A Practical AI-Assisted Workflow for Qualitative Analysis (Full Transcript)

[00:00:00] Speaker 1: I'm going to show you the simplest way to analyze your qualitative data. Not only just developing calls and themes, but developing themes that will help you to address your research question. And this time, we are using AI Assist to do that in MySQL 8. We are going to follow only four simple steps, and then by so doing, you'll be able to reach your destination of having your themes using AI to help you. So we're going to start. First, I want to show you the data that I'm going to use. It's about burnout. I've been using this data all the time. So I have five participants here, and I want to explore the burnout, the causes of burnout, and the solutions of burnout. And also, let me show you, these are the research questions that I want to address in my study. And if you want access to this data set, I'll be sharing this in the description area so that you'll be able to download and use it. First, we're going to use AI to explore our data. Secondly, we're going to use it to develop codes under the research question. We're going to develop themes, and then we can also explore the connections between the themes. So I've opened a new project. The first stage is to bring all the transcripts here. So I'm just selecting all the transcripts. I bring it right here. I have all my transcripts here, five transcripts. Then the next one is to bring demographic information, if you have it. Make sure that it's formatted this way. So as you can see here, we have the first column should be always document group. The second one is document name, which is all the names of the transcripts. So you see the transcript is P1, P2, P3. And the age, gender, and all these are demographic information. So the first two should always be document group and document name. And under the document name, you have all the names of your transcript. If you don't arrange it this way, you will not be able to upload that information into the system. So how do you upload? You go to Variables, you go to Data Editor, and then you click here to Import. And then you look for, so let me look for that information over here. And then it's also giving you a preview about all the variables. And then if you are okay with what you see, you click on Import. And then you have your demographics. Perfect. Now, the first step is to explore the data. We have to assume that you have gone through all the transcripts to familiarize yourself with the data. But you can also further familiarize yourself by going here, click here. You see here, AI Chat, and then you have to log in before you can use it. Another option is you go to AI Assist and then click on AI Chat. And then you go ahead and log in. So I'm going to log in. I have logged in. I just indicate here to Agree and start chatting. Let's look at the first step. What do you have to do for the first step? You are exploring the data. So what do you have to do is you can ask the system some questions based on your document to just get an idea about what kind of information that you are going to review or anything that the system has observed. But before you do that, one thing that you have to do is to go to here.

[00:03:55] Speaker 2: See this place? Right.

[00:03:56] Speaker 1: Click here and also provide information about the study. You can provide the purpose of the study and also the research questions that you want to address in your study. Why do you have to do that? You want the system to get the background information before you chat with AI. So you see here that I brought the purpose of my study and also the two research questions that I have. And then I go ahead and click on Save. You can also give more instructions if you want. It's not really needed. It depends on you. This one is how you want the system to customize its response to you. Right. Do you want things in bulletin? Do you want short answers? You can provide that information. I will leave this place alone and then click on Save. Now this is what we're going to do. We're going to explore the data. But one thing that I really want to explore is called needle in the haystack. It's called needle in the haystack. So it's not, you don't have to do it, but if you want to be curious in terms of to be sure that a system will go through all your transcripts, you're going to hide a specific word or a number in one of the transcripts and ask the system to find where the number is located. Just to get an idea that a system has the capability of going through all your transcripts. It's also to test the amount of information the system can review at a time. So it's always good to do that. You don't have to, but it's good to do so that you are sure that a system does that. This is what you're going to do. We can close this one and we're going to hide. Let's go to P3 and hide something there. So this is, let me see. Okay. So this is P3. We're going to hide a word here, blue, and then click here to, you have to first click here for the system to be able to edit the document. So we're going to hide a word here called blue, and then we're going to see whether the system can find where the blue is located. It is located on P3, P4, P5, or P1. And as you can see, we put it in P3. Now we lock this one. We go to the chat. You click on document. You select all the documents. You click on chat. Now, which of the transcripts are the word? So which of the transcripts has the word blue? So within a few seconds, it will go to the transcript and then it will provide us the answer. We hope it indicates P3, yes. So this is an assurance that the system truly go through all the documents. If let's say that this system doesn't find that, this is indicating that the system was not able to go through all the documents. So what you may have to do is to work with a batch of documents at a time. Maybe if you have 20 participants, you can work on maybe 10 participants transcripts at a time or seven at a time, right? So this one is an assurance that a system went through all the transcripts. So you can ask this, when reading the transcript, observe that having long hours was one of the causes of burnout. Does this apply to all participants? So let me copy that and then put that information here and see what the system will say.

