List of mammals of Guatemala

Transcription

[ Music ] >> We will do-- I'll do a little bit of an introduction of traditional and geometric morphometrics. That's the basis of 3D-ID, the software that you have downloaded on our laptops and that we'll get to use and I'll tell you also the design of the project that we did and it was many years in the making and I'm glad that we finished it. [Laughter] Essentially, what the software does, it will determine ancestry and biological sex of an unknown set of remains or a skull-- from a skull. So, we will transition from the lecture part and we'll actually go and get to go play with some real crania and the other part of the lab, we'll go over the landmarks that we've pared down from a standard set that we-- using and also give you some tips on how to use a software because sometimes it's not that intuitive and some tricks on how to upload your data from the text files, that you don't have to cut and paste them all in there which will be nice and other things that are-- may not listed on the website that will help you sort through some of the issues when you actually use it to analyze your remains, okay? Alright, so, fundamentals of traditional and craniometrics-- craniometrics and geometric morphometrics and if you don't mind, I may have to use my cheat sheet a couple of times with this one. Alright, so historically speaking, traditional craniometrics has been based on sets of caliper measurements. So what you end up using-- you have a set of caliper measurements and what you are measuring is the distance between the endpoints of your calipers, right. So it's 2 dimensional. And what you have here as you can see-- and also use those sets of measurements and you use multivariate statistical methods like discriminate function analysis. Sometimes, you use angles to allocate crania into certain groups. But because it is 2D and because it's only measuring a distance between the endpoints of 2 landmarks, you get an incomplete set of biological information for whatever object it is that you're measuring. So you don't have the full biological archive of that object or form that you're using. Geometric morphometrics on the other hand, it's using coordinate landmark data, it's the one that we used, are various types of geometric morphometrics and we'll just briefly cover it. We use the same-- what's nice about it, we use-- if you use landmark data or landmark coordinates, you use the same anatomical landmarks that you would use in traditional craniometrics so it shouldn't be too much of a leap to get there, but it fully archives the biological forms, 'cause you get the X, Y and Z coordinates for each anatomical landmarks. So for example, when you do traditional craniometrics, there's no way to reconstruct that object in real space, whereas you can reconstruct that object or the skull in real space by using geometric morphometrics. And it's also a lot of more visually appealing so you can see some of the changes, right, visually, whereas when you just have a list of coefficients and statistics, it's a little bit more difficult to actually see where some of these landmarks are changing. For example, in traditional craniometrics, you can say, well, if you take maximum cranial breadth, right, the maximum breadth of the cranium, you can decide, well, this skull is broader than this skull, but in geometric morphometrics, you can say perhaps like the left asterion is more anteriorly placed in this population or X population where you can't actually tease that out in traditional craniometrics. Alright, so a few definitions to get us started. This should be on-- I guess on your jump drive and also you should have the PowerPoints on there and some other things. So shape basically is the geometric properties of an object which are invariant to location, scale and orientation. So, say you take your coordinate data, even if you enlarge it or reduce it, you still have that same shape information. That shape information is not lost, 'cause it's invariant. It's supposed to be invariant to location, scale and orientation of that said object, okay? Shape variable is the variables that we use a lot of times to look at these population differences between groups, between species and other things. It's a geometric measure of the object that's invariant to location, scale and orientation so it's a similar definition than the other one. It's just the variable that we're actually using in our analyses, okay? [ Pause ] >> Okay, so the size measure-- and the one thing also that-- let me go back a little bit. In traditional craniometrics, there is no real adjustment for size. You can do ratios. People have done ratios. You can do Darroch and Mosimann where you-- also, it's kinda ratios where you try to account for size difference but there's no true measure to extract size from what it is that you're studying. Whereas in geometric morphometrics, you can choose to look at only the shape of those groups or the species that you're looking at. If you're looking at fish or some other things that you're looking at, you can extract size from the equation, only look at the shape differences. And that becomes really important when you're looking for example at populations for different time periods. Populations have changed overtime. They get generally larger due to positive secular changes. So, if you are wanting to look at maybe the same population and their time, shape changes. You can extract