Kerrie Mengersen (@KerrieMengersen) is Distinguished Professor at the Queensland University of Technology in Brisbane, Queensland, Australia, and past-President of the International Society for Bayesian Analysis (ISBA) . Her research spans Bayesian statistics, computational statistics, environmental, genetic and health statistics
For more information on the Explaining Bayes contest submit your entry below and tweet @statsandstories with the #BetterBayes.
+ Full Transcript
Bailer: I’d like to welcome you to today’s Stats and Short Stories episode. Stats and Short Stories is a partnership between Miami University and the American Statistical Association. I’m John Bailer, I’m chair of the department of Statistics at Miami University and I’m joined by my colleague Rosemary Pennington, professor in the department of Media, Journalism, and Film. We’re fortunate to be joined today by guest Kerrie Mengersen from the Queensland University of Technology. She is a distinguished professor of Statistics, an Australian laureate fellow, and currently serves as president of the International society for Bayesian Analysis. Kerrie, Welcome.
Mengersen: Thank You.
Bailer: You know, Kerrie, we’d love to talk to you a little bit about some of the work that you’ve done thinking about remote sensing and the broad range of applications from thinking about agriculture, poverty, chemo-brain, treatments for Parkinson’s. So, let me start with a question. What is remote sensing and how could that possibly be useful for all this diversity of applications?
Mengersen: We can think about remote sensing in the broad sense as any kind of digital image and so that ranges from satellites through to brain scans, and cat scans and so if anybody has had an X-Ray on their arm, if they’ve broken it then all parts of their bodies then that can be thought of too as a type of remote sensing so when getting these digital images it can also be signals so we could have acoustic signals where we put a microphone into the forest and listen for animals, but it can also be then brain signals that we obtain from somebody who is undergoing an experiment or a brain operation.
Pennington: How do you enfold this kind of material into your statistical modeling so because I think a lot of people know what remote sensing information is and then have an idea of what statistics is and for some of them there’s not going to be an understanding of how you work these two things together. How do you enfold this stuff into the modeling that you’re doing?
Mengersen: Well let me tell you that through a couple of stories. So, if we’re using satellite data we can use that for a whole range of areas and for example the national statistics offices around the world through the United Nations, trying to use different methods for obtaining their official statistics and so instead of just using surveys and samples they want to be able to use other sorts of information and a rich source of information is through remote sensing so satellite data can tell us about agriculture, can tell us about crop yield and crop production, so imagine that we have a satellite image and its looking down on an area, we can use statistical methods and machine-learning methods to identify what kinds of crops there are growing and what the potential yield of those crops might be. So, we can do that then through a statistical model, a special model, that identifies different objects in that image and we can take the same ideas then to our X-Ray or to our brain scan and we can look at that image and we can identify in that image objects like tumors or bone breaks and so that’s the statistical aspects of that is coming up with the algorithms that allow us to identify features in the data, but then also to be able to use those features in our statistical models or monitoring or analysis of the brain status of a patient.
Bailer: So, you mentioned in a talk before we started this interview the idea of chemo-brain. Can you describe what that is and how that is detected by this type of remote sensing data?
Mengersen: Sure, so chemo-brain is a term for the cognitive changes that happen for somebody who undergoes chemotherapy. So, in a study that I was involved in for breast Cancer patients, we looked at the long-term memory, short-term memory, and executive function, before, after, and then a long time after their chemotherapy. And what we find is that people typically, their short and long-term memory and executive function will decrease markedly just after chemotherapy. So, this is called Chemo-brain, where you can’t find your keys and your short-term memory is short, and people are worried about whether they will recover. Well, most people do recover, but a small group of people don’t. So, this sample we found with three different groups of people. People who recover quite quickly, people who take a year or two to recover to their previous levels of cognition and then a third group that don’t recover for a very long amount of time. Perhaps we can use statistical models, because it’s a real issue, Chemo-brain, but also what are the characteristics of the people who may suffer from this long-term change, and then how can we manage that better? So, this obviously is a team effort. It’s not just statistics. Statistics play a really big role in it, but it’s also medical practitioners, and psychologists, and a whole group of people working in the area.
Bailer: Wow, so there’s just monitoring this through various brain scans so you can look at the impact on the brain following Chemo and then the recovery post-Chemo?
Mengersen: So, it’s through brain scans and also psychological tests, and also there’s symptom data, so there’s a whole range of data that we need to be able to combine in our statistical models. Bailer: And then you’re looking at predictors of that? So, the ages of someone receiving treatment or some other characteristics that might be important for predicting this recovery? Or the impact?
