Transcript#
This transcript was generated automatically and may contain errors.
Hey there, welcome to the Paws at Data Science Hangout. I'm Libby Herron, and this is a recording of our weekly community call that happens every Thursday at 12 p.m. U.S. Eastern Time. If you are not joining us live, you miss out on the amazing chat that's going on. So find the link in the description where you can add our call to your calendar and come hang out with the most supportive, friendly, and funny data community you'll ever experience. Can't wait to see you there.
And I will say that I am so excited to introduce you to our featured leader today. He's a whole lot of fun. His name is Dwight Barry. He's the Principal Data Scientist at Seattle Children's Hospital. Dwight, it's so wonderful to have you here today. If you could introduce yourself, that'd be great. And also tell us something that you like to do for fun.
Thank you, Libby. Good morning, everyone. Good morning from Seattle. Happy time zone to wherever you are. What do I like to do? Hiking is probably my favorite thing to do. You know, I just live an hour from the Cascades, so I get to go hiking a lot.
You live in the perfect place to do that, to have that as your fun thing.
Yeah, I'm just 20 minutes from the beach, then I can go kayaking. So yeah, being outdoors is pretty awesome, even when it's wet.
So yeah, what do I do? What do you do? And what does Seattle Children's do as far as the data science function that you work inside?
Yeah, so Seattle Children's started over 100 years ago. And we've expanded from a small hospital in Seattle to being the primary pediatric hospital for all of Washington, Idaho, Montana, and Alaska. So not the biggest population, but definitely the biggest geographic area of any children's hospital, pretty much anywhere. And we got into data science fairly early on for hospitals. This was in the mid-teens. And I've been working with, I guess, data science is such a weird thing, because we all call it different things and mean different things, but I use it inclusively about any sort of advanced analytics, whether it's biostats or machine learning or AI, whatever. But the machine learning part was fairly new for health care in the mid-teens. And we started doing that, yeah, it was about 2015, 2016. And we gradually expanded to now we have a team of eight. And we have probably 90-odd people in our data analytics community. So it's fairly large.
Oh, and for those folks who are interested in sort of the analyst angle of things, I think we've got three positions open right now. So if you go to seattlechildrens.org, and look for our careers page, and then type data into the search, you can see those. And connect me with me on LinkedIn, and I can talk more about it.
That sounds amazing. Thank you for sharing. Maybe Isabel or Rachel can behind the scenes go find a career website somewhere, and we can share it in the chat with everybody.
Day in the life of a data scientist at Seattle Children's
All right. I think that this is a great spot to ask one of my favorite questions that Natasha usually asks, which is a day in the life question. What does a day in the life or a week in the life look like for you that could help us build context around the data science job function? So that might be like what tools do you work in? Are you an R team? Are you a Python team? Where does your data get stored? Do you deploy things? What do they look like? Do you build things for other users internally? Yeah, I'd love to hear.
Well, the short answer is all of the above. I am primarily an R user. I've been using R since the aughts, back before I even knew what the tidyverse was. But we have other folks who are primarily Python users, and we have our data warehouse hosted in Google Cloud Platform. And we're an Epic organization, so we also have the chronicles and clarity and all that stuff for the various data sources within Epic. We do anything from building a plot for a doctor to setting up a machine learning pipeline to predict our inpatient census every few hours. An Epic for everybody is an EHR, which is an electronic health record, just so we make sure we get all of our acronyms and stuff out there.
Dwight's winding road to healthcare data science
I was wondering also if you might want to touch just a little bit on your background for us, because I think it is super interesting where you came from and how you got how you navigated your way into this healthcare data science space, because you are an ecology guy, right? And that's where your stats background actually came from and got developed. Could you tell us just a little bit about your windy road to data science and healthcare?
Sure. My undergrad degree is in philosophy, of all things, and I decided towards the end of my undergrad that I didn't particularly want to do philosophy professionally. So I got enough hours to have a minor in natural resource management, and then I got a master's in forestry, and then a PhD in environmental science. And after my PhD, I ran a wildlife refuge for about three or four years, and then after that, I became a research professor for about a decade. But along the way, well, in ecology, you end up having to learn a lot of different statistical tools to address all of the very questions that come up. And so, you know, starting with the basics, everything from a t-test all the way up to crazy ordination and cluster analysis, and this was way back when when it was hard to do those things back in the 90s.
