Tag Archives: dataviz

How to Make an Award-Winning Scientific Poster

  1. Decide with one month to spare that you will attend an overseas conference and frantically make travel arrangements.
  2. Email the incredibly responsive conference organizers because you’re not sure if you can give a talk or if you need to present a poster instead.
  3. Make a poster so great that you win the poster competition and get to give a last-minute talk the next morning as a result.

So that happened. You can watch my talk here.

The Cool Stars 19 conference was outstanding for many reasons, and winning the postdoc(!) poster competition was just one of them. Of course, the ability to make a good poster did not appear in my head out of nowhere! One set of great guidelines is here, and AstroBetter has compiled many handy resources about conference presentations (including posters) too.

coolstars19_poster_final.png

Rather than rehash everything that has already been said about making effective posters (you did click on the links above, right?), here is an overview of how I approach poster design together with some tips to make your poster visually compelling.

  • Re-read the abstract you submitted to the conference weeks ago. Is it overly ambitious? Totally off-base? No matter. Your poster is an opportunity to communicate what you’ve done as of TODAY. Hundreds of people will glance at your poster, but only a few will skim your abstract.
  • With the above in mind, copy just the title and author list from your abstract submission (or craft a condensed version if necessary) and forget about the rest of it. Do not copy and paste your abstract anywhere. You will be staring at a giant blank canvas; it will be tempting. Just say no.
  • Use whatever software you are most comfortable with to make your poster. This could be Illustrator, Keynote, Powerpoint, Word, whatever. Does your software let you make a large page size, insert text boxes, images, and shapes, adjust fonts and colors, and precisely drag stuff around? Yes? Good.
  • Follow the conference’s rules for poster size. Sometimes you have a choice between portrait and landscape, and sometimes you don’t. You can make a good poster in virtually any aspect ratio, so pick something within the limit and stick with it. (US-based pro tip: 36 x 48″ is a standard size at FedEx Office and very similar to A0; having one dimension be 36″ also makes printing on fabric easier.)
  • Boil your message down to two main points, and be able to say each point in one tweetable sentence or less. Then divide your poster into three chunks: the first portion is reserved for big-picture motivation and context, and each of your main points gets another third. Use distilled versions of these points as headers for each chunk instead of meaningless words like “Introduction” and “Conclusions.”
  • The focus of each chunk must be figures. Let the figures speak for themselves. For example, you do not need to write “y generally increases with x over the full range of the dataset.” Instead, use clear axis labels that communicate why the correlation is important.
  • Write bullets and single-sentence captions, not paragraphs. Write only the most essential words to explain each figure, and no more. Notice there are no end-of-sentence periods on my poster, because each thought stands alone.
  • Avoid acronyms and jargon. Yes, knowing your audience is critical, but mostly, your audience is tired. Write what you mean, not what your sub-discipline’s jargon-and-acronym-machine expects you to say. If you must use acronyms, define them near the very start of your poster’s first chunk.
  • There is no ultimate poster font size law… but yours is too small because you’re trying to cram in too many words! My default was 52 pt (with some figure labels necessarily a bit smaller). Sans-serif fonts are easier to read than serif fonts.
  • Pick a simple, colorblind-friendly color scheme and stick with it. This means your plots, your text, and anything that goes on your poster should ideally coordinate. If you’re not sure what to choose, consider the colors in the figures you plan to include, and/or consider the color palette for the conference itself. Dark text on a light background is easier to read than the opposite.
  • Don’t be afraid to overhaul your figures. Crop out extraneous axes and tiny labels. Make custom legends. Annotate plots to drive home key messages. Present one illustrative example instead of seventeen nearly-identical panels.
  • Advertise your papers, and give credit where credit is due. Essentially everything on my poster is from my own work or my credited coauthors’ work, so there are very few formal citations. If you borrow any plots or images, be sure to cite the original source. However, a poster is not a literature review; if people are truly interested in your full set of citations, they can read your papers.
  • Take your time and be picky. Ensure elements line up within and between chunks, use the exact same colors and fonts throughout, and get several sets of eyes to proofread.

