Speaker 1: So, in this tutorial, I'm going to be showing you how you can use the help of ChartGPT when analyzing and interpreting and reporting data using the Stata data analysis application. I have here Stata open on the right, and I have ChartGPT on the left. I'm going to use it as my co-pilot, as my assistant in everything that I'm going to do here when it comes to analyzing data. So, we've already covered something like this using SPSS on a video that's on the channel. You can take a look at that here on the link. And now, let's take a look at how you can do the same thing with Stata. The first thing is to provide enough context to ChartGPT, which we've seen that we could do that by providing, say, a number of rows in your dataset. If your data is less than, say, 200 rows, then you could provide all the dataset, just copy-paste it into ChartGPT, and ask ChartGPT to understand it, and then ask follow-up questions. Or you could simply provide information about the variables that you have. ChartGPT is smart enough to understand what kind of data you're collecting, and then you can ask follow-up questions. For me, I have a codebook, which is basically a table showing information about the variables I have in the dataset. And once ChartGPT understands that, then we're going to go ahead and ask questions. So my first prompt here, and I'm using GPT 3.5. If you've paid for ChartGPT, you have the ChartGPT+, then I very much recommend that you use GPT-4. It's more creative. But for the sake of everyone, I'm going to be using GPT 3.5. My first prompt is going to be something like this, where I'm saying, The following is a codebook for a dataset on the nutrition of under five children. Read and understand that, as I will need your help analyzing the data using Stata version 13. It's very important, since there are changes in software between new versions, it's always a good idea to say which version you're going to be using. So I'm going to be using version 13, which is what I have open on the right-hand side. But also, I have provided a link to the dataset in Stata version 13 format. You can download it and follow along with this tutorial. All right, so what I do is I press the shift key and press enter twice to enter a line or a number of lines. And then I can go ahead and paste in the table that contains my codebook. And once I do that, I send it. And of course, it says, sure, I can help you analyze the data using Stata version 13. Please provide a dataset. Of course, I don't necessarily need to provide the dataset. Once again, like I've said, if you have less than 200 rows, you can just paste the whole dataset. It's going to understand it. But if you have more than that, which I do have more than that here on the dataset, you can see we have 413 observations. It's not necessary to go ahead and paste the dataset. The codebook is enough for ChatGPT to help us. If you already have an idea of the kind of analysis you want to perform, then you can always just go ahead and start asking questions. In my case, I have no idea what kind of analysis I want to do. So I'm going to ask for insights first. So the next prompt I have is what insights can I extract from this dataset? Send it. As expected, ChatGPT has given us a bunch of ideas of the kind of analysis we could do. So for example, we can do descriptive statistics of individual variables. So you can calculate various summary statistics such as the mean, median, standard deviation, minimum and maximum of variables like household monthly income, child birth weight and child age in months. It understands those are continuous variables. So we could do that. It gives us ideas about relationship between variables, gender differences. You can investigate relationship between household monthly income and child nutrition outcomes. And finally, we have education and nutrition, where we can explore the relationship between highest education and nutrition outcomes. All right, great. We're going to start with the first one, the descriptive statistics one. So the next question I'm going to ask is for it to give me the commands that I can run for us to come up with that kind of analysis. So I'm saying, what set of commands should I use to run the descriptive statistics in point number one? Okay, let's remove this and I run this. And of course, it says to run descriptive statistics in Stata, you can use the summarize command and it has given us three different commands for doing this. Let's just try the one that we have so far. So for example, summarize HTMI says it should give us summary statistics for household monthly income. I'm going to paste that in and press enter. And indeed, I do have this. Let me actually make this a bit smaller here. So I do have the output. Let me try this again. Okay, this is much better. So we have the variable here, number of observations, mean standard deviations, and so on. So of course, we can go ahead and continue here. We can say maybe, can you come up with a command that gives me the summary statistics of the three variables at once? And of course, as expected, it says to obtain summary statistics for multiple variables, we can still use the summarize command. And this time it has written the correct summarize command for giving us the summary statistics of all the three variables all at once. So I'm going to run this, press enter. And that's for sure. We have all three variables, household monthly income, child birth weight, and child age in months. We have number of observations. We have the mean, standard deviation, minimum, and maximum. That is awesome. One of the suggestions we have from ChildGPT is we can do gender differences. So next, I want to find out the commands that I can use to explore the gender differences for child birth weight, including the chart. So I'm going to go ahead and run this. One of the suggestions we have here is that we can summarize the statistics by gender. So summarize CBW by S. I'm not so sure that's going to work. And of course, we also have the option for charts. Let's go ahead and try this one. And we're going to run this. And obviously, we have an error here that says option by not allowed. As you're running your analysis from the suggestions that you're getting from ChildGPT, you may run into a lot of errors. And the solution for that is for you to just grab the error message that you have and then go back to ChildGPT and tell it that we have this error. It should give you a different idea of a command that you can run. So what I did is I just went ahead and said I got an error and I pasted the error that I have this error right here. And it says, I apologize for the confusion. It seems there was a misunderstanding in my previous response. Instead, the summarize command does not have a built-in by option for grouping a variable. OK, so maybe we can use we can generate summary statistics by gender by using tabulate S, then summarize CBW. So we can copy this and paste it in and press enter. And voila, we do have the correct one here. It says sex of child, female, male. And we have summary statistics of child birth by each of these categories. This is actually great. And then, of course, it did give us a different kind of chart we can run. It says we can use a two-way scatterplot. But we had not had a chance to try out the first option that it gave us. So we had the boxplot. So let's try to copy this command here for the boxplot and run it. And we're going to wait for a moment. And we indeed have a boxplot that shows the differences of males and females by child birth weight. If this is the kind of thing you're looking at, then that's perfectly fine. But let's try again. We could grab this other option where it's saying we can look at that relationship using a scatterplot. And I'm going to throw it in and press enter. It does give us a scatterplot, but I don't think this is the correct one. So you might see here that as you are conversing with ChatGPT, you may have some errors. But you just have to nudge it to the right direction by pasting the error message or by just telling it that the command it gave you is not necessarily working the way you expect it. And explain what you're expecting. And it's going to give you a refined response that you may actually have lack with. Next, we can use ChatGPT to help us in interpreting the output that we have and seeing how we can actually report it. Let's go ahead and actually use this output that we have here. We're going to paste it in. And then we're going to say write the interpretation of this table here and report it in APA format. We're going to go ahead and send it. And for short, ChatGPT gives us this output. It says the analysis of child birth weight by gender yielded the following results. The mean birth weight for children was 2.61 kilograms and based on a sample from 108. In comparison, the mean birth weight for male children was 2.99 kilograms based on a sample of 215 males. Overall, the average birth weight across both genders was found to be 2.81 kilograms based on the total sample size. And it writes the interpretation. And here it's saying the analysis reviewed a statistically significant difference in the mean birth weight between male and female children. It gives us the t statistic and it gives us Cohen's d. We don't have that information here in the table. This is exactly why it's very important for you to see the output that you're having there and actually review it before you can just copy and paste it in. Because we do not have any information about t statistics here. It's also possible that ChatGPT has gone ahead to actually calculate the t statistics. However, I wouldn't necessarily trust the output yet right now. Maybe the best thing is to actually instruct ChatGPT to give us the command that we can use to run t tests. And then we can paste in the output and then we can have the correct information that we can use for reporting.