Prof Mary Towner - "The Potential to Infer the Historical Pattern of Cultural Macroevolution"
Duration: 26 mins 41 secs
Share this media item:
Embed this media item:
Embed this media item:
About this item
| Description: | A recording of Professor Mary Towner (Oklahoma State University) speaking on "The Potential to Infer the Historical Pattern of Cultural Macroevolution" as part of the "Cultures at the Macro Scale" seminar series. |
|---|
| Created: | 2021-07-23 11:08 |
|---|---|
| Collection: | Culture at the Macro-Scale |
| Publisher: | University of Cambridge |
| Copyright: | Mary Towner |
| Language: | eng (English) |
| Keywords: | Phylogenetics; Simulation; Culture; |
Transcript
Transcript:
00:03
Hello, I'm Mary Towner. I'm going to be talking today about the potential to infer the historical pattern of cultural macro evolution. I want to start by thanking the organisers of the seminar series "Cultures at the Macro Scale". It's an honour to be presenting this lecture here. I also want to start by thanking my collaborators and co-authors of the paper that I'm going to be mostly focusing on today: Dieter Lucas and Monique Borgerhoff Mulder. They're both at the Max Planck Institute for Evolutionary Anthropology and Monique is, of course from UC Davis as well. The work I'm going to be talking about is available already as a preprint—as well as data and the code that you can find on Dieter's website—which I've indicated on this page.
00:52
So cultural anthropologists have long been interested in understanding cross-cultural variation across societies. This might be in languages, in archaeological records, in cultures, the cross-cultural social behaviour, housing styles, marriage patterns... any kind of trait that might be measured about societies. I know from glancing at your seminars from last semester that you've already seen this tree picture: Kroeber's depiction of what cultural evolution might look like on the right side, in contrast to his interpretation of organic or biological evolution on the left side. So with cultural evolution, we have the potential to maybe have more intertwining between societies over time in history. And we might be interested in how traits are transmitted both through vertical means, from parents to offspring or parent generation—parent societies—to descendant populations and societies. And also through horizontal transmission from neighbouring groups, for example.
02:11
So my work in this area really began with time at UC Davis as a postdoc with Monique Borgerhoff Mulder in the early 2000s. And she and others had already been working on this for a while. The way I came into this was through review—with Charlie Nunn as well—where we were looking particularly at some of the ways in which phylogenetic comparative methods may not always be up to the task, in certain respects, in trying to understand cultural macro evolution. So for an example, this box on the right depicts two phylogenies from a small group of East African pastoralist societies and maps on a trait: which is the marriage pattern (whether polygyny is rare, or extensive as shown in the darker bars here). On the left side, you can see the depiction of the phylogeny has one fairly older origin of extensive polygyny, that then is passed on to a number of descendant groups that are closely related to one another. This would be through vertical transmission. Now an alternative mechanism could be that neighbouring groups observed each other, transmitted the kind of cultural trait of extensive polygyny between groups living in close proximity to one another. And there wasn't really at the time a good way to tease this apart without having brought in other kinds of collaborative, corresponding, corroborating data. So this is just one example from that paper that reviewed a lot of detailed critiques and evaluations of methods, in addition to phylogenies used to understand cultural evolution and the coevolution of traits across time.
04:27
Now, from this jumping off point, I began working on a project with Monique, and also Mark Grote at UC Davis, to better understand cultural macro evolution using a data set from Western North American Indians. This database, compiled by Jorgensen, but building on the work of many others before him, looks at 172 Native American societies from the west coast of the United States, up through Canada into... going east to New Mexico. And it catalogues a whole variety of traits. Some that might have to do, for example, with pottery style—this is a picture of Santa Clara Tewa, at a Pueblo in New Mexico making pottery. There might be traits having to do with social organisation, with diet, with types of housing, with residents patterns. So we looked at a number of different traits. And we... our goal with this work was to try to figure out if there was a way to look at vertical and horizontal transmission on a level playing field. So rather than prioritising vertical transmission on a phylogeny, would we be able to somehow model a trait across both the historical relationships between societies (what we might use a language for) to understand vertical transmission, but also to better model a trait bringing in the potential at least for horizontal transmission—for example, through societies being close geographic neighbours.