And when I became a professional, a lot of my colleagues realized that I had these stat skills, and they didn't want to develop them. They're like, oh, let's ask Dwight. He'll do the stats for us. So I ended up learning a lot of things on other people's projects. Like, for example, I helped with a soil microdiversity and chemistry project in Costa Rica, where I got a free trip to Costa Rica, which I guess I didn't need to do because I was just doing the computer work, but I appreciated the visit.
And yeah, and then I guess in 2011, I moved into health care. And the reason for that is that in the late aughts, my son was born with some serious health problems. And we were coming over to Seattle, basically weekly, to Children's Hospital, Seattle Children's. And after a while, I was like, you know, after a while, I was like, you know, I have these quant skills, I would really like to use them to help kids like my son. So I took some SAS classes on the side. And because SAS was primarily what was in healthcare at the time. And I learned SAS and got into got into a company called Group Health, which has been bought out since then by Kaiser Permanente. And so it's Kaiser Permanente Northwest. And so at the time, I really wanted to get into children. But even my son's doctor was like, you don't want to work at Children's, our IT is terrible. And so, but I tried anyway, and back then you had to have three years of healthcare experience before they'd even consider you. But I got into Group Health, and I worked at Group Health for three years and one week.
I got into Group Health, and I worked at Group Health for three years and one week.
As long as it took. When I when you told me that for the first time, I just about was like jaw on the floor, because I know so many people who are like, yeah, my plan was, I'm going to go get this experience that I need to do this other thing. And then along the way of getting that experience, their focus would shift. And they would be like, Oh, well, I turned into a different person in those three years. And then I went and did something else. But you had real focus, you were like, I'm going to go get the experience that I need. And you still had a focus of I'm going to go put my effort where my heart is and do this thing. And I think that that is absolutely amazing. I was kind of tearing up as you were talking. I don't think I'm the only one.
Gaining domain experience you don't have
And I think that this is a fantastic example of someone persevering through the I don't have the domain experience, which is a question we get a ton on the Hangout, right? Like people asking, well, I really am interested in finance, but I don't have any finance background, or I'm really interested in insurance or supply chain, I don't have any background, how do I do it? Do you have any advice on that from a career career perspective? From what you did? Did you just go like, okay, well, I need this healthcare experience, let me go find some place that fits this bill, that's not exactly what I what I want, but really kind of gets me in the door or that I have just enough experience to do?
Yeah. So what I found living from ecology to healthcare was that the questions are different, but the tools are the same. And so, you know, just earlier this week, I was doing an ordination problem on healthcare data instead of ecological data. So all of these skills that all of you have are transferable. And the trick is to demonstrate how you can apply what you already know to where you want to move into. And so one of the ways I did that to move into healthcare is that when I got interviewed at group health, well, when I had several interviews before I got into healthcare, but at group health in particular, I brought artifacts of my work. So it wasn't just a website, I actually had maps printed out of some of the map work I'd done.
What else did I bring? I brought a few examples of analytics projects that I had done on paper so that they could look at them right there. And my boss at group health later on said, you're the only person I've ever seen do that. And it just struck me as something you should do because you want to not just talk to people when you're interviewing, you want to show them what you can do. So I think of, even if it's like a pet project that you did for fun on the side, bring some sort of artifact to the interview.
You want to not just talk to people when you're interviewing, you want to show them what you can do. So I think of, even if it's like a pet project that you did for fun on the side, bring some sort of artifact to the interview.
That's wonderful advice. Thank you for sharing. I think I saw some people in the chat saying, I really needed to hear this, that you can really push through that. I don't have that domain experience. I promise that you can everybody. I know that it's, it can be really intimidating to be looking at this big wall of, I need the experience to get the job, but I need a job to get the experience. And it's often so, so disheartening. But I would also refer you to like our episode of the Data Science Hangout with Hubert, who is in the crowd right now, who I'm about to call on, by the way, to ask a question live. Get ready, Hubert. Who was like a computer programmer, a computer scientist, a math person who made friends in this space of research and transplant research and stuff like that. And said yes to, can you help me solve this computer specific problem? And then moved into this space and didn't really have the background there before that either. But there's this power when you make friends and you are putting yourself in a space in a community way.
