Recording good audio for qualitative interviews and focus groups

 

Last week’s blog post looked at the transcription process, and what’s involved in getting qualitative interview or focus-group data transcribed. This week, we are going to step back, and share a few tips from researchers into what makes for good quality audio that will be easy to hear and transcribe.

 

1. Phones aren’t good enough
While many smartphones can now be used in a ‘voice memo’ mode to record audio, you will quickly find the quality is poor. Consider how tiny the microphone is in a phone (the size of a pin head) and that it is designed only to pick up your voice when right next to your face. Using a proper Dictaphone or voice recorder is pretty much essential to pick up the voice of interviewer and respondent(s) clearly.

 

2. Choosing a Dictaphone
Even if you want to buy one, a cheap £20 ($30) voice recorder will be a vast improvement over a phone. Most researchers won’t need one with a lot of memory: just 2GB of storage will usually record for more than 30 hours at the highest setting. There is usually little benefit in spending a lot more money, unless your ethics review board states that you need one that will securely encrypt your data as it records. These might cost closer to £250 ($400). However, you can often borrow one from your library or department.


A recorder should always be digital. There is no real advantage to a tape one – they are expensive, have less capacity, use batteries faster, are larger, and are much more prone to losing your data with erased, overwritten or mangled tape. This is one part where the advanced technology wins hands down! The format they record in doesn’t really matter, as long as your computer and transcriber can play it back. MP3 is the most compatible, note that some of the older Olympus ones use their own DSS format which is a pain to convert or play back on a computers. Digital recorders will have various settings for recording quality, you will usually want to choose the high or highest setting for clear audio. Test before you do a full interview!

 

2. Carry spare batteries!
I’ve definitely got caught out here before, make sure you have a fresh (or recharged) pair of batteries in the Dictaphone, and a spare set in your bag! Every few minutes during the interview, have a quick look to make sure the recording is still running, and before you start, check you have enough time left on the device.

 

3. Choose a quiet location if possible
While cafés can be convenient, relaxed and neutral places to meet respondents for a one-on-one, they tend to be noisy. You will pick up background music, other conversations, clattering plates and especially noisy coffee machines that make the audio difficult to transcribe. A quiet office location works much better, but if you do need to meet at a café, try and do a bit of reconnaissance first: choose one that is quiet, don’t go at lunchtime or other busy times, choose a part of the café away from the kitchen and coffee grinders, and ask them to turn off any music.

 

4. Position the Dictaphone
Usually you will want the Dictaphone to point towards the respondent, since they will be doing most of the talking. But don’t put the Dictaphone directly on a table, especially if you are having tea/coffee. You will pick up loud THUD noises every time someone puts down their mug, or taps the table with their hand. Just putting the recorder on a napkin or coaster will help isolate the sound.

 

5. Prevent stage fright!
Some people will get nervous as soon as the recording starts, and the conversation will dry up. To prevent this, you can cover the scary red recording light with a bit of tape or Blu-Tack. However, it can also help to start the recorder half-way through the casual introductions, so there isn’t a sudden ‘We’re Live!’ moment. You don’t need to transcribe all the initial banter, but it helps the conversation seamlessly shift into the research questions. Also, try and ignore the Dictaphone as much as possible, so that you both forget about it and have a natural discussion.

 

6. Watch your confirmation noises!
Speaking of natural conversation, it is rare while listening for the interviewer not to make ‘confirmation sounds’ like ‘Yes’, ‘Uh-ha’, ‘Mmm’ etc. Yet these are a pain for qualitative transcription (as most people will want to keep the researchers comments, especially for discourse analysis) and it also breaks up the flow of the transcript. Obviously, just staring silently at your participant while they talk can be disconcerting to say the least! It takes a little practice, but you can communicate and encourage the flow of the conversation just with periodic eye contact, nodding and positive body language. If someone makes a request for confirmation such as: ‘So of course that’s what I did, right?’ Rather than actually verbally responding, you can nod, turn your palms up and shrug, and roll your eyes. This way, it shows you are listening and engaging with the conversation, without constantly interrupting the flow of the narrative.

 

7. Use a boundary mic for group discussions
For focus groups or table discussions, use a cheap ‘boundary’ microphone so that it will pick up all the voices: ideally stereo ones that give some sense of direction to help identify who-said-what during transcription. Again, these don’t need to be expensive: I’ve used a cheap £20 ($30) button-battery powered one with great results. It’s something you can spend a lot of money on for high-end equipment, so again look for opportunities to borrow. 