[00:07:50] Speaker 2: Okay.

[00:07:51] Speaker 1: So it looks like it applies to almost all of them, right? And the good thing that it's also provided a link. So when you click on this link, it takes you to the source. So you see how easy the process is. So you can always review. When it comes to using AI to do your analysis, you always have to make sure that you are leading the analysis. This means that you have to be skeptical and review any information the AI generates. Why do you have to review? Because it can give you wrong information. Secondly, you are leading the data analysis process. You don't want to relinquish all your responsibility to AI to control the process. If you do that, then it will be very difficult for you to even explain how AI use it. You are leading the research, the AI is just a tool helping you to make sense of your data. So always evaluate and make sure that every information is good, right? So when it provides information, you just review, you can look at a source and look at whether the system is giving you correct information. Always do that. And also document your actions and your decisions, right? Because after that, you should be able to show to your audience how you use AI. What role do you play in the data analysis process? Did you evaluate the AI output? What decision do you make? You have to always document the process. So the good thing is that this one, you're always going to document your conversation. So you can, after everything, you can review the conversation and then write about the interaction you have with the system and the decisions that you've made. We can ask a second one and then we move on to the next stage. So what is the second question that we can ask? What pattern have you identified? Let's see whether it has identified any pattern. Okay. In terms of career change or the career stage. So those who are in the early career stage are quite vulnerable because they have less experience and then they have to learn new things and a lot is expected of them. And the good thing that the system has provided you a link to also click on to go and see where it's making that observation or conclusion. Okay. So the older and more experienced participants indicated that they have lower level of stress. And guess what? Why do they have lower level of stress? Because they have more experience, they don't know how to handle the issues. They have more control by the time that they use and they have developed a skill to be able to carry out their duties without stress. So or with a little bit of stress, right? So you can click here and then you can see. So the pattern here is that the younger generation, the younger physicians have more stress and burnout compared to the older folks. Okay. So in the case, another observation here is the conflict between administrative burden and patient care. Okay. And then there's a cesspitation gap too that has also been observed. Okay. So you can review and then see whether the observation is the pattern that identifies right. You can also click on the links for you to see the source of the information. The next step is to develop codes, but before we develop codes, we have to create containers for each of the research questions, right? So because at the end of the day, we want to address a research question. So what are we going to do is that we're going to close the chat window for now and go here, click on this one and then indicate, I'll kill one, which is research question one. So we talk about burnout causes. And then you choose a color that you want and click on all, create, and then do the same thing again here. I'll kill two, burnout solution. And then you click on any color that you want. Okay. So now we created containers for research questions. So we go ahead and chat with the system again. So you click on the chat again, you're going to select documents and then select all the documents. You click on start chat. This time, let me show you the prompt that we're going to use. So this is a prompt, the first one that we're going to use. So the system already knows the purpose of the study, but it doesn't hurt for you to talk about it again. All right. Just to remind the system that this is the purpose of my study. So I indicated the purpose of my study and also I indicated that it should generate descriptive codes based on the segment, right? So it should generate codes that correctly describe what participants said. So it will identify information that is significant and then develop a code. So the segment is called quotation and then descriptive codes that the system is going to develop based on the significant information that are selected. Each code should be in between two to five words. You always have to specify this so that the system will do what you expect it to do and then address the following research question. So this is a research question about the causes of bailout and then the system, again, you want the system to group the segment under their respective codes and then can you create not less than eight codes. The reason why you do this is that sometimes you don't do that to create less codes. At this point you want more codes so that you can explore them and categorize them to develop themes. The more codes sometimes it's good because it captures unique information the system has identified. If you limit it too much, you may lose the richness of the information in the data. So I'm going to copy this prompt. If you want to get access to this prompt, I'll put a link there in the description area so that you can click on the link to get access. So let me put it here. So as you can see, I have the prompt here and then click on enter and it will take some time, maybe a few seconds or a minute for it to go through everything and develop. So you see here that the system has developed codes for us and a good thing is that it also has provided quotations that are linked to the code. So what you have to do as a researcher is to review all the codes and ask yourself, is this code addressing the research question? Yes, excessive administrative burden is addressing the research question. And the second one, and is this codes represent the significant information that has identified? Okay. It talks about administrative tax here, administrative tax here too. So you can see that this fairly represents all the quotation that is here. If you want to know the context of the quotation or the source of the quotation, you click on the link here. You can see here and then it will take you to the source and you can read more about that. So you have to review everything and if you are satisfied, then you move to the next stage. What do you have to do? We're going to transfer all these codes and bring them under the research question. And then we're going to also look for the quotation and drop them into the code. So this is where you have to manually do that. So I'm going to select a code, copy that, go to the plus sign, connect it to research question one, click on that, and then put that information here. You can also give a description of this code, what the code means. You can say, this code represent participant expression of doing a lot of administrative work or something like that. It doesn't have to be perfect, but it's not required for you to provide that information. So if you put it here and you remember what this one means, then you can go ahead and click on create codes. So that's what you're going to do. And then after creating code, you click on the first one here and then you select and drag and drop it into that container. You click on the next quotation, you see the next quotation here, and then you drag and drop it into that container. So you see how simple that process is. The system provided all the information. What do you have to do is just manually click on the source and then drag and drop it. I think we finished with the first code. We do the second code, copy that. We go here, we go to research question one, the plus sign, we bring that information here. We put the code here. We can give a description if you want, it's optional. You click create, click on create code, and then you have it here, and then you come back here and just copy, click on that. If you are okay with the quotation, you just drag and drop it into the respective codes. So that's it. That's what you're going to manually do. All right. Let me know whether you have any questions and don't forget to subscribe to my channel. It goes a long way to help. So that's what you're going to do. Let me show you the FANA. Okay. So you see what I've done now. I went through, just ignore this one. This one is the stage where I was developing themes.

