Teaching with ORBIS #lawdi

On the last day of our Linked Ancient-World Data Institute this summer, sponsored by the NEH Office of Digital Humanities, I argued that it was important to show how all this exciting work could have practical implications for what professional classicists (and other ancient-world types) spend a lot of time on, teaching. To that end, I promised to do a post describing how I had assigned my Classical-archaeology students some short homework using Stanford's great new tool, ORBIS. What's ORBIS? In the words of the site, ORBIS "reconstructs the time cost and financial expense associated with a wide range of different types of travel in antiquity." More simply it allows you to map routes between two places in the Roman world given certain constraints for cost, time, and type of route.

Naturally the fine people at ORBIS did a nice upgrade to the service after I made that promise, but before I got it completed, so I had some more work to do before this post. (Fair enough, I dragged my feet for too long anyway. Nemesis strikes!) But finally here it is, suitable for framing (or at least bookmarking).

A Very Short Guide to Using ORBIS in your Classical-Archaeology Course

1. RTFM

Make sure that you understand as much as possible about ORBIS, what it is, how it works, and so on. You don't need to be an expert, but you should at least be able to do more than your students will by the time they finish the assignment. It won't take more than an hour to read the "Introduction to ORBIS", "Understanding ORBIS" and "Using ORBIS" tabs on the website. Don't miss the nifty how-to videos. Although there's more there to read, these three sections will get you far enough for step 2.

2. Make sure you know how to use it

Play around a bit yourself on the "Mapping ORBIS" section. Try to get from one place to another. Change the various parameters. Use all the controls, so you know how to change the views of the route and so on. Click the buttons and links and sliders. Go nuts. Depending on your technological prowess, this will take you a few hours at most.

3. Demo it in class

Once you're confident that you can show your students the basics of the site with confidence, have your student read those same three sections of the ORBIS website that you read up in #1 in preparation for a short demo of ORBIS that you'll do in class for them. Nothing fancy, just enough to show them the basics. I like to point out to them how long travel takes when you don't have motorized vehicles and how much faster travel over sea is than over land, but be sure to walk them through creating a route and choosing the various options, no matter what extra details you cover.

4. Assignment 1 of 2

Have your students use ORBIS to find a simple route between two places that you specify. Have them do it under multiple conditions. (I used three different sets.) Then have them either print out or take a screen shot of the result, with all routes shown. Here's one with routes between three sets of cities, one taking the fastest route, another the cheapest, and the third the shortest.

Since you've set the parameters, you'll know what the correct routes should be, and thanks to the different colors, it's easy to tell at a glance whether the student got it right. Successful completion of this will indicate that your students can handle using the basics of ORBIS. Make sure they all successfully complete this first assignment. Then they're ready for part 2.

5. Assignment 2 of 2

This part is up to you. Depending on which section of your course you're using ORBIS in, you'll want to find some question you can answer, or some issue you can illuminate via ORBIS. I actually did something that was completely out of ORBIS' chronological span.

To help my students understand the rationale behind some of the placement of early Greek colonies in the west, I had them examine routes between Delphi and Naples. The latter was used as a proxy for the earliest colony of Pithecoussae. (This was actually a variation on an assignment I had made up years ago using a QuickTime movie with links to the Perseus website.) The biggest travel difference between the later Roman empire, the time in which ORBIS is "located", and the geometric period was the roads in use. Obviously none of the vast Roman road system was in place, and so I made sure the students used only routes that avoided long portions over land. (I want to use this assignment again, so I'm not giving away all the details!)

6. Put the students to work

If you subsequently have your students come up with their own ORBIS projects, odds are they'll find something useful and interesting to do, perhaps something you hadn't quite thought of. A set of mine, for example, used ORBIS to explore the different travel experiences of three characters from the ancient world with differing socio-economic backgrounds, complete with clever backstories!

And there it is. Hope this encourages you to use ORBIS and other terrific ancient-world-related DH tools in the classroom! I'd love to hear in the notes about your experiences with ORBIS or with any other tool.

Were the Ancient Minoans Europeans?: A Comment on doi:10.1038/ncomms2871

Just got around to reading "A European population in Minoan Bronze Age Crete" <doi:10.1038/ncomms2871>, which argues for a European origin for the ancient Minoans based on genetic analysis of mitochondria from some ancient Minoan bones. Their analysis suggests that the ancient Minoans were most closely related to other ancient peoples and, among moderns, the populations of northern and western Europe. This contradicts Arthur Evans' theory that they were related to ancient Egyptians.