size and only look at those shape changes and not worry about size being a confounding effect on your analyses. And also, it's important due to things like when we're looking at sexual dimorphism, right. So you'll have differences in sexual dimorphism between populations, different groups and those are things that you can look at, also just look at shape whereas not worry about the size component affecting your analysis or your results. So the size measure is any positive real valued measure of an object that scales as a positive power of the geometric scale for the form. In geometric morphometrics, generally we use the term centroid size, okay? And form just means if we're looking at both size and shape, but it's only size and shape and one thing to note is that we do not look at the color of the object, those do not go into it. We do not look at the surface composition. The only thing that we are essentially interested in is size and shape of the object. And I say object because it could be anything. It could be a toy car, you know, we can use this 'cause we actually borrowed this from computer sciences and other biological sciences and we call it the new morphometry but it's not so new anymore 'cause I started doing this in 1998, so how many decades ago is that, right? So, it's not that new but it's still relatively new to forensic scientists because it's very cumbersome to apply. Caliper measurements are really easy. You just measure them, you plug them in to whatever statistical software or equation that you have and it runs. Whereas geometric morphometrics, there is some period where you have to get your data ready to be analyzed and that actually could leave you to throw your computer out the window if you want. Yes? >> This is a quick question. You're using more than just the standard landmarks, right? >> We use-- it depends on what your standard landmarks are. For example, we were trained-- a lot of us from Tennessee were trained using the Howells set so that's a minimum of 77 landmarks that we collect that's using the Howells standard set. But if you're basing your standard landmarks on the forensic set, I think that's 24 not including the mandible. So it depends what standards that you're using, but they are standard anatomical landmarks that are standard in-- for clinicians, for anthropologist, so they are standard anatomical landmarks. Okay, does that answer your question? >> Yes. >> Okay so, geometric morphometrics is the collection of all of the methods in acquisition, processing and analysis of shape variables that retain all of the geometric information and that's one thing about geometric morphometrics. It retains the geometric information of that object. Whereas traditional craniometrics does not retain that geometric information of that object, right? >> Alright, again morphometrics is the study of shape, shape variation and the covariation of shape within extrinsic factors, but that can also include different modalities of analyses. Most of us use Generalized Procrustes Analysis which I'll talk about in a second. But there are other things that you can look at. For example, you can look at outline data, you can look at landmark data, semi-landmark data and I'll go through those things in a minute. The shape again is the geometric properties of a specimen invariant to location, orientation and scale and I know I keep harping on this, but this is one of the main key points and then form is both your shape and your size together and again, maximum geometric information is what geometric morphometrics is doing and this is why it's a different process that we're using in 3D-ID. Some of the good sources are in Slice's paper where he does a really good overview of different methodologies or modalities in geometric morphometrics. Alright, so what kind of data do we have? We have linear distances, right, traditional craniometrics. You can also extract linear distances from geometric morphometrics using EDMA or Euclidean Distance Matrix Analysis. But unfortunately, they are kinda like 2 camps. Right now, you have the camp that does Generalized Procrustes Analysis and the camp that does EDMA. Quite frankly, I don't see why not everyone can play in the same sandbox because one method is better for doing certain things and another method is better doing for other things, but I guess that's a story for another time. Then we have outline data which is part of geometric morphometrics and this is an example of outline data at the bottom here where you look at a leaf. They have used this also to look at things like mosquito wings and other thing. So what basically outline data-- EFA stands for Elliptical Fourier Analysis. So outline data kinda look similar to landmarks, but the difference is when you're dealing with landmark data, you are comparing specific or individual landmarks across specimens and then across populations. So for example, if you're looking at bregma, you will be comparing bregma to bregma to bregma across specimens and across populations. Whereas in outline data, you are comparing the entirety of that shape across populations. You don't take the individual points. And the way you analyze this is, it's in 2 dimensions. You can't do it in 3 dimensions so it is in 2 dimensions. What you do is you get a suite of harmonics. So for example, the more harmonics you get with one harmonic, you see it doesn't represent that shape very well, right. So the more harmonics that you add, you