Audience Q&A: data quality challenges
Okay. So I said I wanted to call on, on Hubert and I absolutely do. Hubert, you had a question in Slido. I will put it in the chat for everybody to see. And I also highlighted it to the top so that you can see it as well. Okay. So can you hear me? Yes. Hi Hubert. Hi Dwight. So my question, what was my question? Oh, there it is. Yeah. Are there data or data quality resource constraints that affect your work? I've worked with data from children's hospitals, various children's hospitals around the country. And it's obvious that some of them do not have some of the informatics or other resources and constructing their data sets or quality of their data. Does, does that impact you at Seattle Children's?
Absolutely. Yeah. In so many different ways. One is just the end size. I mean, any single hospital is going to have a small in any way. And pediatrics tends to be smaller than adult hospitals. And so that definitely impacts what we do. We find that we really don't have the end to do machine learning very much on the clinical side, on the business side. Sure. I work primarily on the clinical side and some other things missing data because patients will come in and out to our specialists. If they can't get a specialist in the time they want to go see somebody else and then maybe come back later. And so we have a gap in the record.
Parents move all the time and their kids go with them. So they're lost to follow up.
And then I guess the biggest one that gets me, and it's the same problem for healthcare anyway, anywhere is that we have longitudinal data where decisions made along the way, change the trajectory. And that's so hard to both conceptualize and deal with statistically.
Yeah. So many sounds like problems that look just like problems I've seen in other sectors too, with customer journey in retail spaces and things like that, where you just have these big gaps between interactions and you don't know why you have a gap in interaction and there is no follow up and you're never going to be able to answer a question. I always think it's interesting that this is, there are so many examples of just how experience transfers to different places. And I wanted to call out in the chat as well, Richard had said, kind of as we were on our last topic, Richard will be joining us by the way, on the data science hangout on November 19th. He's from Hennepin County, but he said, I've been on interview panels and we always like specifics, examples and show and tell. If it's general and vague, the panel might be also struggling with that too.
I wanted to add to this and say, so few people actually do go to the step of having an example that if you do, you are automatically putting yourself in the top 10% of people who just tried, because there are a lot of people who are like, I think everybody assumes that everybody else is doing something that they're not actually doing. They're not actually putting the effort forth. And you might assume, oh, those 400 people who clicked that thing on LinkedIn applied for that job when really maybe only 20 followed through and actually applied for that job. And then maybe out of those 20, only one of them has a portfolio or an example or a link to click through to see work that they've done. So if you just decide to put the effort forward, you might be in the minority already. Thank you, Richard. We're looking forward to seeing you in November.
So if you just decide to put the effort forward, you might be in the minority already.
All right, let's head back over to Slido and let's grab another question here.
AI in healthcare
Because we have a question from Noor and Noor, there are a few here. Hold on. Wait a minute. Noor had three questions. I'm going to put one to the top here because you asked one about AI. And I think that's very interesting. I'm going to put it in the chat for everybody as well. I'm sorry. I was really enthusiastic. We love your prolific question asking. Anyway, this is regarding the question that unfortunately everyone gets. And I apologize for the clichedness. But we're unfortunately for better work in the age of AI, whatever that means. Have you felt a push to incorporate AI into your workflows from stakeholders at the hospital or not? Maybe not. And what are some challenges and or pitfalls with that, especially with the kind of delicate data you would direct with? Right. Good question.
So a few years ago when I think AI was at its peak of the hype cycle, I felt a lot of top-down not pressure, but not request either, more like a sentiment that, yeah, let's use AI. It's going to solve all our problems. And it was exactly what I heard in the mid to late teens about machine learning. So I was like, OK, not a fad. Let's just see how this goes. And so the organization moved forward with AI. And we have a private instance of Gemini. So we do have that on-prem. And it's HIPAA compliant. So we can use it for patient data. And none of it goes back to Google.
So that part's cool. But that's sort of where it stalled out.
So we can access Gemini through the web interface with prompts. We can't do it programmatically yet. So it's like, hmm. So I cut and paste a lot of code.
