Interviewing For Social Scientists

Getting Qualitative Data That Actually Means Something

I spent five years doing fieldwork before I stopped wasting time on interviews that went nowhere. The short version: most social scientists approach interviewing wrong from the start. They treat it like a data extraction exercise. It's not. It's a controlled social interaction where the other person is constantly deciding whether to help you or hedge. Your job is to figure out which one they're doing in real time. Let me be clear about what Interviewing For Social Scientists actually requires. You need to prepare thoroughly, listen more than you speak, and be willing to abandon your interview guide when the participant is showing you something important. The people who get good data are the ones who can pivot without panicking.

Practical Framework for Interviewing For Social Scientists

Here's the process I use, and it's not fancy but it works consistently. First, you draft a semi-structured interview guide with 6 to 10 core questions. Not thirty. Six to ten. Beginners always overthink this and produce guides that take two hours to get through, which means you're rushing the last third of the interview and missing your best material. The guide needs three layers. Open-ended launchers that get the participant talking without leading them. Probing follow-ups that dig into specific claims or contradictions. And neutral check-ins that verify you understood correctly. Something as simple as "So what I'm hearing is... is that right?" This isn't filler. This is your quality control mechanism, and it catches misinterpretations before they become permanent errors in your data. Recording setup matters more than people admit. I use a digital recorder as my primary device and my phone as backup. One time, during a six-hour ethnographic interview in a rural community center, the primary recorder failed at minute two hundred and twelve. Had the phone been running. Without it, that entire session was gone. I don't skip backups anymore.

What Actually Happens During a Good Interview

When it goes well, the participant starts off giving you polished answers. They've thought about these questions before, or they've rehearsed what they think you want to hear. This is normal. Don't mistake it for failure. The shift happens somewhere between twenty and forty minutes in, usually after you've asked a follow-up that catches them off guard because it's genuine rather than procedural. The key moment is when you notice their language change. They stop using institutional phrasing and start using their own words. You'll hear contractions return. Sentences get messier. This is where the data lives. Most interviewers miss this window because they're too focused on getting through their question list. I had a researcher once who was interviewing community health workers about vaccine hesitancy. Her guide had questions about misinformation exposure and trust in institutions. The participant kept giving textbook answers for forty minutes. Then she asked, off the record almost, "What's the part you don't say out loud?" The room changed. The health worker started describing exactly how they navigated contradictory messages between their training and what their own families believed. That single conversation produced more analytically useful material than the previous three sessions combined.

Technical Details That Separate Professionals From Amateurs

Transcription is where most projects stall. Don't try to do it yourself unless you have to. Professional transcription services run about one to two dollars per minute of audio. A typical 60-minute interview costs between 60 and 120 dollars. You will save three to four hours of your own time per interview by outsourcing this, and your attention stays on analysis rather than wrestling with ambiguous speech patterns. If you're working with a tight budget, use Otter.ai or similar tools for initial transcription, then spend 30 to 45 minutes cleaning up the output yourself. The AI gets about 85 to 90 percent right on clean recordings. The gap between 90 percent and 100 percent accuracy is where your ear matters most, especially with accent variations or technical terminology specific to your research population. Sampling strategy determines whether your interview data generalizes or just describes. Purposive sampling picks participants who can illuminate your research question directly. Maximum variation sampling deliberately seeks diverse perspectives across a population. Snowball sampling relies on participant referrals, which is useful for hard-to-reach populations but introduces network bias that most people ignore at their peril.

I worked on a project studying informal economic networks in an urban area. Snowball sampling got us access quickly, but the referrals kept producing similar respondents from the same social circles. We were essentially interviewing the same community through different names. Switching to purposive sampling with strict inclusion criteria expanded our range considerably, though it slowed recruitment by about three weeks.

Common Mistakes That Waste Months of Work

Leading questions are the easiest to make and the hardest to catch in your own transcripts. "Don't you think the policy was unfair?" implies the policy was unfair and the respondent should agree. Rephrase it as "What was your experience with the policy implementation?" The difference sounds subtle but it completely changes what kind of answer you'll receive. Not establishing rapport properly before asking sensitive questions is another frequent error. I once saw a doctoral candidate push into questions about household income during an interview where the participant clearly wasn't comfortable yet. The participant gave a fabricated number, which the researcher analyzed as genuine data for three months before realizing it didn't match any of the other indicators in the dataset. Starting with low-stakes questions for the first fifteen minutes prevents this. Analysis paralysis is real. Some researchers collect forty-five interviews and never code a single one. Set a target before you begin. Twelve to twenty well-conducted interviews typically reaches saturation for most qualitative studies, though complex populations may need thirty. Saturation means you're no longer discovering new themes, only repeating existing ones in different words.

Software Options That Actually Help

NVivo costs money but handles large datasets cleanly. Dedoose runs in a browser and costs less. For smaller projects, Taguette is free and open source. The software matters less than the coding discipline. Consistent codes applied systematically beat fancy features applied inconsistently every time. Member checking — sharing your interpretations with participants — is debated but useful. Not every participant wants to see your analysis, and some interpretations you develop post-interview may not be shareable. But when you do it, it catches major misreadings that no amount of reflexive journaling will correct. One participant once pointed out that I'd interpreted her silence as agreement when it was actually disagreement she wasn't comfortable voicing directly. That changed the entire framing of that case. The hardest part of Interviewing For Social Scientists isn't learning the technique. It's developing the patience to let silence work for you. Most beginners fill every pause. The pauses are where participants decide what to tell you next. Sit with them. The data you get in those quiet moments is usually the data you can't get any other way.