06:29
So to just give you a taste of what we did in this work (and this is again, with Mark Grote, Jay Venti and Monique) we had a method of modelling that built on neighbour graphs where this is... you can think of it like a matrix that captures pairwise between all of the societies, whether or not two societies were in the same language family. So they would be in the same kind of neighbour graph for language if they were in the same group. And also then, trying to do the same thing in a matrix kind of definition—a graph of proximity, geographic distance between groups. So the figures that you see here are more networks, in a sense. (We're on the left under a). This depicts the relationship geographically between societies if they were within a certain distance from one another, and on the right, looking at whether each pair of societies was in the same language, family or not. So long story short (and this is again, published in Human Nature 2012) we found that for certain traits, models that included both of these components—both language and geography as stand ins for vertical and horizontal transmission—those models outperform models that just included one or the other.
08:11
Now, this method didn't rely on phylogenies at all. But it was a way for us to explore just how much might be overlooked if we only assumed vertical transmission of traits. Now, there are other methods out there that have been developed and explored for modelling cultural evolution without solely relying on phylogenies. That said, phylogenies appear to be here to stay—1000s of publications by now. Here, this phylogeny in particular is just meant to illustrate a recent publication. This one is looking at the correspondence between housing size and post-marital residence, and so Hrnčíř, Duda, et al. in 2020 (Plos One) published this paper to look at whether or not there was a correspondence between having smaller or larger kind of floor space and in the household size and residence pattern: whether women or men went to move into. If it's patrilocal, say, wife would move to her husband's natal household, and matrilocal, the husband would move to live with or near his wife's family's household. And this seemed to suggest a pretty strong relationship between the two variables. So one of the goals with phylogenetic analyses, sometimes, it's not just to look at the history of one trait, but to try to look at coevolution between two traits. And again, this is just one example of many I could talk about right here.
10:01
So we still had some some questions, though. And one of the things (because of another workshop on cultural evolution that I was invited to and later, we were brought on to contribute a paper for this) was to ask, you know, do we still worry about phylogeny? So we may have raised some of these issues 20 years ago or more, but are we still concerned? And I think we brainstormed for a while and came up with a list of questions that we could still say are a bit vexing. (And again, these have been explored with others, and you can read more of the the full bibliography in our paper). But just to show you some of the questions (I'm not going to read this whole thing), but there's still some vexing questions that we have about phylogenetic analysis.
11:02
So for example, what i... to what extent can the history of the trait be reflected in a phylogeny? Is a phylogeny that doesn't show horizontal transmission, for example, going to capture that history? Th.. another question here is whether traits might go together—coevolution, perhaps—or have independent histories? How do we model that? How can we reflect that on a phylogeny? Other questions might be whether the the scale—in terms of time of the phylogenetic tree—is really appropriate for the traits that we might be looking at. So if traits are very flexible and change rapidly, the idea of even needing a tree may no longer be quite appropriate. Because if there's so much flexibility and change in traits, it may just be that we need to look at very recent time and not try to recreate a quite older past. Other questions include whether the tree itself is pretty solid? Is it built on appropriate kinds of data? And is it at the right kind of scale? How, how resolved is the tree? And at last, we have questions about the sample size and missing data; you know, is this really sufficient? For example, do we have enough societies to make large inferences? So if we were working with a very smaller sample, say, of societies—and going back to the example of marriage patterns in East Africa, that particular figure was a smaller group—is that.. how does that sample size hold up? Another example: what about missing data, or if we had a number of societies that had traits that are important, but those are extinct? And unlike some kinds of records, like archaeological records, with behaviour with social behaviour, we may not be able to quickly or correctly reconstruct or make inferences about missing data or data of extinct societies.