 

8. Get a group to introduce themselves
For qualitative group sessions, you will almost always want to be able to assign contributions to individual participants. If you are doing the transcription and know the people very well, this can be easy. However, it is surprisingly difficult to differentiate a group of voices that you don’t know just with a recording. For voices to be identified, make sure you start the recording by getting everyone to go round the table and introduce themselves with a few sentences for context (not just their name).

 

9. Backup immediately!
Got your recording? Great! Now back it up! All the time it exists only on your Dictaphone, it can be lost, stolen or dropped in a puddle, losing your data for ever. As soon as you can, get it back to a computer or laptop and copy it to another location. Make sure that your data storage procedure matches your data protection and ethics requirements, and try not to carry around your interview recordings longer than you need to.

 

10. Finally, listen and engage!
Try not to worry about the technical aspects during the interview, shift into researcher and facilitator mode. Take notes if you feel comfortable doing so: even though you are getting a recording, some brief notes can make a good summary and helps concentration. Tick off your qualitative research/interview questions as you go, and write a few notes about how the interview went and the key points immediately afterwards.

 

If you need more advice, you can also read our top 10 tips for qualitative interviews, to make sure things go smoothly on the day. Hopefully, following these steps will help you get great audio recordings for your research project, that will make transcription and listening to your data easier.

 

 

Once you’ve got it transcribed, you'll find that Quirkos is the most intuitative and visual software for qualitative analysis of text data. You can download a free trial for a month, and see an overview of the features here…

 

 

Transcription for qualitative interviews and focus-groups

transcription and a dictaphone

 

Audio and video give you a level of depth into your data that can’t be conveyed by words alone, letting you hear hesitations, sarcasm, and nuances in delivery that can change your interpretations of what your participants say. Yet most researchers and students will want to have typed transcripts of their qualitative interviews.

 

Text gives many advantages during the qualitative analysis process. You can read or skim read text much faster than you can listen to audio, and your eyes are good at quickly picking out keywords. Transcribed text can also be searched for keywords or synonyms, taking you directly to that point in the text.

 

It’s also easier to code text than multimedia data, and since most research outputs (especially journal articles, a thesis or book) tend to remain stubbornly text-based, including quotes is a standard way to embed qualitative data. Text can also be analysed in other ways, including statistical analysis and automated sentiment analysis.

 

Even when working primarily with the video or audio of qualitative interviews, most academic researchers and students will still generate a transcript for these reasons. But at the moment, there is no software that can reliably understand untrained interview audio, find words or create automatic transcriptions. Either the researcher themselves, or often an experienced transcriber will have to listen to the audio and type it up word for word.

 

Thus BEFORE starting interviews, it is worth considering a few ways to make sure that transcription goes smoothly, and cheaply. Finding a transcriber or transcription service is a key part of most qualitative research. But how much will it cost?

 

Well, this depends on the level of detail required. Verbatim transcriptions, especially when there is a need to capture the nuance of the conversation, are very time consuming to produce. These will capture not just the conversation word for word, but also every um, er, pause and hesitation, and sometimes even infliction. When there are gaps or pauses these will be detailed (such as [pause 5 sec]). This level of detail is illuminating, especially for discourse analysis, but expensive. Often researchers would like regular timestamps included (say at the top of each page) so that it is easy to find the position of the text in the audio. This also increases cost.

 

Often you will hear the phrase ‘Intelligent Verbatim’ used by transcription services, which denotes a middle ground where the transcriber chooses which pauses and detail are relevant, but is careful to make sure that the exact wording of the dialogue is recorded. This is what most qualitative research projects use, unless there is a methodological need for more detail.

 

This is still more detailed than what you would get from a standard typing service used in business, where phrases and words may be approximated. These services, sometimes called a ‘clean transcript’ are cheaper and easier to read (since they don’t have breaks or interjections, they are much more like reading dialogue from a novel), but generally lack the rigour and specificity for qualitative analysis. If someone said ‘afraid’ or ‘anxious’ it might represent a difference in your interpretation, so the exact words uttered must usually be noted. For more discussion, there is an interesting paper by Halcomb and Davidon (2006).