The pressure to use more AI, I think, has gone down a little bit as people realize that it does have the same problems in machine learning, that if you don't have a lot of in, it doesn't help and can even hurt. And then probably more concerning is that if you have too many conflicting advice to give the AI, the model is going to collapse back to the average. And if you're in health care, it's like, nobody's average. So we use it in a few areas in the hospital. Like we have something called clinical standard work. And a lot of hospitals have this. And what that is, is a set of instructions for doctors to follow when they have a suspected diagnosis. And it's basically meant to be the best available evidence on treatment pathways. And you don't have to follow it as a doctor. If you have reasons to diverge from it, you can. But we've put that into an AI now. So instead of slipping through thousands of pages of PDFs online, you can actually query it and bring it up. And our AI interface will actually give you references for each one. So it's not making stuff up in theory. But a lot of people are experimenting with it in the hospital. That's the only thing that I know that's actually being used, other than us writing code with it.
It sounds like the security part is very, very important in the health care space, making sure that things don't go back into a system to train, which is a struggle. But there are the different providers, different model providers, do work with people to be HIPAA compliant, as far as I can tell, and sign agreements. So if you have questions about that, for sure, ask around in the Discord if you're in the health care space, because there are some people in the Discord I know who have a little bit of experience about this, or with this, talking with different model providers.
Okay. Thank you so much, Noor, for your questions. And Noor, you had another one that was, I think, really similar to Hubert's. So I think it got answered, which was like, what are some specific challenges that you face in your role that might be distinct from others? So I'm going to mark that one as answered as well. Thank you for all your wonderful questions. I wanted to hop in really quickly, Dwight, and ask you if you voted in Fat Bear Week, because you did not mention this. But when you said that you used to work in the field, in the forest, you were tracking bears, right? Do you want to talk just a little bit about that? And what do you think about Fat Bear Week? There was also a question in the chat from somebody that I saw, just like a brief mention about the reintroduction of grizzly bears into the North Cascades. Any thoughts, bear thoughts?
Bear field work
Okay, so, yay, backpack. Yay, backpack. Backpack one, if anybody is curious about Fat Bear Week. Backpack also has the cutest story to get his name. Yeah, so between Stinson grad school, I worked in southwest Colorado. And at the time, and to this day, people are still wondering if there is a remnant population, a small remnant population of grizzlies in Colorado.
And so my job was wandering around the San Juan Mountains for a couple of summers, looking for them. And the only bears I ever saw were the butts because they were running away. Bears are mostly, and probably for good, lots of good reasons, scared of humans. But grizzlies, of course, can be quite dangerous. And they're not nearly as afraid of humans, especially in Alaska, where you're basically another option for lunch. But so yeah, it was basically, every time you saw some bear scat, which is ecological talk for poop, you picked it up, put it in a bag, recorded where you were, and brought it back for DNA testing. And we also characterized the habitat, you know, what good bear foods were out there in terms of the plants. And we did that over hundreds of square miles. I don't even know, maybe more.
And through all of that, the evidence was mostly equivocal. If you're curious to find out more, there's a book called Ghost Grizzlies by David Peterson. And he ended up writing three different versions of it. And I mentioned in passing in the second edition, I think.
But, but he, it's funny, if you look at all three, I mean, it's funny, if you look at all three editions, he changes his mind about what he thinks about it, about the possibility of grizzlies there. And Dave lives in Durango, he's been there for decades and decades, and is a local basically. So he wrote it from a local's perspective, as well as from an interested environmentalist perspective.
Awesome. Well, I didn't mean to derail us too much into bear data. But it's too much of a coincidence you coming on right at the tail end of Fat Bear Week. I will also mention Fat Marmot Week, which just concluded for the first time this year, which I enjoyed that as well and voted in. Everybody go Google Fat Marmot Week.
A recent data science win
We had an anonymous question that I thought was very interesting as well come up. And we have one from Frank Leyva, too, in the Slido. Everybody hop in Slido, ask your questions. And don't forget that you can upvote questions. So, Frank, if you are available to ask your question live, I'd love it if we could do that. Frank Leyva. Let's see if we can find Frank to unmute him. Maybe.
I think maybe the microphone's not working. Try again. Oh, not yet. Okay, I can read it. No! Okay, Rachel, you're going to read Frank's.
But it was, what is a recent data science win that you are personally proud of? Anything goes.
I like the anything goes. Anything goes. Could be work, could be personal.
Well, there's nothing personal because I don't do data science on my time off.
I don't blame you. You do a lot in your paid time.