have 5 harmonics, it's getting a little bit better, right, at adapting to the shape. And then with 10 harmonics, it pretty much has that shape down. So the harmonics act almost like principle components, so then you would take those harmonics and you would analyze those harmonics, your normal way using multivariate statistics, okay. So I've tried doing EFA. It's very complicated and my brain just doesn't quite work that way so I have done it but it-- I find it a little bit cumbersome but some people really like it. And they have done studies-- actually, Angie Christensen was able to look at the frontal sinus patterns and she used EFA or shape outline data. Alright, Bookstein-- and when we look-- when we talk about geometric morphometrics, we talk about book colors and Bookstein came out with the Orange Book in 1991 and in that book, he defined 3 types of primary landmarks that we utilize. The first one is type 1 landmarks. Those are the most biologically-meaning landmarks, okay? So these are the most biologically-meaningful landmarks. And they're usually at the intersections of 3 sutures. So I have here as an example of [inaudible] and that's a type 1 type of landmark and we'll be able to identify all of these points in the lab. So this is a true type 1 landmark. Then we have type 2 landmarks and these are-- have some pretty good biological meaning to it and usefulness to it, but they're not as good as type 1, but these are still good to use. These are curvature maxima. For example, if you were to measure the height of the mastoid process, you know, the inferior part of the mastoid would be a type 2 landmark 'cause when you're looking at the anterior portion or superior point of a specific landmark, then you're dealing with a type 2 landmark, okay? Type 3 landmarks are the least biologically-meaningful types of landmarks and these are great for traditional caliper measurements, keep using them but I prefer not to utilize them when I am using geometric morphometrics and we have excluded all of type 3 landmarks from 3D-ID, so that's the first thing that we did was to exclude all those. So for example, type 3 landmarks is maximum cranial breadth eurion to eurion, right? So when you're taking your spreading calipers and you're measuring maximum cranial breadth, so let's say you get 143 millimeters, okay? That 143 millimeters can be in a generalized area, right? So there's a larger area that's 143 millimeters. But when you're trying to identify a single area or point with your stylist where that 143 is it's almost impossible, so there's a lot of intra and interobserver error associated with his type of landmark, okay, 'cause there could be a lot of areas on that breadth of the skull that will be 143, alright, so there's no way to point out accurately where that one position is, alright? [ Pause ] >> Alright, semi-landmarks, apparently, you know, unfortunately for me now that I've got landmark coordinate data down, it's taking me 10 years, right? Now, Bookstein comes out and says, "No, you need to use semi-landmarks." Well, you know what, I'm not listening to him, not doing it. This is complicated so you have a lot of corner points so you have 2 different types. You can have semi-landmarks where you have points that are equidistant to each other, so you're kinda looking at an area of outlines where they don't have a lot of-- so these are really useful for areas for example like the eye orbit where you don't have a lot of anatomical landmarks and you wanna get visually for example the shape of the entire orbit. So using semi-landmarks that are equidistant, then you would tale at say 100 points, right, around the orbit. But then you would align those points between the individuals like every 10 points so that they would be homologous across 2 specimens across your populations. The other one that I think Bookstein is a major proponent of is the sliding semi-landmarks. And what the sliding semi-landmarks are-- and this one I like a little bit better, so for example here, you wouldn't take the whole area because we-- honestly, we don't have software that can actually analyze so many points, everything crashes, so we don't have a software that can actually really analyze all those points and when you think about the samples that anthropologists have, our samples are not large enough to actually be able to analyze all those points. So if we do semi-landmarks-- so this is an example, you have 2 points that are fixed in real anatomical landmark, so Pro Osteon and [inaudible], right. So these 2 points would be fixed and then you would have the semi-landmarks and you can have, what 30 points in between, but when you look at and compare different specimens and different individuals or populations, you would align the fixed anatomical points and then that's how you would align your individuals and you would be able to use those to actually look at your data. So this-- yes. >> In the same landmarks, do you get them by using the slide-- caliper to apply them? >> No, you use the computer, so for example, when you're collecting your data, you're just-- you make sure you get your fixed points on this one and the same goes for the other one. You just point, point, point, point, point and then you do it after the fact where you get your fixed points or your equidistant points or whatever it is that you're using. So that-- you know, you have a set number that you just go and mark as you're going and it happens after