Why do surveys and interviews sometimes tell different stories?

The case presented below is hypothetical and has been created to illustrate the concepts being discussed.

A private multispeciality hospital in Delhi completed its annual staff well-being survey. The leadership felt reassured seeing that the survey was completed and that everything looked good. The dashboard seemed healthy and the results of the survey made them feel at ease. One figure attracted their attention: 84% of the respondents agreed with the statement “I can manage the demands of my role effectively.” 

The findings were presented, but within the following weeks, there was a surge in resignations among the clinical staff. However, the organisation didn’t immediately decide that the survey was wrong. They began a follow-up investigation and conducted semi-structured interviews with clinicians from different roles and specialities across the hospital. 

What they found was very different from the survey outcomes. People described many problems, such as regularly working past scheduled hours, excessive workload, administrative challenges and persistent pressure to take on responsibilities beyond their role. 

Why was there such a discrepancy between the survey and the interviews? Surely one of the methods must have been wrong! There might be several ways to try and understand this discrepancy. The survey questions were perhaps designed poorly or maybe respondents were reluctant to answer negatively in a formal questionnaire. 

At the same time, it’s also possible that the interviews included a disproportionate number of individuals with negative experiences, or the circumstances may have changed between the survey and follow-up interviews. 

Such results can be called discordant findings when quantitative and qualitative findings contradict. But they aren’t always a sign of methodological failure because, when the conflicting findings are interpreted thoughtfully, the data can move the researcher beyond the superficial discrepancies and towards contextual explanations, hypotheses and deeper insights into what may be happening in that particular organisation (Younas et al., 2022). So here in our hospital case, they shouldn’t think about which result should be disregarded. Instead, they should think about how they can understand both results. 

Let’s start by looking at what really happens between reading a survey question and selecting a number. It’s useful to know 3 concepts before getting into that:

1. Generalising: Participants may default to broad and pre-established beliefs about their routine instead of answering about the specific timeframe being asked about (example: over the last month). If this happens repeatedly, it can flatten scores. 
2. Compromise: An attempt by the respondent to average out competing or opposite feelings into a single answer that lies in the middle (example: answering 3 on a 5 point scale).
3. Communicative Responding: Individuals don’t just use numbers to record their internal state; their ratings could also be influenced by a wish to send a message. 

Therefore, numerical responses reflect many aspects such as the context, how the item is interpreted, which information is used as a response base, and even reasoning strategies. All these aspects may make information that matters less visible when the score is later used (Grindheim et al., 2024). The study was on patients in treatment settings, not employees, but I’ve used it as it helps in understanding how response processes can work. 

So in the hospital case, the results might tell us how respondents answered an item under a specific response process instead of automatically meaning that 84% of the clinicians are experiencing their work as sustainable. Here, an interview may be able to uncover some of the reasoning behind answers like these. Before moving to interviews, let’s continue focusing on surveys and see how they could be influencing the voice of the employees. Willadsen and Bonk (2025) found that there were differences between closed-form and open-text responses in a large employee survey. In the open-text fields, employees raised concrete, pragmatic concerns that differed substantially from the more abstract topics in the closed-form questions. So what this is pointing towards is that the structure of the question influences what types of employee concerns become evident. The study looked at open-text survey responses, not interviews. 

It’s still important to ask who produced the 84%. It came from respondents but how well did they represent those who didn’t respond? Rhodes et al. (2025) found that some of the nonrespondent groups had a significantly greater risk of turnover in their study of non-response bias. 

Hence, the people who answered the survey and those who could be carrying the organisation’s risk may not even be the same group. The original survey might not have completely represented that risk, and interviews or other evidence could make that risk more visible. 

Once a discrepancy has appeared, how can interviews contribute? 

Interviews allow researchers to understand how people interpret experiences at work, to ask follow-up questions and to investigate experiences that predetermined variables might not be able to completely describe. So qualitative research may help you understand why or how something happened, while quantitative research may tell you more about what happened (Dunwoodie et al., 2022). In the hospital case, interviews offer the opportunity to ask what manageability means to clinicians and how they interpret that question. 

Now to answer the question: How can the contradictory findings be interpreted? There are 2 broad strategies as suggested by Younas et al. (2022):

1. Comprehensive case and variable analysis: this includes breaking down the data by specific subgroups such as roles or departments and comparing the qualitative and quantitative evidence to check where the discrepancy appears. It can help in finding out whether the discrepancy is isolated to certain sections or present across the organisation. 
2. Sociocultural exploration: Conflicting findings are evaluated within the wider social and cultural context to understand why the findings may differ across the 2 methods. 

This brings more questions for the hospital case: was the discrepancy related to a specific department, nonrespondents, a recent organisational change, the time of data collection or something entirely different? In order to answer these questions, I wouldn’t say that organisations should always combine surveys and interviews, as sometimes findings conflict because samples are different, data quality is poor, circumstances have changed or the results are inconsistent due to factors no one can control. 

However, I will say that discordance is only useful when it is examined, not just because it exists. So why do surveys and interviews sometimes tell different stories? They may capture different people, response processes, or ways of describing a similar experience depending on how the question is asked. It’s also possible that the circumstances in the hospital might have changed between the survey and the interviews. The disagreement doesn’t tell us which explanation is correct, but it gives us direction on what to investigate next. 

So in this hospital, the conclusion isn’t that the survey was wrong or that the interviews were right. It’s that the 84% shouldn’t be treated as the complete picture until the organisation investigates where the discrepancy came from. 

All in all, when a survey and an interview have consistent results, the interpretation may be straightforward. When they don’t, the inclination may be to ask which one is wrong, but the more productive question is less straightforward: what has the discrepancy given us an opportunity to investigate? 

References

Dunwoodie, K., Macaulay, L., & Newman, A. (2022). Qualitative interviewing in the field of work and organisational psychology: Benefits, challenges and guidelines for researchers and reviewers. Applied Psychology, 72(2), 863–889. https://doi.org/10.1111/apps.12414

Grindheim, Ø., McAleavey, A., Iversen, V., Moltu, C., Tømmervik, K., Govasmark, H., & Brattland, H. (2024). Response processes for patients providing quantitative self-report data: a qualitative study. Quality of Life Research, 33(11), 2949–2961. https://doi.org/10.1007/s11136-024-03749-2

Rhodes, C. A., Hu, X., Freeman, R. B., Agrawal, R., Cherot, E., Dardarian, T. S., Rouse, S., Chan, T., & Blackburn, B. (2025). Who’s not talking? Nonresponse bias in healthcare Employee Well-Being surveys. Journal of Healthcare Management, 70(5), 337–353. https://doi.org/10.1097/jhm-d-24-00166

Willadsen, H., & Bonk, A. J. (2025). Inferring employee concerns from open text survey responses. Journal of Business Analytics, 1–17. https://doi.org/10.1080/2573234x.2025.2608265

Younas, A., Pedersen, M., & Inayat, S. (2022). Practical strategies to identify and address discordant findings in mixed methods research. Research Methods in Medicine & Health Sciences, 4(1), 24–33. https://doi.org/10.1177/26320843221122342