Remember: the whole reason you travel to conferences is to talk to people about your research. Sometimes you will be standing near your poster and sometimes not. Do you know what I did with all the words I wanted to put on my poster but didn’t? I used them in conversations, and they appear or will appear in papers. So yes, your poster must stand alone, but it shouldn’t have to tell the full story.

Finally, consider how you usually feel during a conference. Mentally sharp? Plenty of free time? Excited to carefully read each poster twice? Of course not. Everyone is exhausted and overwhelmed. That’s really what these tips are dancing around. Could your sleep-deprived officemate distill the essence of your poster after staring at it for 20 seconds? And do those 20 seconds make them care about your results? If not, you still have some work to do.

Mapping Car2Go Portland

As a relatively new resident of Portland, Oregon who doesn’t own a car, I’ve become quite reliant on walking, public transit, and a nifty service called Car2Go (plus the occasional Uber or Lyft). Car2Go is a pretty neat on-the-spot car rental service that I first joined while visiting San Diego a few years ago.

Basically, there are a bunch of 2-seater Smart Cars all around a city, and anyone with a membership can walk up to one and rent it for $0.41/minute. You can park it anywhere on the street in that city’s “home area” when you’re done, and if there’s a meter, you don’t have to feed it. There’s a one-time $35 signup fee and a free smartphone app that shows you where cars are in real time. It’s a pretty awesome deal.

(I could go into more detail about the service, like how you can reserve a car for up to 30 min ahead of time if you want, and how one time it said there was a car and there wasn’t and I was sad, and how parking rules vary a bit from city to city, and even how they have not-unreasonable-but-pricey maximum rates for hourly and daily rentals… but you get the idea.)

The branch of Car2Go in Portland announced today that it was “updating” its home area—that’s the area where you can find cars to rent and end your rental—so I went to take a look.

Here’s the present Home Area.

PortlandHomeAreaMap

And here is the new Home area, effective August 24, 2015.

car2go-Portland-Home-Area-Map-FINAL

I was initially annoyed, because this interfered with my brilliant plan to take Car2Go to within a couple blocks of symphony rehearsals in East Portland this fall. But then I thought a bit more, after realizing that my home and most places I frequent are still well within the new Home Area. I wondered who will actually be most affected by this change. So I pulled up the trusty Racial Dot Map and sketched the old (red) and new (green) Car2Go borders on top of it.

words

I learned three things during this exercise. 1) Manually drawing a bunch of lines on a screenshot is probably not the most efficient way to compare map data, 2) The old Car2Go map is much better drawn to scale than the new one, and 3) Asians, Hispanics, and Blacks will disproportionately be excluded from Car2Go beginning August 24. I took a quick look at a map of Portland income and earnings too, and was unsurprised to see that the areas cut appear to be overall poorer than the areas retained. And for reference, here is a cool map of the “Transit Oriented Development” score throughout Portland. (Note this map is significantly more zoomed-out than all the others here. Source here.) The areas excluded are much less well-served by public transit than the areas retained. Intentionally or not, Car2Go has chosen to offer the convenience of renting a car on the spot predominantly to white, wealthy people who already live near public transit.

About the change, Car2Go has this to say:

Based on careful evaluation of member feedback and historical data, we have determined that members are not able to experience the true benefits of car2go as the Home Area size results in vehicles sitting in areas, where they are idle four times as long as our vehicles in high demand areas. By updating the Home Area, we are confident that members will be now able to find a vehicle in the areas that they frequent most.

Of course, Car2Go can do whatever they want. But I wish they had dug a little deeper into their ridership data. I would love to see them market their service specifically to areas underserved by public transit and to individuals and families who might not otherwise have the privilege of a private vehicle. With an effort like that, Car2Go could make a real difference in the cities they serve rather than just giving lazy, wealthy white folks (like me) an excuse not to take the bus.

High Stakes Dice

Unlikely things happen all the time.