[00:17:56] Speaker 2: I can even move this one here, here, move that one here.

[00:18:00] Speaker 1: So as you can see here, I have all my codes here under the research question and I put all the significant information or significant segments into their respective codes. I did the same thing for research question two. If you want to see the significant segment, you can just double click on this one and then you can see them. We finished the first stage, the second stage, which is developing codes now. If you want to see any codes, as I said, you double click on it, click here and you see all the quotations, right? And then if you decide, you can decide, okay, I don't like this quotation. It doesn't really address the research question or it's not really connected to this code. You can always click on the X and then delete, right? So this is what I have for you for the second stage. Let's go to the third stage. We are developing themes. So the next step that you have to follow is to bring all the themes here so that you can categorize the codes under the themes. Okay. So this is what you have to do. What do you have to do now is to first select the code for research question one and select all the documents, right? So that the system will know that this is a code that you want the system to categorize. Then you see here, AI assist, you click on that and click on chat with retreat segment. Then you can bring the prompt, copy the prompt and bring it here and click on enter. This one, a system will only focus on the codes that are here, right? And categorize them to develop themes. So you see here that the system has done that. So now we have four themes here. So the next step is to copy the theme. Okay. So before you even copy the themes and create codes for the themes, you always have to review all the information here, making sure that they are correct. If they are not, you have to reject it, make an adjustment or do further interaction with the system, right? So one of the codes, the theme here is systemic operational pressures. And these are all the three codes that are linked to the theme. So you look at this name, is it really addressing the research question? Look at the definition. Is it really reflective of what the theme is? And then you look at all the codes, are they linked to this theme? So you always have to, you always have to review before you go ahead and bring the themes to the left side here, right? So imagine you have gone through and everything is right. You just click, select this one, copy and go to the plus sign here, right? And then bring the theme, the first theme. And then you can type here, theme, just to remind you that this is a theme. And then you can give a description if you want to, and click on create code. And now I'm going to look for the codes and drag and drop. So administrative burden, drag and drop here, emphatic work schedule, drag and drop. And then the next one, I drag and drop. So I've done with the first one, I go to the second one. I copy, plus sign behind the research question one, type theme, and then I click on create code. I look for the information, early career, moving work, and then I do the next one. So that's what you're going to do. See how I've created themes and then categorized the code under each of the themes. So that's what you're going to do for all of them. And after that, you do the same thing for the second research question. So imagine I finished the first, I go here, I unselect these, and now I'm going to select there, or we call it activate all the codes under research question two. I make sure that this one also are selected. And I go here, I click on chat with retrieve segment, and then I go ahead. There are 13 segments, and I bring this prompt here. This time I'm asking the system to focus on research question two, and the system will be able to do that for me. So that's what you're going to do for them. And let me show you how the final product look like. Okay. So as you can see here, you see research question one, I have all my themes here. As you can see, this one is a little bit different from the themes name. The name of the themes are a little bit different from the one that we did. Why is it