0:00 Hi everyone.
0:00 I'm going to show you how to analyze
0:02 qualitative data using ChatGPT.
0:05 Imagine you have interviewed
0:07 participants, you are gotten all the
0:09 information that you need and you
0:11 want to make sense of the information.
0:13 What do you have to do?
0:14 You could use chart G P T to
0:17 make sense of your data, and I'm
0:18 going to show you the process.
0:20 I'm going to show you how to
0:21 summarize your qualitative data,
0:24 how to do the summary based on your
0:26 research question that you have,
0:28 how to generate initial codes.
0:31 And also lastly, I will be
0:34 showing you how to develop themes.
0:36 So these are the four things
0:37 that you're going to learn about.
0:39 If you have any question, feel free
0:41 to put in the comments section.
0:42 I'll be happy to address them for you.
0:45 Summarizing your qualitative data.
0:48 Sometimes you need to do the summer
0:51 just to get a big picture of what
0:54 the data is telling you and ChatGPT
0:56 can help you to do the summary.
0:58 Imagine you have interviewed
0:59 participants who are Primary healthcare
1:02 physicians - collecting data from them.
1:04 They are five, this is one of the data.
1:06 You collected the demographic
1:07 information and also questions about
1:09 the experience of burnout, the causes
1:12 and also the solutions of burnout.
1:14 So I have five of these documents,
1:16 see participant one to five.
1:18 I have already analyzed the data
1:21 using qualitative software NVivo.
1:24 NVivo is a software that you
1:25 can use to analyze your data.
1:27 So as, you can see here I have
1:29 my research question here and
1:31 the codes that I've developed.
1:32 For the first research question
1:34 and the second research
1:35 question, I have all the here.
1:38 As we are going through the process,
1:40 I'll be comparing what ChatGPT will be
1:43 giving me to the data I have analyzed.
1:46 And see whether there's a relationship
1:49 or differences between them.
1:51 Can ChatGPT give me similar results?
1:53 That's what I want to find out.
1:54 So first of all, what you have
1:56 to do is to do the summary.
1:59 We are going to copy all the data.
2:02 So I think they are about one
2:04 page for each of the transcripts.
2:06 So we're gonna have about five pages.
2:07 So you're gonna copy all of
2:09 them and put it in ChatGPT.
2:12 One thing that you have to notice, that
2:14 ChatGPT can only take about 3000 words.
2:19 If you put more than 3000 words
2:21 there, the system will tell you that
2:23 it's too much for it to process.
2:25 But there's a way out.
2:26 One thing that you could do is that you
2:29 can send that information in chunks.
2:31 You can send maybe first 3000
2:33 words, second can 3000 words.
2:35 Then after that, you tell ChatGPT to
2:37 review all the information that you have
2:40 sent, and then, develop a summary for you.
2:43 But this one I just want to show you
2:45 how to summarize our data using ChatGPT.
2:48 So let's see.
2:49 So now that we have our data, we
2:51 are going to go to ChatGPT and copy
2:54 and paste that information there
2:56 and see what we are going to get.
2:58 Don't forget to subscribe
2:59 to my YouTube channel.
3:01 Trying to make sure that I give you all
3:04 the information that you need to be a
3:06 very good qualitative researcher, in
3:07 terms of doing your research and analyzing
3:10 your data and presenting your findings.
3:11 So follow my channel and I'll be happy to
3:14 also address any questions that you have.
3:17 I have ChatGPT plus.
3:19 So I have the two versions.
3:21 If you have the free version,
3:23 you have only GPT 3.5.
3:26 So we are going to use the GPT 3.5.
3:30 So let's start.
3:31 First, what I would do
3:33 is to put in the prompt.
3:36 The PowerPoint has all the
3:38 prompts there for you to practice.
3:40 So this is the prompt.
3:42 Summarize the transcript
3:43 in thousand words.
3:46 You can put in a different way.
3:48 You can say, summarize the
3:50 transcript below in thousand words.
3:53 So the thousand words could be
3:54 about two pages or so, right?
3:56 You can say 500 words,
3:58 you can say 200 words.