This isn't a field I'm entirely up on the research in, but it does overlap with my old interest in biochemistry and my current work in archaeology. Some of what I write here may therefore be easily corrected by those in the know, but I had a few problems with the paper (which I enjoyed overall). Please enlighten me in the comments.

Inconsistent Data

Table 1 of the pairwise genetic distances between Minoans and other populations seems to be based on the data presented in Supplementary Table S5, but it doesn't give the same data. For instance, the first group is Bronze Age Sardinians, which have an average pairwise distance of 2.75111 in S5, but 2.89 in Table 1. Maybe that's because these two tables were calculated using two different versions of the same software (Arlequin 3.5.1.2 vs 3.5.1.3), but that appears unlikely, since the Arlequin updates page doesn't seem to list any such change between versions.

[BTW, if you're going to give a supplementary file of data, why not put it in some nice format like CSV instead of a PDF? I fixed it up with Tabula, but that was an unnecessary expenditure of my time.]

"Ancient"?

The authors divide their comparison population groups into two chronological categories, "ancient" and "modern." Their ancient group includes 11 distinct samples: 6 neolithic, 3 Bronze Age, 1 medieval/Iron Age(!), and 1 Byzantine. They seem to be taking the nomenclature of their original sources, so it's worth noting that Byzantine means 11th −13th centuries and Iron Age-Medieval-Nordic means 1st - 14th centuries (and that the original authors did not apply this catch-all term). I didn't look up all their "modern" samples, but the remarkable chronological span of the their "ancient" ones already raises some doubts in my mind. You can't lump together millennia of data like this and act like you have homogeneous groups. This makes map b in Figure 3 misleading as well, even ignoring the way the map is colored in areas where there is no data at all (like all of North Africa, on which more below). Line c of Figure 4 does break these ancient sets up into three distinct groups, leaving out the Neolithic, but even this obscures the fact that one of these groups has a much higher sharing than the others: the one from BA Sardinia with 52%, compared with a maximum of 12% for the other two BA sites, and 15% for all four of the other locations. I'd question this approach, as it's a great example of why the mean can sometimes be a bad choice for comparison: with an N of 3 in the BA group and a standard deviation that's larger than that, you'd be better off reconsidering your choice to group these data points, or at least use the median.

If we break out this Sard group then, we're left with a much more consistent sharing with the Neolithic groups (mean=24%), and fairly low sharing with the other BA, the Byzantine, and the (absurd) Iron Age-medieval group. In short, with that one Sard exception, all the European post-Neolithic "ancient" groups match the Minoans about as well as the modern Middle Eastern/Turkish/Caucasian sets. Since the Minoans are BA Mediterraneans, that makes me wonder why the Minoans are so different...you know, assuming again that that small N isn't to blame.

Line d of that same Figure 4 breaks down the six Neolithic locations into three sub-groups based on a north-south positions. Again, I'm not sure how far you want to go with an N of 1 in the northern set. The largest sharing with the "southern" group—really France and Spain—is not surprising, given their proximity to the Mediterranean and the likely mixing of those populations in the Neolithic (as discussed, for example, in the article they take one of their datasets from).

More troubling though is the complete lack of any ancient sample from anywhere further outside Europe than Asia Minor (modern Turkey). If you're going to claim that Minoans aren't like ancient Egyptians or Middle Easterners, you'd probably want to have some sample of those groups to compare them to. The authors do note the lack of the typical African L haplotype in the Minoan group, but I'd still like to see a direct comparison with some North African Neolithic and BA data.

Note revealing my ignorance

I find the methodology of creating the haplotype sharing quite interesting. Count the individuals from each set who have a haplotype that appears in any of the Minoan individuals, then divide by the total number of individuals in that set to get a percentage sharing. For example, say you have a population with three haplotypes, A, B, and C, distributed in the population at 80%, 10%, and 10%, respectively. Another population with all the same haplotypes distributed 1%, 1%, 98% would have 100% sharing, while a third population with 85% A and 15% B would score 90%, even though the distribution of haplotypes was more similar than the second group's in overall distribution.

So in the data in this paper, we see that the Sard BA sample had the largest overlap of any group, but with only two shared haplotypes, while the North Africa set had a much lower sharing percentage, but 6 shared haplotypes. I realize part of the problem is the small N's involved, which often don't permit for a great representation of the overall distribution of haplotypes within the total population.

Update: A post from a more knowledgeable guy than myself are here.

Apple's Stupid Data Detector: Meetings

Built into OS X is the great capability for detecting when a meeting is being mentioned in things like email messages. I'm calling this by the old System name of "Data Detector", though I suspect that might not be what Apple calls it now.