13:33
So with this in mind, we set a goal for ourselves to look at these kinds of questions and Dieter's idea was to use simulations on existing phylogenies as a tool—the phylogeny wasn't to be our focus of examination per se. And the simulations used arbitrary traits, not real traits. But could we learn more about the robustness of different phylogenetic analyses, if we varied certain things like the rate of change, or the direction of change, or if there were missing data or missing parts of the tree?
14:21
And an analogy that they came up with in particular—I kind of watched it unfold—but if we look at the past as something that we're trying to get a clear picture of, we're looking through our current window trying to get a clear picture of the past. We are trying to recreate that past somehow. And you might imagine a painter with paint brushes and an easel full of colours trying to get a (somewhat) accurate understanding or reflection of the past. But it's not a direct record, it's our our inferences about it.
15:05
So taking this analogy (I'm going to come back to it) the... very briefly, the methods were to use some known phylogenies: one from the western North American Indian database (the one I talked about before), another more recent one from Pama-Nyungan languages. Looking at those two different phylogenies. And then the picture you see on the left is just a little sub-sample here, taking a random trait, in this case, just go by colour—red or black—if we simulated the distribution of that character, under different conditions, and then we know what the actual history would be (so the actual history is what was actually simulated). And then, we use the methods of phylogen... the phylogenetic approach to try to infer that history, and then we can look and see how well those line up.
16:08
So for example, this is just one of many, many examples that we go through in the paper. Let's say you we were exploring direction of change, what we would do is we'd have the actual history through the simulations. And let's say that the trait changed randomly between two states. Under the simulations then, we look at the inferred histories. And if the trait actually changed randomly, but we wrongly inferred that certain traits changed more or less frequently than others—that certain directions were more or less common—we would call that a "Type I" error that were wrongly inferring history by... with a kind of a false positive here: that there was a change that was more or less frequent, with that direction. Now, kind of looking at this again, but from the point of view where the actual history as the selection favoured changes between certain states over other changes. If we were to, again, then wrongly, infer that all changes were equally likelihood that would be a "Type II" error. So a false negative that we think that there's no real pattern there, when there was some kind of underlying pattern.
17:38
So what we did then, (and really, again I keep saying this, but Dieter did so so much of this coding and simulations under different characteristics), we looked at continuous data and discrete traits. I should say... so let's take an example here, a trait might be something like "type of marriage" (polygynous, or monogamous). Let's just say a continuous trait might be something like the proportion of marriages in a society that are monogamous. Now, those aren't specific traits we modelled again, we used just as a sample, a continuous or discrete trait, as like red or black, for a discrete trait, and then a more continuous value for the continuous tree.
18:36
So for each of these... then I'm... we did a lot of different questions and a lot of different variations of this, I'm just going to highlight a few of them. But the big picture result that overall—just to preview that—is that these errors are very common under all kinds of permutations when we create through simulations and actual history, and then we modify things and compare the results. There are lots of Type I and Type II errors in the findings. So as an example (and going... tying it back into the questions), what if, for example, the data that we have are missing or wrongly classified. So let's say the actual history had three colours red, purple and blue. But the inferred history then only had red and blue. That would give us an inaccurate picture of the past—our inferred past would be missing the colour purple. You can also look at this as if, for example, the purple societies were no longer there—had... had been extinct—we wouldn't have any way to recreate those.
20:06
So the analogy, then, here is that we would be trying to use either the wrong colours or not have enough colours on our palette, as we're trying to paint the past and make an accurate portrayal of the past. So high error rates in different categories across either phylogeny (so it varies a bit what we're looking at), but, but there's enough there to be of concern.
20:39
So let's look at another kind of question. What if there's horizontal transmission of traits? So again, two colours in this in this depiction, red and black. In the actual history, let's say that red has made a jump over to a neighbouring society. The inferred history— based only on phylogenies—that don't take that into account would come up with an inaccurate picture. So you can kind of think of this as perhaps painting with the wrong brush. So maybe the brush that you're using isn't fine enough to pick up those fine details. And again, high error rates across the board really in different categories. In this case, were really missing out or assuming that something might have been inherited from an ancestral population, or when in fact, it It came from a neighbouring population.