you've collected your data, okay? So you've probably have to dump a bunch of points afterwards. [ Pause ] >> Alright, so then we have another type of landmarks and these are constructed points and these are also like type 3 landmarks, not as biologically meaningful and this is from a paper from Williams and Richtsmeier, 2003. Some of the descriptions for example, number 4, that's right here. And this one is mental foramen and it's the anteromedial edge of the mental foramen. Okay, so a constructed point is-- had some anatomical meaning but it's constructed meaning that it's an area that doesn't have a true landmark, okay? And then we have fuzzy landmarks. I like the name of fuzzy 'cause I feel fuzzy a lot of days. [Laughter]. So, fuzzy landmarks sounds good. This was actually-- came out in 1998 and I remember when Valeri ET AT actually presented this at one of the conferences and it was like the next big thing. But what you used as a fuzzy landmark is kind of an anatomical area that you can see, but it doesn't have a specific point like the frontal boss. You can see it like your parietal boss as you can see it, so that would be kind of a fuzzy landmark. I don't think too many people use fuzzy landmarks nowadays, but this is another type of landmark that you can look at and I think people were trying to accommodate the areas of vault in the skull where they don't have true landmarks or true anatomical landmarks that you can analyze. Any questions so far? No? Good? Alright, so reliability of landmarks. Best, most confidence type 1, right? Intermediate, type 2. So generally, what we kept for 3D-ID were the type1 and type 2 landmarks. Type 3 landmarks are the least confident and the worst are fuzzy and constructed landmarks, okay? And we're not saying don't use them ever, I mean, every time that you set up your study, you incorporate your own biases, so all of the studies when you're selecting things for geometric morphometrics is based on the samples that you're using, the questions that you're asking, the availability of the data, all those things actually go into play when you're selecting the types of landmarks and the types of study that you're using. So if you can incorporate for example your own error for type 3 landmarks for your own study, there's nothing wrong with doing that, but if you are actually using data I think from multiple individuals, I think that's when it gets a little bit sketchy 'cause you-- there's no way to know how much error is being introduced based on those types of landmarks. Alright, data acquisition devices, that was the old school MicroScribe digitizer. It used to be the low red machine. Now, it's a black machine. They have different types. You can even have one now with a scanner on it to scan which I haven't use. They have the tablet form, the Polhemus, all of these can be used to collect your X, Y and Z coordinates and if a lot of people also have used point extraction of scanned images that you can get your points off of your scanned images as well. Unfortunately, when you do the scanned images, you don't get 3D, you're actually doing in 2D and I think it's better to keep all of the biological information so I would suggest just going ahead and digitizing your skulls, but if you don't have that available to you, you can use-- and a lot of times these are automated, they have software out there that you can automate and extract your landmarks. These are listed so if you're interested in these digitizers, you can go and look at them. Okay, so you've collected your data, now what do you do? You have pages and pages and pages of 3D data and remember, your 3D data, every landmark that you have, you have an X, Y and Z coordinate, right, for how ever many landmarks that you have, so that's when it gets cumbersome. So unlike traditional craniometrics, you can't just take those numbers and plug it in to whatever software program to run a discriminate function. It just won't work. So, we used Generalized Procrustes Analyses where you have to scale each individual to the same coordinate system. So what does that mean? You have to scale, rotate and translate each individual to the same coordinate system. So for example, you take whatever specimen, say your first specimen, it'll just align everything together so that everybody's on the same coordinate system. We'll see it in a second here. So first, we superimpose our landmark configurations. You can derive Euclidean distances. This is another method that's the EDMA method. You have to maintain a constant orientation or you can partially superimpose landmarks. This is for the error rate that we were studying right now. So we did a study that looked at-- with all of these new software and all of these new studies that were coming out doing geometric morphometrics, nobody actually tested the landmarks that we were selecting for use in these projects using a digitizer or using different methodologies. So this is what we need. But in order to do an error test or a precision test, you have to-- if you're using geometric morphometrics, you have to have the skull in the same orientation. You can't move it. So that's one key thing to remember. Let's say if you're digitizing your skull and if slips are moved, you have to start over again. You can't just continue digitizing because you're introducing all kinds of error, okay? So what we ended up doing is using Euclidean distances, the other camp, right, we used interlandmark distances