It's pretty impressive on the whole. Given a bit of text like the one in a message I got this morning...

The meeting will be on Friday, April 26th at 9:30am in BC 106.

on hover it highlights the date and time bit and give me a little pop-down menu to make an appointment in iCal out of these data:

I can edit the appointment before it gets created, putting it in the right calendar, for example. Great, right? Yes, great.

But...

It also does a few stupid things. First it uses the subject of the message as the appointment text, which would be fine, except the subject very clearly has the same date and time info and it doesn't bother to try to strip that out, even as it removes the leading "Fwd: " from that same text. Second it's apparently too stupid to ever recognize any location info, even when that info is pretty clearly indicated by adjacent and obviously locational prepositional phrases like "in..." or "at...". Sometimes it's going to get those wrong, but a lot of the time it won't, yet it clearly hasn't been written to scrape that info out of a message. Finally this capability isn't active in text that you're typing. So if I'm in Mail, writing a message to confirm an appointment with someone, I can't just click on my own date and time text and make an appointment out of that. I have to go into iCal and do it.

This is one of those insanely great things I love in OS X. It's frustrating when it's senselessly crippled.

Back to 3D

I did a little experimenting with some 3D after Christmas 2010 when my son (not me, I swear) got an XBox with Kinect. Fun, but labor-intensive, as the tools were still fairly unpolished, especially for the non-programmer. (Search for posts with the Kinect label.) Then a month or two ago, I started getting into it again, thinking about how I could use it on the dig, and including some research into various other kinds of photography. I was looking at the freeware stuff (like VisualSFM), which now looks pretty good, but still wasn't quite working on my MacBook. Then I got into using Homebrew instead of MacPorts, and one thing led to another and I got busy with other stuff (like my actual day job).

Then a few weeks ago, my friend and colleague Sebastian Heath started tweeting a bit about stuff he was doing in 3D, using the inexpensive (but not free) AgiSoft PhotoScan. Looked pretty good, and he poked me a bit about not writing up what I was doing, so here I am.

Well, almost here. I bought the software today and spent a little time with it and my iPhone. My first test model was a head of Michelangelo's David, which graces a column in my living room, but that turned out to be a bit shiny, so I moved on to another iconic figure who makes an annual appearance in the house. I snapped 14 photos of him, not quite going all the way around. Then I tested it out in PhotoScan using the speedy low-res settings and since that looked good, I cranked it up to 11 to come up with a fairly nice model, especially given how lazy I was about it. A still image is to the right. Click to enlarge. You can mostly read the numbers of the various pockets (though that's a 6, not a 5 right in the front) and the detail isn't bad. The whole thing took a little over an hour to render and then I moved it into MeshLab, smoothed it out, and exported.

As I said, more soon. I'm hoping to have some fun over Christmas break on our visit to the in-laws.

Election 2012: What if the House were bigger?

As usual after a presidential election, there have been a lot of maps showing how the vote went in the US electoral college. One thing that I haven't seen get much discussion is the effect of limiting the size of the college itself.

Thanks to Nate Silver, lots of people now know the number of votes in the US electoral college (whether they realize it or not): 538. With the exception of the three given to Washington, DC, the electors are distributed to each state according to the number of congressmen it has. (DC's total is equal to the number it would have were it a state, not to exceed the number held by the least populous state.) This formula sets a minimum of three electors for each state, equal to its two senators and one House representative. Currently the seven least populous states have three electors. This minimum perpetuates the equality of states' voting power in the senate, where population is irrelevant and every state gets two votes. It's also the case that every state must have at least one representative, no matter how few citizens it has. In other words, the smallest states have more voting power than their populations would warrant otherwise. (There's also an inherent inequality in that there needs to be a whole number of electors, which means there will be other inequalities due to rounding.)

So what would happen if the House were bigger and there were therefore more electors and also a more proportionate relationship between a state's population and the number of its electors? In effect this would give more influence to the more populous states.

To check this out, I made a little spreadsheet to recreate the distribution of electors. The tricky part is assigning members to the House of Representatives. This assignment isn't quite straightforward (here's a nice little paper on alternatives to the current method), and my reconstruction spreadsheet of the current situation (435 members total) doesn't quite get it right (MN and RI are missing one member each), so my projection of what the House would look with more members is likely a tiny bit off too. (I could spend some more time on this, but I think I'm close enough and a little internet searching hasn't helped me out. Suggestions welcome in the comments.)