3:59 But I like thousand words because
4:01 it gives you a lot of information
4:03 that you could get a big picture
4:06 of the data that you have.
4:08 So now that I've finished writing this
4:10 one, I'm going to copy and paste all
4:13 the information from the transcript.
4:15 Let me copy here.
4:17 This is the first one I paste here.
4:20 And if you click on
4:21 enter everything you go.
4:22 So don't click on enter yet.
4:24 Copy and paste everything first.
4:26 I'm looking for the second transcript.
4:31 Let me open the second transcript
4:33 and then copy and paste.
4:35 Normally, I use this form of summarizing
4:38 to help me to better understand,
4:42 what exactly the data is all about.
4:45 So let's say if you want to get initial
4:47 understanding of the data, this is
4:48 the good practice for you to do.
4:51 So now that I've copied everything.
4:54 The next step is to click this one or
4:57 just press or enter on the keyboard.
5:00 So you see here that the system
5:02 has summarized everything for you,
5:04 and it's about maybe a thousand
5:06 words I haven't checked yet.
5:07 I can copy and put it on a Word
5:09 document to count the words.
5:11 But as you can see here, the system
5:12 have provided you information
5:14 about the first interviewee.
5:16 And the second participant,
5:18 the third participant, fourth
5:19 participant, and fifth.
5:21 It has summarized each of
5:25 the transcripts for you.
5:27 I think that's very good information
5:29 because now you have a brief information
5:31 about each of the participants.
5:32 So I like this one.
5:34 You get an understanding of
5:35 each of them before even you
5:37 go into the data and analyze.
5:39 You don't have to use chatGPT
5:42 to fully analyze your data.
5:45 You can use it now like what I'm doing,
5:47 to just summarize and learn about the
5:50 data before you go through line-by-line
5:54 and manually develop codes to address
5:57 your research question that you have.
5:58 So you can use it and at the
5:59 same time doing the manual
6:01 coding or using the software.
6:03 So now it's good you have brief
6:06 information about each of the participants
6:09 and instead of you going through all
6:11 the transcripts and learn about them.
6:12 At the beginning of the stage of
6:14 your data analysis, having this
6:16 one will be very helpful for you.
6:18 This kind of summary we call it
6:19 general summary because we didn't ask
6:21 ChatGPT to address a question for you.
6:24 You ask it to just summarize.
6:26 Imagine that you have a research
6:27 question and you go back to ChatGPT
6:29 to review the transcripts and address
6:32 the research question that you have.
6:34 So that's what we're gonna
6:34 do for the second step.
6:36 This is the prompt that
6:37 we are going to use.
6:38 Summarize the transcript based
6:41 on the research question.
6:42 What are the causes of burnout
6:44 among primary healthcare physicians?
6:47 So let's try and see
6:48 what we are going to get.
6:51 We put the prompt here, and let's see.
6:55 So we can see here that, ChatGPT
6:57 has provided a summary addressing
6:59 the research question that we have.
7:00 So you can see here that they talk about
7:02 heavy workload categorized by long hours,
7:06 numerous clinical and administrative task.
7:09 So you can see here that the system
7:11 has provided some of the information
7:13 about the causes of burnout.
7:15 This one is also talks about
7:18 lack of work-life balance.
7:20 It is also mentioned a coping
7:21 mechanism, but that's not our focus.
7:24 Our focus is more about
7:26 the causes of burnout.
7:28 It said overall, the causes of burnout
7:30 among primary healthcare physicians
7:31 identify in the transcript include heavy
7:34 workload, administrative burnout, lack
7:38 of work life balance, high expectations...
7:41 that's very excellent.
7:43 Right?
7:43 So ChatGPT is very good in terms
7:46 of summarizing your data and
7:49 addressing your research question.
7:51 You see that there's some
7:53 similarities between the summary
7:55 ChatGPT has given us and also what.
7:57 I analyzed.
7:58 Trouble carrying out duties, having
8:00 long hours, less time for themselves
8:02 and family, no work family balance.
8:05 I think that you see the summary
8:07 covers almost everything that I coded.
8:10 These are initial codes that I did
8:13 right when I was coding the data.
8:15 So you see how.
8:16 You could even use as a way of validating
8:23 the code that you have created, right?