At work, I've done several really cool and impactful things over the years. And the most recent has probably been a paper that we have in review in the journal Neurology right now. And so this is looking at a relatively new medication for stroke and pediatric stroke. It's really devastating unless you can get to it quickly. And when you are able to get to it quickly, there's a much better chance of having a good outcome. And the pediatric world usually lags behind the adult world, simply because of the sheer amount. And so, I guess adults become the guinea pigs for children.
I'm going to say this wrong. The newer med for helping stroke care is called Tenecteplase, I think. It's abbreviated as TNK, so I always call it TNK. And the paper we have in review is talking about initial use of this in children, which it's been used for a little while in adults. And the risk profile is fairly small, but it was completely unknown what it is in children. And so, we had 43 patients over three or four years. So, a fairly small N. But I just basically did a beta binomial Bayesian model, and we used a prior based on the adult risk profile. And so, we determined the estimated risk for these kids. And we think it's going to be as safe for children as it is for adults.
I just basically did a beta binomial Bayesian model, and we used a prior based on the adult risk profile. And so, we determined the estimated risk for these kids. And we think it's going to be as safe for children as it is for adults.
That is amazing. Frank, I'm so glad that you asked this question so that we could hear about this. I put a little link so that people can see what Tenecteplase is, and you did say it exactly the way that I would read it. So, I think you got it exactly right. And then I will also put a link to a chapter of Bayes' Rules, which is a great book. Oh, I love that book. Yeah, which as soon as you said the beta binomial Bayesian model, I wanted to share this. So, I'm going to put this in the chat for you as well. So, everybody can see it if you're curious. Yeah, Bayes' Rules is great. Everybody go check out Bayes' Rules, the book. Single best intro to Bayesian statistics there is. Yes, I'm a baby Bayesian. I'm very frequentist, and Bayes freaks me out. So, anytime there's a friendly and fun introduction to it or information about it, I think it's wonderful.
Handling difficult statistical suggestions
Okay, well, we had a fantastic anonymous question. Not sure who asked this, but it's great. I would love to read it aloud. It says, when I present my work, sometimes people suggest really difficult statistical analyses that are over my head. I don't have a stats background, so I never really know what to say or do in that situation. What would you suggest? Nor how to reply on this one as well. You can reply in Slido in case you didn't know. Nor says, ask them why or what their motivation for that suggestion is. Force them to think about why they are making that suggestion and talk about it more.
Dwight, as a person who doesn't have a super formalized stats background,
what advice would you give? Nor's advice is perfect. That's how you start it. You push it back on them to explain their reasoning. What I found, and this gets easier as you get older with more experience, if you've got a solid background of basic stats and probability, you can usually learn these more advanced techniques. It may take a little time, but it's possible.
I guess one example is a while back I had to do a partial proportional odds model, which I had never done before. So I had to do a lot of reading to see do a partial-proportional odds model, which I had never done before. So I had to do a lot of reading to see Frank Carroll, thank you, if you're out there. Yay, Frank Carroll! We love Frank Carroll. I love Frank Carroll. To learn about it. And then I had Jim and I help me with the code. And so that combination was pretty helpful.
Anybody who does not know Frank Carroll, lots of great books, but also I think Frank is on Blue Sky. You can go follow him. Somebody find him on Blue Sky or some books. I'm pretty sure I had a Frank Carroll book in undergrad. Oh, so many questions just popped up in Slido.
All right, any other thoughts on this one question? Since I interrupted you, I did the end girl over Frank Carroll. No worries. Yeah, the other, just real quick. Yeah, so push it back on them. Ask them the reasoning. Learn it if you have to. And think about the context between what they need to know and what they're asking for, because a lot of times a simpler approach will answer the questions they have. And if you can demonstrate that, you may not have to go to the more complicated technique.
Oh, funny that you say that because Zach said the same thing in the chat. Zach said, I agree with Noor. Something else to keep in mind is that simple methods are usually preferable because they're easier to explain. So be wary of using that complex model that the stakeholder might have suggested because they themselves may not even understand it or understand how important maybe model parsimony might be to making sure that all of the results of your analysis are even interpretable. Fantastic. Oh, our community is so good and smart. I love y'all. Yay. Okay. Let's hop back into Slido where there are so many questions now that I cannot even choose, but I saw that Olaya had asked a question.
Question on dream formats for scientific communication
Olaya, would you like to unmute and ask yours live? Let's see. Do I see her? Where are you, Olaya? There she is. Hey. I was having lunch, so that's why I didn't have video. Hi. Hi.