Mengersen: That’s right. Exactly. And so this use of remote sensing data gives us direct observations but we also have the scans of people and other information that we can combine for this. So, we’ve used similar kinds of brain scan data and brain image data and brain signal data and symptom data for looking for example at the Parkinson’s disease as we know is a very debilitating neurological disease that is characterized by the shakes. And these shakes can stop people from having a normal life. So, one of the treatments for this is deep brain stimulation where they put a probe into the brain and the signal from the probe can counter these shakes. So, the problem is that this works well for a group of people but not for other people. So, one of the characteristics of those people, so we had a neurologist who came in and said “I’ve got all these records of patients that I’ve been treating but I know if I sit across the table from somebody who is going to respond well, but how do I translate that to other people and how do I learn from this big pile of records who this treatment is going to be effective for.” Well that’s a statistical problem, a problem for statisticians and so we can use brain scans, we can use brain signals, we can use symptom data to tell us that. As a statistician, I found myself in an operating theater looking at brain signals, brain scans, the symptom data and watching a person have a probe put into their brain and when it was put into the right place in the brain suddenly their shaking stopped and they could open their hand and say ‘Wow, I haven’t been able to do that for many years.’ and the next day they could walk and they could think about going back to work and it was amazing. That works for some people and doesn’t for others and that’s that statisticians job to work out who’s in which group.
Bailer: Remarkable work. Well Kerrie, it’s been our pleasure to have you join us on Stats & Short Stories. Stats and Stories is a partnership between Miami University Department of Statistics and Media, Journalism, & Film, and the American Statistical Association. Stay tuned and keep following us on Twitter or Apple Podcasts. If you’d like to share your thoughts on our program, send your email to firstname.lastname@example.org and be sure to listen for future episodes where we discuss the statistics behind the stories and the stories behind the statistics.
Pennington: You know John, since we spoke to Kerrie I’ve been having a really hard time explaining what Bayesian analysis is.
Bailer: You know Rosemary I think you’re the only one.
Pennington: That’s definitely a lie.
Bailer: I’m sure that others struggle with the same challenge.
Pennington: So how do we get people to explain it in a way that’s understandable to people who don’t study it?
Bailer: Well this sounds just like an opportunity, Rosemary. I wonder if the Stats and Stories podcast could do something to answer that question that you’ve just posed.
Pennington: Well we have been considering launching our first contest.
Bailer: Well with that really slick website we now have...
Bailer: …and looking at that you might even be able to submit something with this new website. Perhaps an entry in a contest. I think a contest to challenge someone to see if they could explain what Bayes is better. We might even call that with a hashtag #BetterBayes.
Pennington: I like this, so what are we going to ask them to do exactly?
Bailer: I think it would be neat if they could maybe write a story headline and the lead for such a story and that lead would be a sentence of less than 30 words, describing what Bayes is.
Pennington: I think since Richard isn’t here, I’m the journalist at the table. I like this idea and it seems like a nice way of melding the stats and stories together.
Bailer: Yeah because we’re going to see if you can explain this to your grandmother what this is. So what’s a good headline look like?
Pennington: You know it’s short and it’s sort of a quick summation of what a new story’s about and then the lead, that first sentence, basically captures the who, what, where, when, why, so all of the nuts and bolts of a story. That way if someone is grazing the new story they can get what they need from those two things if they aren’t going to finish reading the story, so it’s everything you need to know, the nuts and bolts, the very basics of the story.
Bailer: It sounds like an abstract boiled down to the essence of what’s going on
Pennington: That’s exactly what it is.
Bailer: Well are you ready for that folks? I mean this is the challenge that we’re posing to you
Pennington: Right. Go to statsandstories.net where you will see a form where you can submit a headline and a lead explaining what Bayesian analysis is. Again, the way you might explain it to your grandma, to maybe your journalist colleague who has no idea what it means. We’re not looking for a deep explanation, it’s got to be the nuts and bolts of what Bayesian analysis is.
Bailer: And we’re going to be looking at these. And actually we’re going to get a journalist and we’re going to get a Bayesian specialist, a statistician, and they’re going to be reviewing these entries and what we’re going to do is we’re going to identify a winner and they’re going to have a chance to chat with us and perhaps the judges of this event, talking about what kind of insights you had and what led you to describe Bayes methods the way you did.
Pennington: So beyond becoming the first winner of the stats and stories contest and being crowned the authority on better Bayes. What else are we going to offer them?
Bailer: Thank you very much Rosemary for asking and the winner of this incredibly exciting opportunity and competition will be a year’s membership in ISBA, the International Society for Bayesian Analysis, I know be still your hearts this is, what a neat opportunity and you will have the adoration of all of the stats and stories listeners. What more could you ask for Rosemary?
Pennington: Not much John, not much. So, the contest is going to be open for a month. We’re going to close it on October 6th and then we’ll take a couple of weeks to look through all of the entries and decide on a winner. Then we’re going to call you and talk to you and learn about how you came up with your definition.
Bailer: So, get ready. We’re looking forward to seeing what you think!
Pennington: And you can share your definition on twitter if you want to, using the #BetterBayes hashtag. But importantly, if you’re going to submit an entry to this contest you have to do it at the stats and stories website, statsandstories.net
Bailer: That’s it! We look forward to seeing it!