21:46
So one more example here, that we looked at was what if the structure of the tree itself matters? Now an example here is where there's a late burst phylogeny. So, if we went back and deeper time, much time would go on without much change here (what we're seeing), and then there's kind of a quick fluorescence of societies that have variation in a trait. And this particular picture—on the left the actual history—you can kind of see where the red is, that the ancestral node is red. But if we looked at inferred histories, we might make different inferences about those ancestral nodes. If we were thinking that this wasn't just a very recent burst now.
22:48
(Whoops, let me What did I do here? Hold on a second.)
22:54
Okay, um, how does this... How does this kind of fit in with our analogy? So again, high rates of errors across the board, but this could be a case where maybe the canvas that we were painting on was really coarse, so that we weren't able to have as much fine detail as we needed, as we recreated our understanding our inferences about history.
23:26
So what do we do? And this is the the checklist, I'd say, that we suggest. But I'm going to say that it's just the start of one—and there are many other ideas about this. One is to make sure that our research questions are informed by good theoretical models to start with. Make sure that the trees that we use are at the right time scale, and built on solid and diverse sources of data, considering how traits are transmitted, and in particular, the potential importance of horizontal transmission, and what that might do to some of the inferences from just phylogenetic methods that don't take that into account. One of the pitfalls that may come from ignoring some of these problems is that the internal nodes that are reconstructed—even though many researchers will say that the exact value of those nodes isn't what matters—that there's a robust kind of outcome regardless of what an internal node has been estimated to be. But that often may get lost in the details where then it's easy to want to try to make inferences about those internal notes. I mean, there were curious about these things, what was the trait in the past—so just to be aware that sometimes based on what what particulars in our tree may be playing out, we may not have as much confidence as we'd like to in those nodes.
25:06
And then the most kind of... the plug here, is that for any particular sample—any particular phylogeny—there may be some unique pitfalls. One way to explore those could be through relatively straightforward simulations on a phylogeny, where if researchers are exploring a certain hypothesis, they can simulate that with data that they create, and then look for where there might be tricky spots where they might be walking into situations where they would be at high risk of making Type I or Type II errors. Or where even if they don't find a pattern, for example, they really can't say much—with much confidence that that couldn't be there based on the intricacies of their particular sample or phylogeny. And again, one more push to check out our preprint. The data and code is available on Dieter Lucas's website which I will sit here and I also want to thank Monique again and thank the organisers for inviting me to lecture—and looking forward to our conversation!
Hello, I'm Mary Towner. I'm going to be talking today about the potential to infer the historical pattern of cultural macro evolution. I want to start by thanking the organisers of the seminar series "Cultures at the Macro Scale". It's an honour to be presenting this lecture here. I also want to start by thanking my collaborators and co-authors of the paper that I'm going to be mostly focusing on today: Dieter Lucas and Monique Borgerhoff Mulder. They're both at the Max Planck Institute for Evolutionary Anthropology and Monique is, of course from UC Davis as well. The work I'm going to be talking about is available already as a preprint—as well as data and the code that you can find on Dieter's website—which I've indicated on this page.
00:52
So cultural anthropologists have long been interested in understanding cross-cultural variation across societies. This might be in languages, in archaeological records, in cultures, the cross-cultural social behaviour, housing styles, marriage patterns... any kind of trait that might be measured about societies. I know from glancing at your seminars from last semester that you've already seen this tree picture: Kroeber's depiction of what cultural evolution might look like on the right side, in contrast to his interpretation of organic or biological evolution on the left side. So with cultural evolution, we have the potential to maybe have more intertwining between societies over time in history. And we might be interested in how traits are transmitted both through vertical means, from parents to offspring or parent generation—parent societies—to descendant populations and societies. And also through horizontal transmission from neighbouring groups, for example.