8:25 So you can analyze the data and use
8:28 ChatGPT to get a summary, addressing
8:31 the research question and then
8:32 you can compare what you found.
8:34 I can see that there's a lot of
8:36 similarities between what I did
8:38 and also what ChatGPT gave me.
8:41 Another thing that we can do is that
8:43 after you have done the summary,
8:45 addressing the research question,
8:46 let's go into developing codes.
8:49 This is where it is going be very
8:50 interesting because I'll be comparing
8:52 the calls that I'm going to get with
8:55 what ChatGPT is going to give me.
8:57 Another prompt that we can use
8:59 is review the transcript to
9:01 identify relevant excerpts that
9:04 address the research question.
9:06 What are the causes of burnout
9:08 among primary healthcare physicians?
9:10 Generate phrases, or codes that best
9:13 represent the excerpts Identified.
9:15 Each code should be
9:16 between two to five words.
9:19 You know.
9:20 This one we call it
9:21 structured prompting, right?
9:22 Structured prompting is where you
9:24 give, the system adequate information
9:27 for you to get what you want.
9:29 You tell the system to look
9:30 at the transcript, right?
9:32 So every answer that the system is
9:34 going to give you should be based on
9:36 the transcript that you have provided.
9:39 Then you tell the system that
9:41 address this question for me, and
9:45 you ask the system to develop codes
9:47 addressing this research question.
9:49 Also you want the code to be
9:51 in between two to five words.
9:54 So you see how you're provided a
9:55 lot of information for ChatGPT to
9:58 get the information that you need.
10:01 So let's try and see
10:02 what you're going to get.
10:03 So we copy this one and paste it into
10:06 ChatGPT and see what you are going to get.
10:10 This is interesting.
10:11 This is what a ChatGPT has given you.
10:14 It extracted accepts from the
10:17 transcript and also developed codes.
10:20 So you see how interesting it is?
10:23 ChatGPT goes through the data extracts
10:27 significant information that can address
10:30 the research question and develops
10:31 codes for you and see all these codes.
10:34 We have a lot of codes here.
10:36 We have about 20 codes.
10:39 Another thing that we have to really look
10:41 into is that we have to verify whether the
10:46 significant information that the system
10:47 has given us is truly from the transcript.
10:52 Sometimes ChatGPT can make things up.
10:55 You look here, ChatGPT may
10:57 provide inaccurate information
11:00 about people, places or facts.
11:02 So you have to be very careful.
11:05 Don't over depend on the output.
11:08 You always have to verify.
11:10 So one way of verifying is going
11:12 back to your transcript and
11:14 see whether the significant
11:15 information were truly from the data.
11:18 It's not like the system made it up.
11:20 So, let's see whether I can
11:22 find this one from the data.
11:24 So the first participant,
11:29 oh, it looks like it's here.
11:33 You see, long hours, numerous
11:35 clinical and administrative tasks.
11:37 So this shows that ChatGPT
11:41 directly extracted that
11:42 information from the transcript.
11:44 Let's try the second one, a lack
11:46 of balance with work and home life.
11:49 Oh, I think it's here.
11:51 You see a lack of balance
11:53 with work and home life.
11:54 So just checking too.
11:56 I think I now have confidence
11:58 in the significant information
12:00 the system has extracted.
12:01 So, you always have to check because
12:04 you don't want to just take this
12:07 information and start presenting it
12:09 to your audience without checking.
12:11 So another thing that we can also check
12:14 is the code that ChatGPT provided.
12:17 Let's try to compare it with what I did.
12:23 Although my codes are a little
12:25 bit different in terms of the
12:27 construction, they are quite similar.
12:31 Work life balance.
12:32 I have something like
12:33 no work family balance.
12:35 So there are some similarities
12:37 concerning what I developed
12:39 and also the system developed.
12:41 So I think that it looks like
12:43 ChatGPT is doing a great job
12:45 concerning making sense of data,
12:47 so you see how you can work hand in
12:48 hand with ChatGPT, especially when
12:52 you reach a stage where you're having
12:53 difficulty coming up with a code, right?