because our study was based on crania that were at the University of Florida, my colleague there digitized-- Shanna Williams, digitized her crania. She shipped them to me. I digitized the same crania. So obviously, that would introduce quite a bit of error, right? So we used interlandmark distances to look at the error. So Generalized Procrustes is the preferred method. So what does that is that you select a specimen to approximate the mean of that specimen. And then you fit all of your other specimens to that one mean. So that's when you scale, translate and rotate. So the one specimen X as your approximation mean. It doesn't matter which one it is, okay? So it could be random. And then you recompute the mean as the simple average of all the fitted coordinates. So it's a good thing that with 3D-ID, this is all done for you. You don't have to worry about these things. So that's why we simplify that it was Geometric Morphometrics for Dummies 'cause we knew that if we ask people to go through this and I know Kate has as well as I have, you have been reduced to tears at times when trying to work this-- Joan too, right? [Laughter] So you fit your entire new sample to the new estimate and then you do the last 2 steps and this is all done on the computer until you have convergence in that so everybody's aligned and I have here this shows it really nicely. So, the picture over there is you're raw landmarks, right? So you have 2 specimens. Okay, same landmarks, same anatomical landmarks. Then what do you do? You have to center your landmarks, okay, right in here. So all your landmarks are centered, right? And they're not in the same coordinate system that's why they look like they're different sizes, but they're really not. They were just collected in different coordinate systems, okay? So then you center and you scale your landmarks. Now, they're all kinda the same size, right? So they're all scaled. And then finally, you center, scale and rotate so that this is what your end result is and that's what you do with every specimen, okay? Easy peasy, right? >> Long as I don't have to do it? >> Right. [Laughter] Right, that's-- yes, as long as you don't have to do it. So Euclidean distance, what does it mean? It's a straight line distance between objects and you just take as many interlandmark distances from whatever anatomical points that you're studying as what Euclidean distance uses. So it's invariant to location. It's a coordinate-free system so you don't have to scale these, right, because if you do, all the interlandmark distances for all of your specimens, you don't have to worry about that scale thing. So that's why we used EDMA when we did our precision study because the skulls were shipped from one state to another, alright? So see, it does have its uses. These are some area-- if you go to the life bio at SUNY Stony Brook, you have all of the software for free. It's all free ware. You have morphologic-- a past system, one that we used for interlandmark distances. Morphometrica is for Mac, Morpheus et al. and now there's MorphoJ which is another good one to look at population differences. But unfortunately with those, you still have to mess with your data, you still have to run it through a GPA system and learn how to deal with it, okay? But just in case, if you feel you have a free weekend or something, right? >> So here comes 3D-ID and we developed this pretty much Geometric Morphometrics for Dummies. All you need to do is collect your data in Excel whatever it is that you wanna do and you plug your X, Y and Z coordinates into the software. You don't need to know what's going on in the background, okay, that's thanks to Dennis Slice. So why did we do this? We wanted to develop population specific classification criteria using the newer, more robust methods, the traditional methods still work really well but-- and they actually are very comparable to the older methods, but the nice thing about 3D-ID is that you can find subtle differences that you may not be able to detect in more traditional means, so we're still working on it. We're still adding and actually Dr. Slice wanted me to let you know that there's gonna be a new update with new samples soon for you to download, but he said to let you know that the new version of java I guess introduced a slight glitch in the system so he's cleaning that up. It's-- to compile-- we wanted to compile an extensible population database derived from 3D data and Kate has been gracious enough to donate the Guatemalan samples and various other researchers have been able to donate too. So right now we have close to I wanna say 1300 individuals in that database and what we've done is also group them by Mesoamerican, South American-- some of our samples make sure before you run it though to look at the sample sizes for each one because some of them are still kinda empty. So, if you're having problems when you run the software, it may be because some of the reference populations are still lacking some individuals, but what we're doing is working to keep collecting the data so we keep updating and filling those gaps and as we see-- as we need them. Okay, so when we did our reliability study, we looked at-- we wanted to look at type 3 landmarks, type 1 and type 2 landmarks so this is the first test that we did to look at-- well, you know, we're saying everybody should do 3D but you know, what is the error and the precision of using this technology. So this is the 