02:11
So my work in this area really began with time at UC Davis as a postdoc with Monique Borgerhoff Mulder in the early 2000s. And she and others had already been working on this for a while. The way I came into this was through review—with Charlie Nunn as well—where we were looking particularly at some of the ways in which phylogenetic comparative methods may not always be up to the task, in certain respects, in trying to understand cultural macro evolution. So for an example, this box on the right depicts two phylogenies from a small group of East African pastoralist societies and maps on a trait: which is the marriage pattern (whether polygyny is rare, or extensive as shown in the darker bars here). On the left side, you can see the depiction of the phylogeny has one fairly older origin of extensive polygyny, that then is passed on to a number of descendant groups that are closely related to one another. This would be through vertical transmission. Now an alternative mechanism could be that neighbouring groups observed each other, transmitted the kind of cultural trait of extensive polygyny between groups living in close proximity to one another. And there wasn't really at the time a good way to tease this apart without having brought in other kinds of collaborative, corresponding, corroborating data. So this is just one example from that paper that reviewed a lot of detailed critiques and evaluations of methods, in addition to phylogenies used to understand cultural evolution and the coevolution of traits across time.
04:27
Now, from this jumping off point, I began working on a project with Monique, and also Mark Grote at UC Davis, to better understand cultural macro evolution using a data set from Western North American Indians. This database, compiled by Jorgensen, but building on the work of many others before him, looks at 172 Native American societies from the west coast of the United States, up through Canada into... going east to New Mexico. And it catalogues a whole variety of traits. Some that might have to do, for example, with pottery style—this is a picture of Santa Clara Tewa, at a Pueblo in New Mexico making pottery. There might be traits having to do with social organisation, with diet, with types of housing, with residents patterns. So we looked at a number of different traits. And we... our goal with this work was to try to figure out if there was a way to look at vertical and horizontal transmission on a level playing field. So rather than prioritising vertical transmission on a phylogeny, would we be able to somehow model a trait across both the historical relationships between societies (what we might use a language for) to understand vertical transmission, but also to better model a trait bringing in the potential at least for horizontal transmission—for example, through societies being close geographic neighbours.
06:29
So to just give you a taste of what we did in this work (and this is again, with Mark Grote, Jay Venti and Monique) we had a method of modelling that built on neighbour graphs where this is... you can think of it like a matrix that captures pairwise between all of the societies, whether or not two societies were in the same language family. So they would be in the same kind of neighbour graph for language if they were in the same group. And also then, trying to do the same thing in a matrix kind of definition—a graph of proximity, geographic distance between groups. So the figures that you see here are more networks, in a sense. (We're on the left under a). This depicts the relationship geographically between societies if they were within a certain distance from one another, and on the right, looking at whether each pair of societies was in the same language, family or not. So long story short (and this is again, published in Human Nature 2012) we found that for certain traits, models that included both of these components—both language and geography as stand ins for vertical and horizontal transmission—those models outperform models that just included one or the other.
08:11
Now, this method didn't rely on phylogenies at all. But it was a way for us to explore just how much might be overlooked if we only assumed vertical transmission of traits. Now, there are other methods out there that have been developed and explored for modelling cultural evolution without solely relying on phylogenies. That said, phylogenies appear to be here to stay—1000s of publications by now. Here, this phylogeny in particular is just meant to illustrate a recent publication. This one is looking at the correspondence between housing size and post-marital residence, and so Hrnčíř, Duda, et al. in 2020 (Plos One) published this paper to look at whether or not there was a correspondence between having smaller or larger kind of floor space and in the household size and residence pattern: whether women or men went to move into. If it's patrilocal, say, wife would move to her husband's natal household, and matrilocal, the husband would move to live with or near his wife's family's household. And this seemed to suggest a pretty strong relationship between the two variables. So one of the goals with phylogenetic analyses, sometimes, it's not just to look at the history of one trait, but to try to look at coevolution between two traits. And again, this is just one example of many I could talk about right here.
10:01
So we still had some some questions, though. And one of the things (because of another workshop on cultural evolution that I was invited to and later, we were brought on to contribute a paper for this) was to ask, you know, do we still worry about phylogeny? So we may have raised some of these issues 20 years ago or more, but are we still concerned? And I think we brainstormed for a while and came up with a list of questions that we could still say are a bit vexing. (And again, these have been explored with others, and you can read more of the the full bibliography in our paper). But just to show you some of the questions (I'm not going to read this whole thing), but there's still some vexing questions that we have about phylogenetic analysis.
