Ethical Thematic Analysis with ChatGPT: Step-by-step Tutorial
HTML-код
- Опубликовано: 6 июл 2024
- Step-by-Step explanation of how to use MS Word and Chat-GPT to deliver a high-quality analysis that is academically robust and adheres to ethical standards
I demonstrate how to create codes (initial, open coding), using ChatGPT and Microsoft Word and what to do next, switching between these tools until we develop the final themes.
#qualitativeresearch #academia #researchpaper
👉 My Ebook on how to use ChatGPT for thematic analysis (including a list of GPT prompts) payhip.com/b/KmzOL
➡️ Feel free to ask me questions in the comments, and if you feel that you may require a more guided assistance, you can book a lesson with me through the following link drkriukow.com/qualitative-res...
00:00-02:52 Overview before we start - watch this!
02:52-11:05 Ground work in ChatGPT
11:05-14:41 Using Microsoft Word
14:41-17:24 Focused Coding
17:24-20:59 Developing themes
![[ GUILTY GEAR STRIVE: DUAL RULERS ] Teaser Trailer](http://i.ytimg.com/vi/e4TDqaE3XAg/mqdefault.jpg)
👉 My Ebook on how to use ChatGPT for thematic analysis (including a list of GPT prompts) payhip.com/b/KmzOL
This got me started at coding a new project. Easy to follow.
Thank you for taking the time to share this! Really glad I could help
This is great food for thought! Let me share some reflections: I think that going through the transcripts and reading, coding, re-coding, etc. is actually the most fundamental part of the whole qualitative process when coding is involved (there are some other qualitative methods which do not involve coding), and using AI sometimes seems like encouraging the researcher to skip such coding process, getting directly to the results (that should be checked, of course). A diligent or curious researcher could also go through the traditional deep reading-and-coding process and THEN use AI, but it would look like a double effort. I think the process is more important than the results, because it brings you a better understanding, which is crucial for developing themes, besides guaranteeing high quality codes. However, maybe the time saved with AI can balance the cons of skipping the hard part and its benefits. So my questions for you: 1) What do you think about the reading-and-coding process with AI? ; 2) How much time would you save using AI for a high quality thematic analysis (following a robust iterative process such as that outlined in your recent e-book)? Thanks for your attention and congratulations for your work.
I agree with what you said - by reading the transcripts by yourself, you are developing understanding of your data that will also contribute to the subsequent iterations of your codes and, finally, theme development. I can still imagine doing this without having read the transcripts, and relying solely on the codes, however - for if these are good, descriptive codes, in principle they are essentially "summaries of the data" as I often say. This is provided, however, that you can trust ChatGPT completely (that it indeed generated extremely accurate and detailed codes), which of course is the main concern - can we trust it that much? In short, at the moment I would definitely NOT trust ChatGPT to do the work for me like this, unless it is a very small dataset and I can easily control what it does, but as you said, what is the point then? I can definitely see it as a useful second coder, as you suggested above. In the future, I do think that (very soon in fact) we will be able to hand our data over to such tools completely and not worry about our own understanding at all - because if you think about the goals of data analysis, it is not for us to understand the data, but rather to answer the questions about the data that we have. We only need all these processes (coding, subsequent coding, etc) to make sure that we manage this amount of data with our simple human brains :) When we can trust AI completely, we won't need to ask it to create, organize and list codes - it will be enough to upload the data and ask the research questions. I believe we are very close to this point at the moment, possibly even there already, and the only reason we are not Officially there is simply that academia is not ready to accept the fact that this aspect of research has changed forever
@@qualitativeresearcherInteresting perspective! It seems like you are saying that reading-coding-recoding is a means for an end/result (answering the research question) and I agree with that. Anyway, I also think that such process lead to different results if done manually, allowing reflections and research questions to develop inductively and abductively, thus changing the end/result. As they say: the journey sometimes is more important than the final destination (i.e. you learn new things and know better where to go once in the field yourself). Furthermore, contextual / personal knowledge of the researcher makes the process of qualitative analysis essentially different from what happens in quantitative analysis, making AI contribution inherently different. But that's a story for another discussion. Thanks again!
Jumping in quickly here (have not watched the video yet) but as a business school academic, this is the way forward. However like you I suspect there will be some degree of "expertise" loss, similar to knowing how to manually calculate the square root of large numbers vs. asking your phone. LLM will shortcut a lot of our accumulated expertise, but just like the @qualitativeresearcher says, maybe the magic is not in the tools (coding) but gaining insight from the data analysis itself. Ok off to watch the video!
