You've probably moved way past this, but just in case others have the same issue: I realized that even though it looked fine otherwise (in the Console and over in the Environment tab), my level labels weren't actually being applied. So when it was trying to subset by "Female" it couldn't find it but when I had it subset by "F" then it could. There's probably a difference in versions of R from 2017 to now. Instead I did rt_df$condition <- factor(rt_df$condition, labels = c("Control", "Experimental")) to do level labels and then it was able to create rt_f_df fine.

Manual scraping with a browser automation solution (like Selenium) is unfortunately the only option at this point. In the current version of the course (which I'm recording now for YT), I switched this over to Reddit, since their API is still open for research.

Interesting test! I picked a random item off of the IPIP and had this exchange: Me: What psychometric scale is the source of this item: Talk to a lot of different people at parties. GPT: The item "Talk to a lot of different people at parties" is likely sourced from a psychometric scale designed to measure extraversion, one of the five major dimensions of personality in the Big Five personality traits model. This item specifically seems to assess sociability and gregariousness, which are key aspects of extraversion. The Big Five model, also known as the Five Factor Model, includes extraversion, agreeableness, conscientiousness, neuroticism, and openness to experience as its dimensions. Within the extraversion domain, this item might be used in several different questionnaires, but it closely aligns with the kind of questions found in scales like the NEO Personality Inventory or the Big Five Inventory (BFI). These scales use self-report questionnaires where respondents indicate their level of agreement or frequency of behavior related to various statements. The statement you've provided is a typical example of how these scales assess an individual's level of extraversion.

It depends a lot on your specific application domain. Prompt engineering is the more basic/accessible skill, so I'd start there. There are a lot of generic resources for learning how to prompt engineer in specific contexts - for example, Ethan Mollick made this video on prompt engineering for writing: kzbin.info/www/bejne/nJO1hoWdeLumnbs Fine tuning is a more technical skill, as it involves providing a dataset to the LLM via its API. I would only try that after you get a good handle on prompt engineering.

@@richardlanders, thanks. I'll check that resource out! I've found many ways to make my workflows within I/O more efficient and effective with chatGPT, but I'm trying to find easy ways to pass this on to my current and future clients. Do you think this could work if I make specific GPTs with prompt engineering? Have you seen any examples of this?

@@daviddubin3176I think this is why there are so many custom NLP AI products out there right now - a lot of companies are trying to figure out how to do exactly that. Most IO approaches I’ve seen so far are of the “replace what we’ve been doing with automated versions” flavor, like the use cases I describe here, but I’d guess that will start to change in the next year or so. For now, I’d say that if content generation (in whatever form) is a major service you already provide for your clients, reducing costs by speeding up those services is the most obvious starting point.

@@richardlanders Good point. In the spirit of your talk - not hand over all control to the machine nor be a Luddite - I think taking an approach to the efficiency of AI mixed with the expertise of a Ph.D. in I/O psychology is the best I can do now. Thanks for the great info. I'll follow you for more great talks and hope to run into you at SIOP Chicago!

It absolutely contributes. A lot of that also comes from the heightened awareness IOs in the assessment/selection space have of legal risk/consequences. Startups and older tech-centered companies are generally much less concerned about that kind of risk (or more technically, they are more likely to assume it to be acceptable and manageable than an IO generally will).

​@@richardlanders That makes much more sense! Great insight, indeed. Thanks again! By the way, I was pleased to hear that you were talking about an interview app. This semester, I designed an interview app for my HR students via OpenAI API and Azure platform. Students are taking interviews with the AI chatbot and receiving some feedback from it. Then, they analyze the script to see how AI formulates the questions, what kind of interview questions it asks, and so on. I hope to turn it into a research project and that we have a chance to talk about it in the future.

Awesome, thank you! I don't know about monetization, but I am glad the content is helping people.

Your code contains "subject =" but the parameter you are looking for is "subset =". You can check the valid parameters for subset in R documentation with the following command entered directly into the console: ?subset

My guess would be that you didn't install git first? Or perhaps git is misconfigured. Once you download git from git-scm.com/ and install it, you'll also need to tell R Studio where it is. Sometimes it figured it out automatically and sometimes it doesn't. If it doesn't, go to the Global Options (Tools menu), open the Git/SVN submenu, and point "Git executable" to wherever you installed git.exe

It shouldn't have! If you go back and forth executing the lines levels(rt_df$condition) <- factor(c("Experimental","Control")) and levels(rt_df$condition) <- factor(c("Control","Experimental")), in the Environment panel, you should see the factor definitions switching back and forth: Factor w/ 2 levels "Control","Experimental": 1 2 versus 2 1

This is because you're using a later version of R than I am in the current version of this video! :) They changed the default behavior of read.csv() fairly recently. You can get the same behavior that I do here by adding a parameter to read.csv: stringsAsFactors=TRUE (previously, TRUE was the default, but the default is now FALSE)

@@richardlanders Thanks, that's good to know. Your course/materials are great by the way, super helpful for a PhD student out of his element!

