible explanation, show what supports it. Finish with a statement that fits the level of certainty, such as specifying the group examined or identifying what neither source can establish.

Four paragraphs for the fictional study

The example shows how Natalia could bring her findings together. Use the data, references and concepts from your own study when drafting your discussion.

Participants' ratings of access concerned one stage of using an online service. In the survey, 84 of the 120 people who rated the sign-in process found it very or fairly easy. The interviews documented cases where participants signed in themselves and then needed help preparing a required file. These findings complement each other because they concern successive steps in using the portal. For the research question about access, the survey identifies the reported ease of signing in. The interviews identify additional demands after that step.

A narrower comparison concerned who carried out the work needed to send the same application. Of the 92 valid responses from people who attempted the procedure, 58 indicated independent submission. Of the nine linked pairs, six agreed, two differed and the interview in the final pair contained insufficient information about another person's involvement. In both discrepant cases, a report of independent submission appeared alongside an account of someone else doing part of the required work. Comparison identifies discordance between the two accounts. The available data do not independently verify the steps taken.

In one case, the interview suggests that a sense of independence could reflect making the decision and dealing with one's own affairs with technical assistance from another person. This reading draws on how the resident described their own role in the decision and another person's role in preparing the file. What produced the other discrepancy remains unclear. The distinction between stages can be considered in light of Hargittai's (2002) distinction between internet use and the skills needed to carry out online tasks. Her study investigated information search tasks. It provides a conceptual reference here, while the roles of help in the application process come from the local accounts.

Taken together, the data support a description of digital access that includes reaching the portal, completing successive steps and the role of assistance. Given voluntary survey participation and purposive interview recruitment, the study cannot estimate how frequently residents across the municipality give discordant accounts of independence. The quantitative result also depends on who answered the item about the outcome of their attempt. Future research could ask about each action separately and record who performed it. That would allow researchers to examine how distinguishing decision-making from technical execution affects what participants report.

How do you connect the findings to theory and existing research?

Choose a paper for what it contributes to the specific interpretation. Hargittai (2002) moves from asking whether a person uses the internet to examining differences in their ability to carry out online tasks. This helps explain why reaching the portal captures only one part of access. Her study did not examine the fictional municipal portal, the local application or the linked accounts. For a real study of public services, look for research on similar administrative tasks and groups to strengthen the empirical comparison. Check whether each citation supplies a definition, a possible mechanism or a genuinely comparable finding.

Malewski (2023) examines debates about the grounds for combining research strategies and quantitative and qualitative data in the social sciences. In a master's thesis, you can address that issue by justifying how you draw a particular inference. If a questionnaire organises reports by stage, while interviews reveal how people understand independence, explain the level at which these contributions come together. A methods reference can explain the analytic choice. Use your own evidence and relevant studies of the phenomenon to justify your interpretation of what residents reported.

Make clear when the comparison happened and how it relates to your study's purpose. If the strands were planned separately and comparison arose during writing, describe the timing and scope of that decision. Fetters and colleagues (2013) include integration at the level of interpretation and reporting. Their paper can help justify why you brought together the findings in a particular way, where your design permits it. The reader should be able to distinguish the original recruitment and data collection plan from subsequent comparison, especially if only some participants contributed to both strands.

Match the strength of your conclusion to the evidence

Keep three types of statement distinguishable. “Two people gave differing accounts of the same episode” describes a finding from the comparison. “For one person the disagreement may reflect a distinction between making the decision and receiving technical assistance” offers an interpretation supported by that interview. “Distinguishing these concepts could improve measurement” proposes an idea for a subsequent study. Using “the study demonstrated” in all three cases would obscure what was checked and what remains open.

How should you handle awkward or incomplete findings?

• A direct contradiction in one case. Keep a record of both findings, confirm the event and explain how it limits the conclusion. One case can challenge a claim that everyone completed the application independently. A statement about how often it happens needs separate evidence.
• An item without a counterpart. Record why there is no comparison. A finding from one source can still answer part of your question. If interviewees were never asked about their feelings, silence tells you little about what they experienced.
• An imprecise quantitative result. If you report an estimate and an interval, interpret them alongside sample selection, group sizes and missing values. In descriptive analysis, show the denominator. Check that the precision and scope of a number suit the claim it supports.
• A theme concentrated in a particular group. Check who was asked the relevant question and how recruitment may have made a particular account more likely to appear. Explain the conditions in which the pattern appears in your data and leave its frequency open if the study did not measure it.
• An explanation that arose during analysis. Say what suggests it and what would help you distinguish it from the alternatives. If you agreed hypotheses with your supervisor earlier, keep their evaluation separate from later interpretation and state when a new explanation emerged.

Be precise about causality. “Participants described needing help when attaching documents” carries its source within it. “Help increased the probability of submission” requires evidence that can distinguish the effect of help from such differences as the difficulty of the application, earlier experience and the time available. People who seek assistance could also have more difficult cases. Bringing together numerical results and accounts gives you the context in which support occurred and suggests questions about its role.

How do you check the chapter before sending it to your supervisor?

Choose a key paragraph and trace every claim back to its source. For a number, check the denominator, the treatment of missing answers and the measurement period. For an interview account, check whether the quotation or paraphrase has sufficient context. When you call findings contradictory, review the conditions you matched. For each possible explanation, give the evidence that suggests it. For each citation, check whether the authors examined a comparable phenomenon or whether you are using their work as a conceptual aid.

Then read from your notes to the text. If a note covers a single case but your discussion says “residents”, make the subject more specific. If the paragraph opens with successful submission and then cites a sign-in result, explain the connection or use a finding about the relevant stage. If your account of linkage suggests all cases were paired, state how many comparisons you actually made. Keep those same limits in the general conclusion.

For a meeting with your supervisor, bring one research question, two comparison notes, a passage about limitations and a paragraph linking your findings to previous research. Identify whether you need a decision about the group, help interpreting a particular account or advice on where the discussion belongs in the thesis. In the example, Natalia could ask whether her justification for comparing nine cases is clear and whether readers can distinguish two incompatible accounts from the finding that different stages had different demands. This gives the supervisor a basis for discussing the analysis and the way it is written.

Sources

• Fetters, M. D., Curry, L. A., Creswell, J. W. (2013). Achieving integration in mixed methods designs: principles and practices. Health Services Research, 48(6 Pt 2), 2134-2156. DOI 10.1111/1475-6773.12117.
• Hargittai, E. (2002). Second-level digital divide: differences in people's online skills. First Monday, 7(4). DOI 10.5210/fm.v7i4.942.
• Hornowska, E., Brzezińska, A. I., Kaliszewska-Czeremska, K., Appelt, K., Rawecka, J., Bujacz, A. (2012). Paradoksalny efekt triangulacji? Edukacja, 4(120), 72-83.
• Malewski, M. (2023). Badania mieszane w naukach społecznych – o konstruowaniu metodologicznej prawomocności. Edukacja, 2(165), 17-30. DOI 10.24131/3724.230202.
• O’Cathain, A., Murphy, E., Nicholl, J. (2010). Three techniques for integrating data in mixed methods studies. BMJ, 341, c4587. DOI 10.1136/bmj.c4587.
• University of Białystok, Faculty of Sociology. Zasady dotyczące zakończenia studiów II stopnia na kierunku Socjologia w Wydziale Socjologii Uniwersytetu w Białymstoku. Section “Praca dyplomowa (magisterska)”, point 11. Guidance specific to that faculty.