11:02
So for example, what i... to what extent can the history of the trait be reflected in a phylogeny? Is a phylogeny that doesn't show horizontal transmission, for example, going to capture that history? Th.. another question here is whether traits might go together—coevolution, perhaps—or have independent histories? How do we model that? How can we reflect that on a phylogeny? Other questions might be whether the the scale—in terms of time of the phylogenetic tree—is really appropriate for the traits that we might be looking at. So if traits are very flexible and change rapidly, the idea of even needing a tree may no longer be quite appropriate. Because if there's so much flexibility and change in traits, it may just be that we need to look at very recent time and not try to recreate a quite older past. Other questions include whether the tree itself is pretty solid? Is it built on appropriate kinds of data? And is it at the right kind of scale? How, how resolved is the tree? And at last, we have questions about the sample size and missing data; you know, is this really sufficient? For example, do we have enough societies to make large inferences? So if we were working with a very smaller sample, say, of societies—and going back to the example of marriage patterns in East Africa, that particular figure was a smaller group—is that.. how does that sample size hold up? Another example: what about missing data, or if we had a number of societies that had traits that are important, but those are extinct? And unlike some kinds of records, like archaeological records, with behaviour with social behaviour, we may not be able to quickly or correctly reconstruct or make inferences about missing data or data of extinct societies.
13:33
So with this in mind, we set a goal for ourselves to look at these kinds of questions and Dieter's idea was to use simulations on existing phylogenies as a tool—the phylogeny wasn't to be our focus of examination per se. And the simulations used arbitrary traits, not real traits. But could we learn more about the robustness of different phylogenetic analyses, if we varied certain things like the rate of change, or the direction of change, or if there were missing data or missing parts of the tree?
14:21
And an analogy that they came up with in particular—I kind of watched it unfold—but if we look at the past as something that we're trying to get a clear picture of, we're looking through our current window trying to get a clear picture of the past. We are trying to recreate that past somehow. And you might imagine a painter with paint brushes and an easel full of colours trying to get a (somewhat) accurate understanding or reflection of the past. But it's not a direct record, it's our our inferences about it.
15:05
So taking this analogy (I'm going to come back to it) the... very briefly, the methods were to use some known phylogenies: one from the western North American Indian database (the one I talked about before), another more recent one from Pama-Nyungan languages. Looking at those two different phylogenies. And then the picture you see on the left is just a little sub-sample here, taking a random trait, in this case, just go by colour—red or black—if we simulated the distribution of that character, under different conditions, and then we know what the actual history would be (so the actual history is what was actually simulated). And then, we use the methods of phylogen... the phylogenetic approach to try to infer that history, and then we can look and see how well those line up.
16:08
So for example, this is just one of many, many examples that we go through in the paper. Let's say you we were exploring direction of change, what we would do is we'd have the actual history through the simulations. And let's say that the trait changed randomly between two states. Under the simulations then, we look at the inferred histories. And if the trait actually changed randomly, but we wrongly inferred that certain traits changed more or less frequently than others—that certain directions were more or less common—we would call that a "Type I" error that were wrongly inferring history by... with a kind of a false positive here: that there was a change that was more or less frequent, with that direction. Now, kind of looking at this again, but from the point of view where the actual history as the selection favoured changes between certain states over other changes. If we were to, again, then wrongly, infer that all changes were equally likelihood that would be a "Type II" error. So a false negative that we think that there's no real pattern there, when there was some kind of underlying pattern.
17:38
So what we did then, (and really, again I keep saying this, but Dieter did so so much of this coding and simulations under different characteristics), we looked at continuous data and discrete traits. I should say... so let's take an example here, a trait might be something like "type of marriage" (polygynous, or monogamous). Let's just say a continuous trait might be something like the proportion of marriages in a society that are monogamous. Now, those aren't specific traits we modelled again, we used just as a sample, a continuous or discrete trait, as like red or black, for a discrete trait, and then a more continuous value for the continuous tree.