Yeah. So you mentioned the publication before, which is what made me think of this question and my massive conflict of interest because I work for a preprint server. And my question is, what would be your dream display for all the analyses and not just the final products with the graphs or whatever other end point you may have, but also sort of the process? Does that make sense? What sort of format? Is that what you're asking? Yeah. I mean, a normal publication now is we're on the online world, so it will be a PDF that is online, and that's kind of like the standard, right? There have been some experiments with the life, not life code. I don't know what they called it, but essentially like a notebook type format. So either that or even beyond, what would be your dream for someone who has done those analyses to communicate to a fellow scientist who could like really see what you did?
Way back in the heyday of Stats Twitter, which we all mourn its passing, I saw this absolutely brilliant tweet. It said it was on the research side and it was, how was it phrased? It was phrased like, what is a habit that you or your lab does for purely superstitious reasons? And somebody replied, publish in journals. And given the abilities of what we can do with interactive graphics these days, PDFs are stupid. So I saw a brand new journal. I think Alberto Cairo is involved with this. The Journal of Visualization, Data Visualization, Data Presentation, something like that. It's a brand new one. They've only been around for a year or so. And you can publish interactive articles there. And I think that's brilliant. I really like what they're doing.
And given the abilities of what we can do with interactive graphics these days, PDFs are stupid.
It has everything from source, the code that you can go get. It has the written part if you just want to see something static. And for those who want to share interactive stuff, you can. And so far, that's the best example I've seen.
Cool. I'll look it up. So it's something about Journal of Data Visualization-ish and it's new. Yeah. Okay. Great. Thank you. I think that there's a link in the chat from Aaron. Oh, good. Somebody found it already. Let's see. Journal of Visualization and Interaction. There it is. I actually had it on my browser. Awesome. You just happened to have it up. So sorry I was not muted when I cracked up at you saying what you said. I try to mute myself when I know I'm going to be laughing. Did not expect it. Okay. Thank you so much, Alaia, for your wonderful question. I appreciate it.
Exciting developments in data science
We are going to hop over to Zach's question. We have about 15 minutes left, and Zach has asked with a little asterisk, so I'm going to ask it for him. This is a fun one. What developments in data science are you most excited about, not including AI? That's a good one. It is. Putting you on the spot.
Yeah. I think the recent advances in Bayesian workflows is really cool. I think folks like Gelman and his group and the BMRS folks are doing some really amazing things on making Bayesian modeling as easy to do as machine learning. And that's Andrew Gelman, right? Yeah. Yeah. Just for anybody who doesn't know who Andrew is, maybe we can find a link to Andrew's website. Yeah. They have a new book that just came out that's online for free called Bayesian Workflows. Amazing. I was just going to link to the YouTube for Bayesian Workflows. I'm going to let somebody else handle that, though. And also, I will pitch the same thing that Dylan pitched in the chat, which is McElroy's.
He is great. He's just hard to get past the basics if you don't already have the basics. Yes. I would say statistical rethinking can be a lot to chew on if you are just getting into that. But I think that the more widespread adoption of Bayesian workflows and Bayesian approaches to problems is also really exciting. And I want to learn more. I hope that we can have some more Bayesian folks on the Hangout soon to chat more about it. Yeah. Yes. And thank you to everybody who's helping with links. You guys are amazing. Thank you, Nora, for the Bayesian Workflow link. Thank you, Zach, for the question.
Working with small sample sizes
Okay. Logan, I saw you had a question, and I know that you're here because I can see. Would you like to ask your question live? I'll put it in the chat. Yes. Hello.
So I work in higher education doing data for colleges and universities. And we have a lot of cases where we have small end sizes, students responding to course surveys or anything. And I've been brought on to a new team to help them build models and really try to use their data in better ways. And I wanted to know how you approach having useful outputs when you have end models, but also just in a simple data analysis, this is what happened kind of thing. Because, yeah, like I said, we just have students not responding to things. And I can't glean too much of, you know, one student said one bad thing about a professor.
Yeah. Yeah. Rate My Professor is pretty evil. The worst. We don't talk about Rate My Professor. No. We don't use it.