Here’s an example. Let’s say you are rolling a 20-sided dice. You probably won’t roll a 20. I mean, you might, but you have a 1-in-20 chance, which is only 5%. This argument works for any number on the dice. Yet, you will roll some number between 1 and 20. No matter what you get, it was unlikely… but at the same time, you were bound to get an unlikely result. Weird, huh?d20

Now let’s say you have a very funny-looking dice with 100 sides on it. Each number only has a 1% chance of coming up. So, let’s raise the stakes a little. Each time you roll, getting 1–99 is just fine. Nothing happens. But, if you roll a 100, you have to pay $10,000.

So, don’t worry! 99% of the time you will be just fine. Just don’t roll the dice any more than you have to—it’s a pretty boring game without any apparent reward, anyway—and try not to worry too hard, because statistics is on your side. Right?

d100

You’re curious, though. You wonder… how many times would you need to roll the dice for it to be more likely to get that 100, just once, than to avoid it completely? If you do the math1, you’ll find that 69 rolls puts you above the 50% mark. In other words, you are more likely than not to get a 100 if you roll 69 times.

Feeling lucky? Want to keep rolling? By the time you’ve rolled that strange 100-sided dice 700 times, you are more than 99.9% likely to get the dreaded 100.

Contraception fails much more often than 1% of the time.

Every time a woman has sex with a man, she rolls a dice. Depending on her contraceptive method of choice, or lack thereof, her dice has a different number of sides on it. But each roll always holds the possibility of pregnancy. Depending on her work, health, and insurance situations, she could be out a lot more than $10,000 in the coming year, not to mention having a child to raise.

Is your dice a condom? If you use them perfectly, that’s a 2% failure rate over one year. You only need to roll 35 times to be more likely than not to get pregnant2.

Is your dice a birth control pill? If you use them perfectly, that’s a 0.3% failure rate over one year. You need to roll 231 times to be more likely than not to get pregnant2.diceplot1

This is the absolute best case scenario for these common contraceptive methods. It is why methods like implants and IUDs with extremely low failure rates of 0.05–0.2% are gaining popularity. It is also why emergency contraception exists—think of this as a second “bonus dice” you can roll if you get unlucky with the first one.

We can play this game all day. Women play this game their whole reproductive lives. You can’t take our dice away. You can’t tell us not to roll (well, you can try, but it does absolutely no good). But apparently some employers can deny us access to certain dice and virtually all bonus dice based on a “sincerely-held belief” in junk science.

And yes, women could ignore our employers’ preferences, save our hard-earned money, and go buy whichever dice we like. But this game has a different set of rules. Suddenly we have to be able to afford the dice we want. Suddenly it is not the same game other women can play for free.

Someday, I hope all women (and men!) can have free access to all manner of highly effective, side-effect-free, reversible birth control. I know that doesn’t seem very likely to happen any time soon. But then again, unlikely things happen all the time.


The math is actually pretty easy. I’ll use the notation P(something) to indicate the probability that something will happen.
P(not rolling 100) = 99/100 = 0.99
P(not rolling 100, with n rolls) = 0.99n
P(rolling 100, with n rolls) = 1 – P(not rolling 100, with n rolls) = 1 – 0.99n
For this last probability to be more likely than not, it needs to be greater than 50%. So when we solve this equation for n number of rolls:
1 – 0.99n = 0.5
We get n must be 69. In other words, if we roll 69 times, we’re more likely than not to get a 100.
If instead we want to be 99.9% sure of getting a 100, we write it like this:
1 – 0.99n = 0.999
Which tells us n must be 688 (nearly 700). If we roll 688+ times, we are 99.9% likely to roll at least one 100.

Statistics from this site. Note that per-year failure rates are not necessarily the same as per-roll failure rates. Contraception failure rates are typically calculated as “the difference between the number of pregnancies expected to occur if no method is used and the number expected to take place with that method,” so while this analysis may not be completely sound, the take-home message is unchanged: highly effective birth control is incredibly important.