12:56 So you can use ChatGPT first to generate
13:00 codes, and then you use the codes to
13:03 go through the data that you have.
13:05 So it helps you with the beginning
13:06 to start the coding process.
13:10 So that's what I would use if I
13:12 were to analyze my qualitative data.
13:17 So now that we have been successful
13:20 developing codes, let's go ahead
13:23 and do something interesting.
13:26 What we could do to follow a systematic
13:29 process is to ask ask system to
13:31 review the codes and generate
13:33 themes out of the codes, right?
13:37 Or another option is to.
13:40 ask system to use the summary to
13:43 generate themes to help you to
13:46 address your research question.
13:47 I'm thinking that if you feel satisfied
13:51 with the initial code s ChatGPT has
13:53 given you, then you can go ahead and ask
13:57 ChatGPT to categorize the codes for you
14:00 so that it can develop themes for you.
14:03 I'm fully satisfied with
14:04 what ChatGPT has given me.
14:06 So what we are going to do is that
14:08 we are going ask ChatGPT to generate
14:11 themes, addressing the research
14:12 question based on these codes.
14:17 So we can say, can you categorize
14:21 the codes and generate themes
14:25 addressing the research question.
14:26 Let me paste the research question here.
14:30 Addressing the research question,
14:32 what are the cause of burnout among
14:34 primary healthcare physicians?
14:36 And I said each themes should
14:39 be in between two to five words.
14:42 Okay, so let's see what
14:46 the system will give us.
14:55 Okay, so we have seven themes, right?
15:00 As you can see here,
15:02 workload and time pressure.
15:03 And we have these under
15:05 workload and time pressure.
15:08 And then we go to work life imbalance,
15:11 and we have all these codes.
15:13 So what you could do is after you have
15:16 gotten this one, you can review the codes.
15:19 Because at the end of the day, you also
15:22 have to give a human touch to it, right?
15:25 So if the system gives you
15:26 this one, don't run with it.
15:29 Always verify, examine, question the
15:32 results, and make sure that the codes
15:35 that are under each of the themes
15:37 are supposed to be there, right?
15:38 So you see how easy the process is.
15:41 We started with a summary and then
15:44 we went to initial codes, and now
15:46 we have asked system to categorize
15:47 the codes for us to develop themes.
15:51 And also you can ask the system to
15:53 put the results in the table form.
15:55 So you can say, can you put this
15:59 recent output in a table form.
16:02 Let's see what this, see, nicely.
16:06 The system has put
16:07 everything in a table form.
16:10 You see how amazing ChatGPT is?
16:14 Putting everything in
16:14 the table form for you.
16:16 Let's ask another question.
16:18 It's gonna be very interesting.
16:19 It's getting interesting.
16:20 Can you add a third column
16:25 to the table and the column.
16:29 The new column should have excerpts from
16:34 the transcript supporting the themes.
16:39 Let's see what we're gonna get.
16:47 Wow.
16:48 You see how, it's interesting that
16:52 now you have the theme you have codes,
16:54 you have excerpts from the transcript,
16:58 but remember, always check whether the
17:00 excerpts are truly from the transcript.
17:03 So it's very important for you to check.
17:04 You finish your analysis.
17:07 Amazing work, what you
17:09 can do with ChatGPT.
17:11 You can use it, especially when
17:12 you have a huge amount of data.
17:14 You don't have time and
17:15 resources to analyze all the
17:17 data or analyze everything.
17:19 So you can use our gpt as a way of
17:22 getting your codes, your themes and
17:24 excerpts, and presenting your findings.
17:26 Amazing.
17:27 I hope this one was so helpful.
17:29 If you have any question, feel free to
17:33 put Your question in a comment section,
17:35 I'll be happy to address them for you.
17:38 And I hope this one was so helpful.
17:40 If you want me to come and speak to your
17:42 organization about ChatGPT and doing
17:44 research, I'll be happy to do that.
17:46 Email me and I'll be happy to respond
17:49 to your email and if you want a
17:51 consultation, if you want to learn
17:53 how to use ChatGPT to analyze your
17:55 qualitative data, I'm there to help.
17:58 Enjoy the rest of your day.
18:00 Thank you so much for your time.