So what happens? In this election, Obama with 50.5% of the national vote got 332 electoral votes to Romney's 48% and 206. Had the House 485 members (an average of one more per state), Obama would have gotten 364 and Romney 224. Throw in another 50 and Obama's at 396 and Romney 243 (you'll note that my model is one over the real total of 638). In all cases, Obama wins a rounded-off 62% of the electoral college, so no change in that metric.

That doesn't mean that there would no changes at all in the way the election might go. For example, it might be possible to put together a different collection of states to win, or to neglect more of the smaller states and still win. It does however look like the picture wouldn't change much even with a much bigger House.

(It would be interesting to consider what would happen to the split between the parties in a bigger House, but given all the complications with drawing districts, that's far from straightforward to work out.)

Post-Sandy Update

It's not really post-Sandy for a lot of people. There are still hundreds of thousands without power and many - including a bunch of people I know at Breezy Point and a bigger bunch that I don't, but see every summer - are in worse shape than that. Still we've got power back and things are close enough to normal that I thought I'd revisit what I wrote the day before the storm hit.

1. The Model - Wow, the model nailed it. About the only thing it got wrong was the speed of the storm, which moved faster than expected...fortunately. Instead of having a landfall early Tuesday morning, it landed late Monday evening and by daylight on Tuesday was mostly gone.

2. The Rainfall - As predicted, rainfall was minimal around here. It didn't even hit the lower end of the range that had been predicted when I was writing. Instead we got a little more than an inch. My sump, which was dry before the rain came, was dry after too, and that was a good thing because we didn't have power from about 7:40 pm on Monday night. So a flooded basement was the one thing I didn't have to worry about. (And likewise I was wrong that we would get flooding without power.) As I wrote, this lack of rainfall was not something anyone was talking much about on TV, even though it meant one thing a lot of people did not have to worry about.

3. Outages - Al Roker was right. In the coastal areas that got hit hard with ocean surges, power seemed to be out uniformly, while elsewhere outages were spotty. Here in Maplewood, for example, the Village never lost power, while a few families are still out. We lost it for nearly eight days. I think I was right that damage in this immediate area was less than last year's Halloween storm, but there was so much damage along the coast and in heavily wooded areas, that the various utility companies couldn't repair it as quickly. That said, some of the damage around here was pretty spectacular, and a lot of trees went down, taking power lines with them.

4. Timing - The good forecasting and speedy action by public officials was indeed aided in great measure by the weekend. Lots of colleges and universities (like mine) were able to send kids home in plenty of time, and the various public-transportation systems had time to take action without worrying about a lot of people getting home from work. Individuals were also able to do a lot of shopping...a lot.

Once the full extent of the damage had become clear, there were a few lucky things too. The fairly low rainfall meant that those who hadn't been flooded didn't get water damage on top of the power loss. Also temperatures in the first days were moderate, with highs in the 50s, so people without heat could get by more easily. The major roads were fairly clear too, so on Tuesday and Wednesday it was possible to get around (or out of town) without too much trouble. Finally outages in a lot of areas were spotty. As I said above, Maplewood Village never lost power, and the public libraries were able to open up right away too. This meant that it was possible to escape a dark home to recharge one's literal or figural batteries. Cell service was a bit spotty, but good enough that most people I know had telephone and internet at least intermittently, which was good for keeping in touch.

At our house, we learned that our water heater doesn't need electricity, which was a nice surprise. That meant hot showers and clean dishes. Out gas stove kept working too, even if we had to light it by hand. No oven though, since that is controlled by the electronics in the control panel. We have a propane grill outside, which we didn't use, but could have. We also keep the house pretty cool in winter, so indoor temps in the high 50s were familiar.

Once it was clear that we didn't have to worry about flooding the basement, and that school and work were going to remain closed at least through the end of the week, we decided to make that visit to friends in upstate NY that we'd been putting off. Off we went, and discovered that just a little over two hours away, in an area that had been hit hard last year by the remnants of Irene, life was going on pretty much as normal. We caught up on the TV coverage of the disaster at home and counted our blessings. It seems like about half of our local friends performed some version of an escape.

This is two years in a row that we've lost power. Last year it was brief for us, but friends went without for several days. So I've got some plans to be better prepared. As I said, as long as our gas stays on, we have hot water and can cook indoors. Without gas, we'd be cooking outside. No electricity is a drag, but the appropriate candles along with the battery-powered small lights we have already (yay, LEDs!) can get us through the nights. For entertainment, we certainly missed more accessible internet, but cell service was good enough for keeping in touch and the radio was great (kudos to WNYC for terrific coverage of the storm and its aftermath). Our laptops could run all evening as long as we had charged them up during the day, but TV would have been nice (still finishing up The Wire on NetFlix).