How can we do this when we must maintain confidentiality and our subjects have signed privacy documents? I've always been told NOT to input anything into GPT because it scrapes the data.
I have commented to this under another video - you would need to double check it, but I am pretty confident that no data is being kept in ChatGPT, OpenAI explains that the text you input into the chat is not used to train the model. privacy and data protection is therefore not an issue, but whether it is ethical to "share" the information that you promised not to share, with ChatGPT, will largely depend on the wording of the individual NDA agreement.
So this will not be red flag in Turnitin for plagiarism?
According to Open Ai, the data from these chats is not stored anywhere, which means that it wouldn't be flagged as plagiarism later.
### Detailed Steps for Using ChatGPT for Thematic Analysis Coding
#### Preparation
1. **Collect Data:**
- Gather all transcripts or textual data that you want to analyze.
2. **Set Up Tools:**
- Ensure you have access to ChatGPT (preferably ChatGPT Plus for uploading documents) and Microsoft Word or any text editor for organizing and color-coding your files.
#### Initial Coding
3. **Prepare Data for Upload:**
- If using ChatGPT Plus, prepare your transcript files for upload. If using the free version, prepare to copy and paste your data into the chat.
4. **Create Initial Prompts:**
- Develop a detailed initial prompt for ChatGPT. Include the context of your study but avoid sharing specific research questions or aims to prevent bias.
- Example Prompt: "This is a transcript from a study exploring teachers' experiences of remote work. Please generate detailed descriptive codes for this transcript."
5. **Upload/Copy Transcript:**
- Upload the transcript file to ChatGPT Plus or copy and paste the transcript into the chat.
6. **Generate Initial Codes:**
- Run the initial prompt with ChatGPT and review the output.
- If the output is unsatisfactory, refine your prompt and try again until you are happy with the detailed codes generated.
7. **Extract and Organize Codes:**
- Ask ChatGPT to generate a list of codes without quotes for easier management.
- Example Prompt: "Please provide the same codes but without the quotes."
8. **Document and Color Code:**
- Create a text file for each participant with both the codes and associated quotes, using a unique color for each participant.
- Create a separate file for the list of codes (focus coding file) and paste the codes into this file, maintaining the same color scheme for consistency.
#### Focus Coding
9. **Group and Refine Codes:**
- Manually group similar codes from different participants into broader categories.
- Use the color-coded focus coding file to track the origin of each code.
10. **Iterative Adjustment:**
- Adjust and refine the codes based on emerging patterns. Ask ChatGPT for adjustments if necessary.
- Example Prompt: "Based on these refined codes, generate more detailed descriptions."
11. **Review and Validate:**
- Continuously review the codes to ensure they are descriptive and cover all aspects of the data.
- Manually check the quotes to validate that they accurately reflect the content of the codes.
#### Theme Development
12. **Develop Themes:**
- Manually identify themes from the grouped codes. Ensure these themes accurately represent the underlying data.
- Example Prompt: "Based on these groups of codes, suggest potential themes that accurately describe the data."
13. **Manual Finalization:**
- Manually finalize the themes to ensure they comprehensively and accurately convey the data insights.
- Ensure that you understand the context and meaning of each theme by revisiting the original quotes and codes if necessary.
#### Reporting
14. **Document the Process:**
- Document every step of the process, including prompts used, decisions made, and changes implemented.
- This documentation ensures transparency and allows others to follow and validate your analysis.
15. **Create the Report:**
- Compile the final list of themes, supported by detailed codes and quotes.
- Ensure the report clearly communicates the findings and the process used to achieve them.
#### Final Review
16. **Peer Review:**
- If possible, have peers review your process and findings to ensure validity and reliability.
- Incorporate feedback and make necessary adjustments.
17. **Ethical Considerations:**
- Reflect on ethical considerations throughout the process, ensuring bias control, transparency, and validity.
By following these detailed steps, you can leverage ChatGPT for high-quality, ethical thematic analysis while maintaining control and transparency in your research process.
you've been busy :)
is this ethical to do ??
Of course it is - as I stress in the video, this approach specifically prioritizes research integrity. I am constantly reviewing what ChatGPT gives me, then structuring my codes and themes, etc - there is human involvement and ChatGPT only serves as an additional tool that helps us maximize our own effectiveness and save time - just like a calculator does. ChatGPT is not interpreting anything for me, not developing any findings, etc. - notice that I am only employing it for the relatively automatic and objective task of doing the initial coding, which is essentially summarizing the text
Thank you for your insight. You are really an amazing teacher. I love to learn from your videos and highly appreciate your valuable time and efforts to educate us.@@qualitativeresearcher