3 skulls from the Pound lab. We had 2 observers. We had 3 digitizing sessions per skull so each individual digitize each skull 3 times. The skulls were not fixed, hence we did interlandmark distances and how you determine your interlandmark distances is you take your N. So N minus 1, right, we had what, 17 landmarks, right? Divide it by 2, or I guess it was-- yeah, divided by 2 and then you take your landmark number minus 1 divided by 2 and then you get 100 so we ended up with a total of 171 total interlandmark distances and that was the suite of distances that we used to analyze the data. So we looked at all of these interlandmark distances and 32 percent showed digitizing error in excess of 5 percent or 54 interlandmark distances out of the 171, okay? So 37 of these included eurion, oops, okay? Twenty eight percent included alare, alright. And the radiometer point and opisthocranion were problematic. And the radiometer point are not part of the forensic set but some of those you who take the house measurements will see that those are very problematic. And opisthocranion, right? It's the maximum length where you run your-- it's kinda like eurion where you have to run your calipers to get your maximum length. So all of these as you can see was the type 3 landmarks that gave you error in excess than 5 percent which is unacceptable. And between observer variation, all of the variation here that you see that was significant also had radiometer or a type 3 landmark so this is between observer. So not only was it intraobserver but interobserver variation was significant when using type 3 landmarks. So, what did we say to that is you have to be cautious when using type 3 landmarks so we decided to take all of those out for 3D-ID. Alright, so we originally started out with 75 landmarks. First of all, can you get busy people to take 75 landmarks, probably not and some of the landmarks with very obtuse descriptions at best, right? Accuracy and repeatability of type 3 errors was poor so that was the first thing we decided. Okay, so let's start by taking out the type 3 landmarks. We still continue-- when I still go on data collection trips, I still continue the entire suite just because that's what I was trained to do and I feel guilty if I don't do it. So that can be for future use or for my own personal use later or to share data or what have you. So, our 3D-ID reference data now uses 34 landmarks for classification, okay? The definitions come from Howells and they come from the Moore-Jansen, Ousley and Jantz, 1994, so there's a standard set. So it's a few more than the standards but not too many and these are our standards that's in here and one thing, we have included now I collect the right orbit traditionally, we didn't collect the right orbit only the left orbit when we did data collection. So when you run analysis in 3D-ID, make sure that you omit your right orbit superior and inferior border because most of our reference sample unfortunately were collected in the early days and some people still have not moved to colleting both right and left. So, when you want a larger sample, we have a larger set using just the left orbit. So we have it in there 'cause we're trying to move towards collecting both orbits, okay? So as you can see, they all are standard sets. This is the posterior aspect, the inferior, and we will go through all of these in the lab. We have-- this is at the time of when we released 3D-ID, we had 1089 individuals. I've added about almost 100 Angolans, Portuguese, Cambodians as well are included-- that's gonna be released soon. And we would like-- so these are the people that actually helped with providing either collections or assisted with providing data. So this is the data pane, it's pretty easy. It's easier to collect your data in the format. It's in alphabetical order so you start out with asterion and all you do is literally cut and paste into coordinate data. But if you want to import, you can also import the data sets. So if you're collecting in Excel, you make sure you collect your data, actually not in the new version of Excel but in the compatibility mode of Excel 'cause we found that out yesterday that MicroScribe will not work with the new Excel. It will only work in the compatibility version. So, if you collect your data that way-- and the best way to do it is just type in your landmarks in your Excel sheet on the left and you can just start collecting and it'll just collect all the way down in a row. Then what you will do is save just your landmarks, just your X, Y and Z landmarks without the landmark definitions or-- excuse me, the landmark names on the left. Save it as a text tab delimited file and then when you go in here, you will go into program and you will say read file, and it will pull up your file. So-- oh, one other thing to make sure you leave a space for the right upper and right lower orbit and another space between-- it was nasion and left orbit also. So it has to be a space in there or the data will come in in the wrong area. So then you save it as your text and all you have to do is read file and then your data will be in there. And that's because we bothered Dennis over and over again. It's just crazy when you have several cases you wanna put in to cut and paste every landmark and putting it in there. So he listened thank goodness. Now, we can import our data from a text file, okay? So we'll work on that in a second. So then you can go in here, just-- I