The really big barrier to staying comfortably in the house was heat. Our furnace uses gas to create steam, so it doesn't use a lot of electricity, but it is hard-wired into the house line, so it couldn't be run without some playing around. It doesn't use a lot of juice, so it could easily have been run off an inverter from the car (which is how I charged up my phone), had it been wired for that. So item #1 is a transfer switch, which will let us run the furnace off the car inverter...or a generator, which we ordered on the last day we had power and arrived on the last day we didn't. The generator runs on propane, which we now always have for our grill. It isn't big (2kW), but it's enough to run the furnace, our sump pump, the stove, and various appliances, at least one at a time. The Prius in the driveway means that we have a fairly large supply of electricity around the house, so in a pinch we don't need to fire up the generator. We get about 400 miles to the tank, so we haven't been bothered by the rationing still in effect in NJ (and due to end in a few days). I may get a battery backup for the sump pump too. Those can last a few days at best, but I'm more concerned about short-term outages, like Irene last year, when the sump pump was running virtually non-stop for a few hours as the storm front passed through our area. I'd rather not have to run out to fire up the generator in those conditions. A back-up water pump is another possible purchase. I'm looking at smallish transfer pumps which can run off the car inverter (ours maxes out at 100W) and can be moved around to handle whatever flooding we might get. If nothing else, it could also help reduce the load on the sump pump during those surges.

All in all, we were pretty lucky. Lots of people lost their homes, other property, even lives. We had to do some labor, but got to visit dear friends and watch the destruction from a safe distance on TV. One cold night in the house was a small penalty in comparison to what might have been.

Ramblings on Sandy

Public officials in my area (northern NJ) seem to be doing all the right things: evacuating low-lying areas, shutting down public transportation, sending out all kinds of warnings. For my part, I find myself driven to skepticism by the slightly creepy enthusiasm of the media weather people...as usual. In this instance, this was fueled by the way a rather hasty mention on the Weather Channel Saturday morning of the injection of some dry air from the southwest which resulted in a serious drop in rainfall south and east of the storm.

So a few thoughts, not all of which are cynical:

1. The Modeling - Everyone keeps talking about the uniqueness of this storm. It's a strange kind of hurricane now, and is likely to lose hurricane status soon. It's also supposed to take an unprecedented left-hand turn, from what I can tell. Here's a plot of historical hurricane paths for hurricanes that were very close to Sandy's position. You'll notice that none of them takes such a sudden change in direction. So how good are the models at handling this kind of thing? We'll find out tomorrow morning, when Sandy either turns or doesn't. I'm genuinely curious...and not only because I'm not interested in bailing out my basement tomorrow night. (For the record, right now the consensus puts landfall somewhere in southern NJ.)
2. Rainfall - Last year we got seriously dumped on my Irene, about 6" in less than a day in my neck of the woods. In addition we'd had a fairly wet lead-up, so the ground was already full of water. This year in contrast has been average or even a bit dry, and Sandy is forecast to leave a total of 4" of rain over several days, starting some time tonight. I'm sure the coastal and riverine areas are gong to be much worse, but we're likely not to have as much potential for in-house flooding as last year. (If the power goes, we're definitely going to get water, but right now my sump is totally dry.)
3. Some people on the air (e.g., Al Roker) are giving warnings for 7-10 days of power outages. Now I'm sure that it's very possible that some small isolated areas could experience such a long absence of power, but most of us are very likely not to. In part this is because of the response to all of the problems after last year's Snowtober event, which resulted from many, many fallen trees and limbs, and the subsequent difficulty of getting around in the aftermath. I'm sure the state hasn't solved all the problems of last year, when only the rare spot had outages even approaching a week, but it's likely taken care of some of them, so we should see fewer problems this year.
4. Finally this timing of this year's storm is a lot better than last year's pair (Irene and Snowtober), both of which hit on Friday/Saturday. This year we started getting warnings on Thursday, and so had the entire weekend—instead of a few workdays—to empty out the shelves at our local hardwares stores (and surely others were better behaved than we've been and prepared in advance).

Bottom line? While I won't be surprised if the models are right, and the coasts really get hit hard, and the power goes out for a while, I do think the overall situation may well be better than last year (either big storm), and our personal one won't be any worse. But we'll all know much more in just a day.

Meanwhile...gotta find those AA batteries for my alarm clock.