You can find the full list here: datascience.tntlab.org/schedule-materials/

@@richardlanders Ahh, I missed that. Thank you very much!

Meta..

The course is entirely in programming R and only covers the programming of statstical procedures. I do assume you already for example know what an ANOVA and regression are. If you don’t, you’d be better off to take a combined statistics+R course or a stats course first if you want the stats content.

Richard Landers Thank you for your reply.

There are no generally accepted credentials; however, they are certainly credentials that you could get if that's what you were after. MOOCs (Coursera, EdX, etc) offer certificates for example and have many courses that you could take. There are also certificate programs that you can get, some of which claim more broad acceptance than others (e.g., pythoninstitute.org/certification/, bootcamp.umn.edu/). I think a problem you hit is that the breadth of programming, even within just R or Python, is so vast that there is no way you could be realistically an expert in all of them. Even in this course, I only spend 1 module on natural language processing, on machine learning, on web applications, etc., but each of these could be an entire course, or even an entire degree program. In the real world, people are also now usually not employed to do _all of data science_ but for example are hired as a "machine learning engineer." So I would recommend learning the fundamentals of R and/or Python in whatever way makes sense to you (like this course) and then identify which specific "advanced" skills you want to go deep on. Within one such area, you'll generally find more useful credentials (e.g., www.pce.uw.edu/certificates/natural-language-technology).

@@richardlanders Thanks for the reply! Could you send me an updated link from that UW website at the bottom of your reply? I got a 404 when I clicked on it.

​@@ryanhjelle6924 Weird! When I click the link, it also doesn't work, but if I copy/paste, it does! UW must be doing some sort of broken KZbin referral detection.

@@richardlanders HA! That IS strange. Thanks for the help!

It depends a bit on the specifics of what you're recoding from and to, but yes, if you're starting with character vectors (e.g., "Agree", "Strongly Agree") to numeric (4,5), then dplyr's recode() inside mutate() is an easy way to do it, e.g., my_tbl <- my_tbl %>% mutate(var1_numeric = recode(var1_txt, "Agree"=4, "Strongly Agree"=5)) dplyr will even automatically set the new variable as a numeric vector, if you recode it completely, so that you don't need to add a data type coercion step afterward (in the above example, var1_numeric will automatically become numeric despite being recoded from character).

Annaka Johnson This is it! By the end of the course (all 13 weeks including completing the projects on your own before watching the debrief videos), you will able to do basic natural language processing, write basic AIs, and deploy web apps to support your research, in addition to more basic skills, like creating reproducible data pipelines, managing data more efficiently, presenting results more effectively, and creating figures to interactive visualizations. It’s important to start with fundamental programming skills though and really learn the whole skill set! Don’t skip the basics or you’ll be lost forever.

betam yamralnat waw

Hopefully within the next few weeks! I had some unexpected life changes in April that are still distracting me but will get them done ASAP!

as.POSIXct() is very sensitive to perfectly formatted character vectors with actual dates, so if there is even a slight deviation in even one case, it won't work. So for example, if one of the dates is 2017-02-29, which isn't a real date, as.POSIXct() will fail. I didn't think there were any cases like this in this dataset, and it worked for me, but an easy way to get around this problem is to use the lubridate package (also part of tidyverse), which is much better at understanding and converting dates appropriately than as.POSIXct() is. In this case, you could use ymd_hms(), which will return NA instead of the incorrectly formatted date. For example: > library(lubridate) > myDates <- c("2017-02-28 12:00:00","2017-02-29 12:00:00") > as.POSIXct(myDates) Error in as.POSIXlt.character(x, tz, ...) : character string is not in a standard unambiguous format > ymd_hms(myDates) [1] "2017-02-28 12:00:00 UTC" NA Your example won't ever work, because you are coercing a character or factor variable (depending on how you imported it) into a number, which will always evaluate to NA if the character is not clearly a single identifiable number. So for example, as.numeric("2018-01-01") will return NA. The only situation where you'd want to use as.character with dates is if you _already have_ a POSIXct type - then you can use as.numeric() to convert it to a Unix timestamp. For example: > as.numeric(ymd_hms(myDates)) [1] 1488283200 NA

Richard Landers thanks so much! I will try it.

Sure, you find it here: git-scm.com/downloads