 

If you are conducting focus groups, this can also increase cost and difficulty because of the need to identify the different voices in the room. Typically this can add 20% to the transcription costs, some services will charge for each additional participant. Many transcribers will justifiably add an additional 20% or more for bad audio. We are going to look in the next blog post article about how to make sure this doesn’t happen, but noisy environments and bad recordings make the process much more time consuming, as it is necessary to keep going back and forward to correctly hear muffled words.

 

In general, you should expect to pay between £1 (often $1 in the States) at the absolute minimum and £3 ($2.70) per audio minute for transcription. This means a one-hour interview will cost around £80 ($60) to be transcribed, depending on the quality of the service and number of people speaking. For fast turnaround (ie within 24 or 48hrs) expect to pay a premium. After salaries, it is often the most expensive part of a qualitative research project. So if you have 20 interviews, you will need to budget £1600 ($1200). This is why many students end up doing transcription themselves, and while this is good for keeping close to your data, it is not easy, and can be a false economy.

 

As someone who has also done transcription before, it is vital to stress what a difficult and specialist job this is. Almost no-one can type at the same speed that people speak, and so the work takes much longer than the length of the interview. You are not paying someone £60+ an hour, they will work two or three times that long to get everything typed and corrected. It is also exhausting, and mentally draining. I’ve tried automated software for transcription, like Dragon Dictate or Microsoft’s Project Oxford, but these are not yet geared up for this type of work. They struggle with words that run together, require perfect audio recordings with no extra noises, and can’t identify different voices. I know some transcribers that use trained dictation software in their work: the computer recognises their own voice, so they have to listen with headphones to the audio and clearly repeat every word, one by one.

 

There are many online services offering transcription services, easily found with a quick Google search. I don’t have any specific ones to recommend at the moment, but if you want to use a company I would suggest you choose one that specifically works with research interviews, and offers the options above. It is also a good idea to choose one that works in your native dialect! If you are not used to hearing British, Scottish or Indian accents as an American transcriber (or vice versa), there can be odd misunderstandings and discrepancies that arise.

 

A transcriber that is used to working in your field of study is also useful: they will spot commonly used terminology and abbreviations. My favourite transcriber had worked in a medical field before, so was used to most of the NHS acronyms, and if there were terms or phrases he hadn’t come across before, he would Google them to make sure they were right. Good people like this are hard to come by!

 

Personally, I have always used a few freelance transcribers who work exclusively with universities. Ask your department or colleagues for local recommendations, and if part of a research project, one who is already on the university payroll system can save major headaches and delays. Don’t be afraid to give a new transcriber just one transcript to see how they do, before you commit yourself to giving them all the work. It’s also not a bad idea to have a back-up, especially if a transcriber gets sick, or you need a large batch of transcripts in a hurry.

 

Finally, there will always be errors and uncertainties. You still need to have at least a cursory read through of the transcript to make sure it makes sense and there aren’t typos. A feedback loop is a valuable thing to set up with a good transcriber, so they can learn about common phrases and terms they are mishearing, and the accuracy will improve. Words misheard will usually be marked with [inaudible] and you will need to go through and fix these. Often, it will be obvious to you as the researcher who was there in the room, but not for someone else, especially when it occurs just as the noisy expresso machine turns on!

 

I hope this is illuminating, it’s one of those things that is difficult to find much written advice on. Very few articles discuss this essential part of the research process – Davidson 2009 is a notable exception. Check out some of our other blog post articles for more on this stage, including how to get good quality recordings, and 10 tips for qualitative interviewing, and let us know if you have any suggestions or tips of your own!

 

 

Once you’ve got a transcript, you will be ready to start qualitative coding your text data, and Quirkos is an ideal software tool to bring your interview and focus-group data to life, with a visual and intuitive interface. Download a free trial, or watch a video overview showing you how to start a new analysis project in just 20 minutes.

 

 

Building queries to explore qualitative data

qualitative analysis with queries in Quirkos

 

So, you’ve spent days, weeks, or even months coding your qualitative data. Now what?

 

Hopefully, just the process of being forced to read through the data, and thinking about the underlining themes has revealed a few likely points of interest. Now is a good time to step back, put your research questions in front of you, and think about what the data is telling you about the main topics, and how you can work this into a convincing argument.

 

But it’s also a good time to try something different: to challenge your assumptions and come at the data sideways.