that? Because the system, every time that you run the analysis, it gives you a different kind of name of the theme, but the meaning are similar, the meaning are the same, right? The most important is the meaning. So you see here, all the meaning are quite similar. So now I did the same thing for the second research question. This is the first research question, and then this is the second research question, right? Now what is the next step? The next step, you don't have to do it. You do it, especially if you think that the theme that you have developed cannot individually address the research question. Sometimes to adequately address the research question, you have to explore the connection between the code, the connection between the theme. If each of the theme can individually address the research question, like this one, I don't have to explore the connection because each of them are individual causes of burnout. I don't have to link this one to this one to address the research question. In this case, I don't have to go for it. But if you want to explore more, yes, you can do that, especially if you have enough data to support the connection. You can do that. Or if you are using a granite theory approach where you want to develop a theory to explain a specific phenomenon, then you can go further with the last step. So this is the bonus step. This is the fifth step, which is optional. You want the system to explore the relationship between the themes. So is there any concurrent relationship between the themes? Concurrent. Are the themes happening at the same time? Do they happen at the same time? Is there a hierarchical relationship where one theme comes under the other theme? Is it embedded? Is one theme nested in another theme? Is it sequential? Is one theme following another? Is it causal? Is one theme causing another? So all these are areas that you can explore. We can do that right now and see what we're going to get. So what we're going to do is we're going to select this prompt. And then we can go to here. And then let's try for research question one. So I select all of them. I select here. And then I go to this place. And I select Chat with Retrieved Segment. And then put that information here, the prompt, and click on Enter. So it will give you first whether there's a relationship, what kind of relationship, and also evidence in supporting the relationship. Is there any evidence from the data to support? As we can see here, the system indicates that there is a causal relationship between two themes. And it has provided some evidence for us to explore. The fact that the system is saying that there's a causal relationship doesn't necessarily mean that you just have to accept. You have to explore and see whether, based on the data, this relationship exists. If it doesn't exist, you just leave it alone, right? Or you modify it. This will show an overlapping relationship between work-life imbalance and headache work schedule. So it has made an argument. So think about this as a claim. The system has given you claims. Your role is to look at the evidence in supporting the claims and see whether it's appropriate. If you don't have enough data, you just reject the relationship. So that's what you're going to explore and then see what is really going on. And then you can go further. You can copy this and paste it on your document. If you want to see your chat history, you can click on here and see all your conversations you have had with the system. And then so it doesn't vanish. You can always go back and review the communication you have had with the system, right? So this is what I have for you. Let me know whether you have any questions. I'll be happy to address them for you. And I provide one-on-one consultation if you want to meet me and I will be able to teach you how to use this software to analyze your qualitative data. And if you are in an organization and you want training about how to use AI tools to make sense of your qualitative data or how to assess the quality of the AI output, I can come and give you training for you to gain the skills so that you'll be able to use AI in an ethical way. You can use AI in a responsible manner. Thank you so much for your time.