18:36
So for each of these... then I'm... we did a lot of different questions and a lot of different variations of this, I'm just going to highlight a few of them. But the big picture result that overall—just to preview that—is that these errors are very common under all kinds of permutations when we create through simulations and actual history, and then we modify things and compare the results. There are lots of Type I and Type II errors in the findings. So as an example (and going... tying it back into the questions), what if, for example, the data that we have are missing or wrongly classified. So let's say the actual history had three colours red, purple and blue. But the inferred history then only had red and blue. That would give us an inaccurate picture of the past—our inferred past would be missing the colour purple. You can also look at this as if, for example, the purple societies were no longer there—had... had been extinct—we wouldn't have any way to recreate those.
20:06
So the analogy, then, here is that we would be trying to use either the wrong colours or not have enough colours on our palette, as we're trying to paint the past and make an accurate portrayal of the past. So high error rates in different categories across either phylogeny (so it varies a bit what we're looking at), but, but there's enough there to be of concern.
20:39
So let's look at another kind of question. What if there's horizontal transmission of traits? So again, two colours in this in this depiction, red and black. In the actual history, let's say that red has made a jump over to a neighbouring society. The inferred history— based only on phylogenies—that don't take that into account would come up with an inaccurate picture. So you can kind of think of this as perhaps painting with the wrong brush. So maybe the brush that you're using isn't fine enough to pick up those fine details. And again, high error rates across the board really in different categories. In this case, were really missing out or assuming that something might have been inherited from an ancestral population, or when in fact, it It came from a neighbouring population.
21:46
So one more example here, that we looked at was what if the structure of the tree itself matters? Now an example here is where there's a late burst phylogeny. So, if we went back and deeper time, much time would go on without much change here (what we're seeing), and then there's kind of a quick fluorescence of societies that have variation in a trait. And this particular picture—on the left the actual history—you can kind of see where the red is, that the ancestral node is red. But if we looked at inferred histories, we might make different inferences about those ancestral nodes. If we were thinking that this wasn't just a very recent burst now.
22:48
(Whoops, let me What did I do here? Hold on a second.)
22:54
Okay, um, how does this... How does this kind of fit in with our analogy? So again, high rates of errors across the board, but this could be a case where maybe the canvas that we were painting on was really coarse, so that we weren't able to have as much fine detail as we needed, as we recreated our understanding our inferences about history.
23:26
So what do we do? And this is the the checklist, I'd say, that we suggest. But I'm going to say that it's just the start of one—and there are many other ideas about this. One is to make sure that our research questions are informed by good theoretical models to start with. Make sure that the trees that we use are at the right time scale, and built on solid and diverse sources of data, considering how traits are transmitted, and in particular, the potential importance of horizontal transmission, and what that might do to some of the inferences from just phylogenetic methods that don't take that into account. One of the pitfalls that may come from ignoring some of these problems is that the internal nodes that are reconstructed—even though many researchers will say that the exact value of those nodes isn't what matters—that there's a robust kind of outcome regardless of what an internal node has been estimated to be. But that often may get lost in the details where then it's easy to want to try to make inferences about those internal notes. I mean, there were curious about these things, what was the trait in the past—so just to be aware that sometimes based on what what particulars in our tree may be playing out, we may not have as much confidence as we'd like to in those nodes.
25:06
And then the most kind of... the plug here, is that for any particular sample—any particular phylogeny—there may be some unique pitfalls. One way to explore those could be through relatively straightforward simulations on a phylogeny, where if researchers are exploring a certain hypothesis, they can simulate that with data that they create, and then look for where there might be tricky spots where they might be walking into situations where they would be at high risk of making Type I or Type II errors. Or where even if they don't find a pattern, for example, they really can't say much—with much confidence that that couldn't be there based on the intricacies of their particular sample or phylogeny. And again, one more push to check out our preprint. The data and code is available on Dieter Lucas's website which I will sit here and I also want to thank Monique again and thank the organisers for inviting me to lecture—and looking forward to our conversation!