Yeah. My brief time in academia, well, I guess it wasn't too brief. It was about a decade. Yeah. Patient surveys are hard because they're optional. They tend to be, you get the whale distribution where everything's pushed up in the four and fives, and then you got a little tail and it pops up again at one. And the professor could probably name the ones. But how does it help at the population level? It's hard to say. One of the things I got to do when I was a professor was set up my own exam, my own surveys. And so the first thing I did was make sure that in a or no opinion was set off to the side and you had to go one way or the other. And so I only had a four point scale instead of something in the middle. And so then I could just look at the ones who had actual opinions and get a better sense for things and which way they're going.
But that was a long time ago, and I don't know how well that works anymore. I think a great approach now that you can do is because ordinal models are so easy to do in Bayesian statistics. And if you string that along with marginal effects, you can get some really good outputs that aren't stupid odds ratios. You can actually get probabilities. And if you set your priors reasonably enough and don't use the default flat priors, you can get some reasonable inferences. And you can say, OK, look, the credible intervals are really wide, but the balance of the of the posterior distribution is in this direction. So it can be directionally correct. Awesome. Thank you. Yeah. Thank you for your question, Logan. I appreciate it. I think there are other people in the chat as well who are in a low in zone where they just don't have a lot of people all the time. And so that's a that's a widespread question.
OK, thank you so much. I wanted to call some stuff out from the chat that was fun. And yet the chat is moving around so much that I think I've lost most of what I was going to was going to like point out. Oh, I'm so sad. Well, there was a hilarious thing that Richard said. Small ins for students evaluations, but they still hurt my feelings with a big effect size.
Hilarious. I will say I used my professor for I just used it in the opposite way in undergrad to pick professors that I would like, because they would have really low scores on difficulty. And if somebody if a whole bunch of people were complaining that a professor was really challenging, I'd be like, that's the one for me. I want somebody who's going to be really hard. And then I just learn a ton. And a big part of it is that I don't have a lot of time to other people would like drop the course. And I would just stay stand on my ground and learn a massive amount from a really challenging professor. Yes. Yeah. Student evaluations rough.
Interpersonal skills in data science careers
There is a question from Alejandro in the Slido that is fantastic for our sort of signature career advice question as well. And so, Alejandro, I see you just came on camera. Would you like to ask that live? Yes. Hey, can you can you hear me OK? No, Alejandro. How are you doing? I'm doing very well. I had a question for you because I just joined. OK, first of all, I got the job. Thank you for the recommendation. I appreciate that. So there's some career advice right there. So seeing as this is like my first statistical or official position in like the statistical workplace, I was wondering if you had any advice for interpersonal skills that has been that have been useful to you in your workplaces. And I think I'll say that, like anything goes with this one, you know, doesn't have to necessarily relate to your role as a statistician on your team. But I don't know, I feel like, you know, a lot of the time for young people and for students, like a lot of the emphasis is on the technical skill. And then maybe that helps you enough to get into a job position. And then you kind of have to like figure out the interpersonal on your own. So anyway, this is wondering what your thoughts were. That is a great question.
Well, so I guess just as a start, and this is this is sort of a
across-the-board advice. If you have an attitude where you always want to keep learning, that's the single most important thing, not just for technical things, but for interpersonal things.
I guess the thing, I was really shy as a kid. I think the thing that really broke me out of that was having to do presentations in college. So the more speaking I had to do, the more comfortable I got talking to a room. And of course, at the beginning, it was totally scary.
By the time I was out of grad school, I was like, I'm not going to do this. I'm not going to do talking to a room. And of course, at the beginning, it was totally scary.
By the time I was out of grad school, I was like, I'm totally confident now. You know, I still get butterflies, but I'm confident. And to the extent that when I was working at Group Health, my boss told me later, because I had talked back to a VP or something like an SVP. And she says, you don't care who's in the room, do you? I was like, no. But that's not great advice, though. I wouldn't necessarily recommend talking back to an SVP, unless you're really confident in your answer.