 

If you have been using a qualitative software package like Quirkos, you may already be able to see some trends and connections in the data. For example, what are the themes or nodes which you have coded most to? Are these surprising? Step back a little more: what has been coded more or less than you thought? Don’t forget to look beyond the numbers as well, click on major themes and see what is actually being said.

 

Next, it’s a good idea to try and look at connections between your coded topics. Most CAQDAS software will let you do this: in Quirkos the ‘overlap’ view will visualise which topics were coded together for any of the Quirks in your project. You might be intrigued to see that quotations coded as being ‘Negative’ were particularly correlated with statements about ‘Diet’ or ‘Experience’. Again, make sure you look at the actual quotes, so you can understand the substantive reasons for these associations.

 

Contradictions are also a good thing to look out for. When are people making comments that contrast from the general view? Are there particular contentious issues?  Hopefully, by looking at the quotes in context, and thinking about the narrative or demographics of the person, there will be an obvious reason for this. But don’t bet on it! It’s always OK to flag things that aren’t understood, either for further discussion with colleagues, revisiting the literature, or the fabled ‘more research is needed’.

 

 

A really useful way to explore the themes and trends in coded qualitative data is to do subset analysis. Is there something different about responses from people in difference age ranges? What about gender, or from people who have children? If it’s a literature review, what does one particular author have to say in their later works? Are articles in one journal defending a particular view more than others? A great advantage to using any qualitative software package is the ability to bring up results from just certain sources or responses.  Most CAQDAS will let you do this in some way, but I’m going to go through the process in Quirkos, and how it can illuminate the expected or unexpected.

 

Most of this functionality is built around how you have described your data, also known in Nvivo as ‘Source Classifications’ or the ‘Document Variables’ in MAXQDA. In Quirkos, these are ‘Source Properties’ and can be used to describe anything about the source; be it demographics of the respondent, circumstances of the recruitment, or where and when an article was discovered. See this blog article for more detail on how this can be useful.

 

In Quirkos, you can use the ‘Query View’ to explore your data, by seeing only coded results that meet certain criteria. For example, you can see results just from women, or people in a particular age range. However, you can also see work coded by particular people working on the project, or coding done during a specific period of time. If you’ve created a grouping of Quirks (such as different emotions) you can just see results from these topics.

 

Default query view in Quirkos

 

By default, the first button in the query view shows ‘PR’ for Properties (the source properties you have described). These will automatically be shown the next drop down box to the right, allowing you to select one of any of the properties and values defined in the project. By clicking on the PR button, you can choose from any of these filter options:


PR            Properties                     Which source properties match
HA            Highlight Author           Who coded a segment
QA            Quirk Author                  Who created the Quirk
HD            Highlight Date             When the coding was done
QD            Quirk Date                    When this Quirk was created
QL            Quirk Level                    Which level grouping the Quirk belongs to

You will also see the   =   symbol, by clicking on this, you can change the logic matching to  !=  or ‘not equal to’, so you can get results where Gender was not equal to Male.

 

There are also standard comparison options:


<                      Less than
<=                    Less than or equal to
>                      More than
>=                    More than or equal to


These are useful for date ranges, or numerical source properties like age. So you could get results from all respondents older than 42.

 

But wait, that’s not all!

 

You can also add up to 9 extra criteria to the search, by clicking on the (+) button at the end of the row. This means you can stack search criteria, for example where Gender = Male AND Age > 42. You can even change the AND operand to OR, and thus make your searches wider, rather than more specific. This would give you results from all sources that were Male, as well as respondents (regardless of gender) that were over 42. The example below shows what the response from a search with two criteria might look like:

 

Quirkos qualitative query results

 

Using tools like this, you can explore your qualitative data through many different lenses, and see what interesting things might emerge. You might have a theory, backed up by your own experience or the established literature that certain respondents will behave in certain ways. Or (if methodologically appropriate) you can experiment and try looking at different groups of participants until an interesting pattern emerges. Maybe you will discover that women like toast more than men (as in our example project) or hopefully, something with deeper significance for your research question!

 

Regardless of your discoveries, CAQDAS software like Quirkos can make it easy to filter your coded data, and get an extra level of insight into the underlining intricacies. For further reading, try this chapter from Lewis (2001), and the section on how queries can be used to "test ideas and theories".