would leave the shape dimensions to use, it's just a leave what's in there. >> You can determine group and sex. One thing I will say that when you add sex to the mix, it lowers your allocation because once you start looking at different populations, sizes of factors, so that happens to traditional methods. When you start doing discriminate functions based on ancestry and sex, your correct classifications will be reduced. So I would go in here and then look at-- we only have a few inserted Caribbean area right now. I believe it's Panamanians, a couple of Bahamians that are in there. So, know your demographics of your population and then you can click some of these areas. Africans might be skimpy still. So click what you want and then you just hit process, okay. And it should give you the group that your skull is more closely related to. And we have used it and I know Aaron Kimberly has used it in Florida and it has worked really well with Mesoamericans and with the young boy-- it doesn't do-- it can't do sex for juveniles, but it has been classifying are juveniles into the correct ancestral group or geographic area especially with Hispanics. Here's your data option and then what you would hit is the process button at the bottom. This is what your data will look like. You'll have 3 columns of X, Y, and Z. It can do missing variables so if you have a fragmented skull, you just plug in what you have. But if you're importing it as a text file, make sure you have those missing rows in there so that your data goes in to the correct row, okay? And this is typically what you would get. You'd get a posterior probability, you'd get typicality and then this will give you the sample sizes of the groups that you picked, okay, for allocation. Alright, some of the applications, I've been testing it on the cases from the medical examiner's office that we get and we positively identified one case as European-American versus European. We have a European group 'cause as we know, European-Americans do not look like Europeans. And it had a posterior probability of 0.65 in a typicality of 3 so that worked pretty well and we also tested it-- we're still using Fordisc so we can compare our results and actually Fordisc is how we align ourselves and make sure that we're working how we're supposed to work. As you can see, the results are very similar and they had it in white male. And then we also were able to allocate, we have an adult group that will be concluding there. We tested this using subadults of Portuguese and it correctly allocated the Portuguese juveniles into that European or the Portuguese group, so it is working on which is nice to say working with the juveniles as well. We used canonical variate analysis. Dennis is working on alternative fitting methods. He also has where he tests everything using cross validation. So we're trying to get age adjusted but that's gonna be in the future as well and we're gonna try to incorporate some of Joan's postcranial data in this as well. Future applications, what I'm very excited about are the juveniles because like I said, we were able to look at the juveniles from Portugal and they allocated correctly-- the Mesoamerican 10-year-old, allocated correctly. And I wouldn't use them in the really young, but I think once you get to that 10-year period, right, because number 1 we have reached, what, 85 percent of our size by 10. So I think that's okay and also ancestry Unna determined that you could actually distinguish that in utero so a lot of those things, so-- I think you use it for the older kids or teenagers that will be safe to do it so the applications have been from going just using for adults that we may be able to use it now for children. This is a study that we did that we looked at Cuban, historic African slaves. We had children for that Africans, Americans, Portuguese adults, and Portuguese subadults and as you can see that our Portuguese subadults, 50 percent classified into the subadult Portuguese and 50 percent classified into the adult Portuguese so we got the correct allocation for the group so we were pretty excited about that. So again, we can correctly characterize ancestry and subadult crania. So these are some of the things that we would like to add in the future, but again it all depended if we can have samples of the same population of subadults and adults so maybe we can move to maybe CT scans or something in the future to get some data on that. Thank you. Any questions on geometric morphometrics. No? >> I have a question. >> Yes? >> When you were doing the error testing with the type 3 landmarks, will you still have some marking where your eurion was or-- >> Yes. >> Okay. >> Yes, I was 'cause-- yeah, you can't-- yes, still problematic, yeah. But still pencil marking in. Yeah and there were-- and opisthocranion also pencil mark. >> Its tough. >> Yes. You can't eyeball that one fortunately. >> Alright. So Anne, when you're saying they're not fixed-- >> Yeah. >> How do you setup. I mean I'm sure I get [inaudible] to see it but-- >> Oh, they were not fixed meaning that they did not move between observers and between sessions, that's what I meant by fixed. >> Okay. >> Yeah, so I do have them in a stand but if you move them from place and then put them back in or between skulls and stuff, that means it's not fixed, okay? Yeah, sorry, yeah. Okay. Thank you. >> So this is all goning magically make sense when go next door[inaudible]. >> Yes. [ Music ]