Conferences and community
Going to conferences is really cool, especially Posit Conf, you know, used to be our conf, our user groups, because the community is so wonderful, and so open, and so welcoming, that I hate networking. I absolutely hate networking. But I love being around our folks, because we just go and start talking to each other about whatever. And it's comfortable. And it's the only professional semi-professional group that I've ever been around that is like that. So that's really appreciated. I mean, like, things like JSM and those types of meetings are not like that. It's very clicky. And JSM is Joint Statistical Meeting, for anybody who is not aware. Usually in a different spot in, I think, the US or at least in North America each year. Someone correct me if JSM is frequently in Europe, I'm not sure, but I think it's mostly in the US. But I love the suggestion to just make friends, like, don't worry about networking, just make friends with people, Alejandro. Go be yourself. You seem like you are an absolute delight and just go make friends. I wanted to point out some advice from the chat from some very smart people, because Ellis Hughes is with us today. Ellis is a fantastic organizer up in the Pacific Northwest where Dwight is organizing Cascadia Com. Hey, Ellis.
And Ellis says interpersonal skills equals being comfortable reaching out to folks across the org just to learn about them is a massive unlock. Just to learn about them. What do you do? What do you like? What do you have going on? You know, how's the family? How are things going? What are your hobbies? What do you like to do for fun? Just get to know people as much as you can. Makes it easier the next time you need to reach out to them for something that is work related.
And Ellis says interpersonal skills equals being comfortable reaching out to folks across the org just to learn about them is a massive unlock.
And then you have absolutely unlocked a superpower within yourself, which is just being willing to ask for help. One of the biggest barriers that I have ever experienced in myself, in my own brain is going, I really want to reach out to this person and ask them for help. And I'm just so worried I'm bugging them. And I'll ask Rachel, Rachel, what should I ask this person for help? She's like, yes, just go message them on Slack or something. They're very nice. Get out of your own head about it. And then Gonzalo also mentioned something that I thought was very important to recognize. Gonzalo says, for me, the interpersonal is different since moving to Latvia because I'm shy as an American, but I'm too extroverted for a Latvian. And so now it's like learning about the cultural differences. Those differences can exist inside of organizations too. If you pop into an organization that's super, I would say more academic than you're used to, or less academic than you're used to, that difference can be a little jarring as well, or much bigger, much smaller. Thank you Alejandro for asking that question. It was great.
Cascadia conference
Yeah. I want to, with our last, oh, go ahead. With Ellis here too, we should give a shout out to the Cascadia conference. Heck yeah. I don't know if Ellis, you want to jump in and tell us a little about it. I would definitely be putting you on the spot, or if you want to share about it in the chat. But I just always hear so many good things about this conference. One of my favorites for the whole year. Yeah, I guess I can speak to it for a second. Cascadia is an annual regional conference that we run every year where we just try to have folks from the area. So from the Washington, Oregon, North California, BC, everyone's invited. And we want to just get people talking about stuff that they're doing in R. We try to focus on students, give them an opportunity to present. We're usually, I mean, the last several years we've been between like 120 to 160 people that are all just coming together on their Saturday to talk about R and be massive nerds together. It's a ton of fun to be together. The last few years we've run workshops as well. The day before, we really try to focus on making it very cost-effective as well. So that folks, if they have to self fund, it's not a massive burden. The other conferences just can be really expensive. And we try to run a lean ship as much as we can. Really, really great community. Everyone's incredibly friendly. If you are in the Pacific Northwest Cascadia area, we have a LinkedIn page. We have a Mastodon and Blue Sky. We're not the best at social media. We work on it every year, but we're just not that great at it. Join, follow them. And you'll see some updates for when we put out the call for papers. We're going to be in Seattle. Next June is our aim for a timeline. We're trying to figure out where we're going to actually host the conference inside of Seattle.
But yeah, please reach out if you're in the Pacific Northwest area, because we'd love to have you. Yay. It's so wonderful. I go to Cascadia every year, even though I am nowhere near the Pacific Northwest. We also have virtual tickets as well that we provide to everyone. And we do actively try to make it a very fun thing for folks that aren't in the area or for some reason can't make it physically in person. We do offer that virtual option as well. And Libby's been a fantastic advocate for us for years. I love it. Best little RConf in the US, I think, for just like a one day. You don't have to take off work to do it. It's on Saturday. It's so much fun.
Okay. Well, with our last two minutes left, I think that we have to thank Dwight for hanging out with us, for talking about bears, for giving us all of the advice about finding experience in an industry when we don't have any domain experience, about adopting Bayesian methods and learning all about stats and incorporating them when you are not a hardcore stats person from the get-go, talking about your windy career path. This was so much fun. I hope you had a wonderful time here as well, Dwight. I did. Thank you so much for the invite. And thank you, everyone, for the wonderful questions.