“And I’m like ‘Honestly, I don’t know nothing about mopeds’” -Macklemore
Mopeds are a thing. A thing that exists. A thing that has had approximately zero influence on my life. People in Europe drive mopeds, people in America drive literally anything else. But then I watched the music video of Macklemore’s & Ryan Lewis’s new song “Downtown.” The video begins with Macklemore naively purchasing a moped which he proceeds to ride around the ‘Downtown’ for the remainder of the video.
“Oh yeah mopeds, those exist” I thought, and headed to Google to see if a moped is right for me [Spoiler: It is not]. But then I wondered how many other people were similarly inspired to dive into the fabulous(?) world of mopeds.
Google, of “Don’t be Evil” fame, obviously tracks every search performed on its site and the aggregated data is available through Google Trends. Below is the Google Trends plot for the search term “moped” from 2004 - present in the United States. An interactive version of this plot can be found here.
The most prominent feature is the large spike in the summer of 2008. When the financial crisis hit people were thinking of ditching their Hummers and opting for something that gets 70 mpg. What is curious is that this was not a persistent trend, even though people were supposedly strapped for cash for the next few years of the recession. Everyone did their moped research in 2008 and decided that mopeds were not for them. Conversely, everyone bought a moped in 2008 and they all still have them and have no need to ever search for anything moped related. Anecdotally, the number of mopeds I see is still approximately none so I will choose to believe the former.
Because Google Trends only displays relative search numbers, and the 2008 spike was such a huge outlier, I chose to rerun the analysis from December 2008 until the present to achieve greater “resolution” to answer the question of interest. Explicitly, how has “Downtown” affected searches for mopeds?
Because searches for mopeds are cyclical, we can average these to get an idea of how Macklemore’s song has deviated interest in mopeds. Here is a plot of the average number of relative moped searches by month (bold). Each individual year is also shown, with 2015 highlighted in red.
As you can see, 2015 has been a bit below average, but there was a small increase at the end of the month in August, which is when “Downtown” was released (indicated as a dotted line). Then in September, instead of dramatically plunging as Fall approaches, the number of searches for mopeds stayed relatively stable. September’s search traffic scored a 74, almost 17 points higher than the average for this time of year, and statistically different than the mean. More people than usual are actually looking for mopeds because of this ridiculous song.
There is no way to know if this increase in traffic will actually lead to more mopeds on the streets, especially when Winter is Coming. Nevertheless there appears to be a causal relationship between this song and moped searches. As “Downtown” continues to rise in the BillBoard Top 100 (it has only been on there since the September third and is currently sitting at number 16 at the time of writing) the last question we have to ask ourselves is: How much is Big Moped paying Macklemore?
Click Below to Watch the “Downtown” Music Video.
[Dusk. Three 20-something white men sit around a rectangular kitchen table, several piles of white and black cards stacked in front of them. Oh and some Busch Light cans.]
This all started when no one would play Cards Against Humanity with us. So we started pulling pairs of black and white cards off the stacks and sharing the ones that we thought were funny. “Someone should make a Twitter bot that automatically posts random combinations of these,” I thought aloud. No one responded because that is a stupid thing to say at a party. When we finally finished all of the black cards I went home and did it (two days later) anyway.
To be fair this HAS been done before. @CAH_bot is a fine example of one. And its about 1600 tweets ahead of me. But I did it anyway because the world deserves more Twitter bots (and because I didn’t find that account until after I did all this work).
How did I do it? Well I’m glad you asked. After I tracked down some text lists of the cards, I imported them into Excel and used its RANDBETWEEN() and INDIRECT() functions to pull cards from each list and paste them together. Once I copied them into a new Notepad document I used the code in the appropriately titled “How to Write a Twitter Bot with Python and tweepy” tutorial to automatically post to Twitter. All I had to do then was create a Twitter account and away I went.
So here it is: @bot_CAH
This little guy is more of a rough approximation of a Twitter bot. First, I should probably write up some Python that automatically generates the posts. Also it currently posts every 15 minutes, but only when I am using my Surface. So it won’t completely spam your Twitter feed. In a perfect world I would have it post every hour from a constantly running Raspberry Pi (basically just a tiny $35 computer that’s useful for things like this).
This bot works in a completely different way from my first foray into Twitter automation. My other bot, @PH_papers, is based off this post and uses dlvr.it to automatically update the account based off a Google Alerts style search from PubMed. I would recommend you follow it if you are interested in hearing about the most up-to-date Pulmonary Hypertension research. So far most of my followers are doctors from Mexico. That’s how you know you’ve made it.
In closing I would just like to say that this was a fun little experiment and that it has helped to reveal some deep truths about the universe.
This post was originally hosted on the American Society for Pharmacology and Experimental Therapeutics Blog - PharmTalk. The original post can be found here.
Computer Skills for Scientists
The dawn of the computer age is over. Computers are no longer just hobbyist toys or specialized equipment; they have become an integral part of the scientific process. In fact you are almost certainly reading this on a computer. As science’s reliance on computing grows, so too will its need for able-minded computer users. Graduate school or post-graduate training is not only a time to develop your bench top skills, but also a time to focus on honing the “soft skills” of independent critical thinking, teamwork, networking, and scientific writing. Allow me to propose that you should also spend some of your (limited) time expanding your computing prowess.
Streamline Workflow and Save Time
Many commonly used programs, like Excel or ImageJ, allow you to write macros that will repeat a task over and over. Macros are sets of instructions that tell a program to repeat a specific sequence of tasks. Need to perform the same analysis on hundreds of images? A macro might work. Do you need to import, manipulate, and format a series of data sets in Excel? Try a macro. Some require a bit of coding knowledge, while others are able to record a sequence of clicks. You could potentially save yourself hundreds of hours by automating some of your more monotonous work. Take a day to look through your favorite program’s documentation or help files to learn what options are available to you. When in doubt, Google it. Chances are someone has already done something similar.
Improve Clarity of Scientific Reporting
Mistakes are inevitable, but they are a huge problem for data reproducibility. Automated computing tasks help to remove unintended biases in data analysis by essentially “blinding” the researcher leading to increased reproducibility. Journals are pushing for researchers to publish their raw data and program code. Peer review only works though if you, as a trained scientist, can interpret that computational information. Systemic statistics illiteracy is also a common problem that is closely tied to computational proficiency. The more you practice and explore a particular program, the more positive you can be that you are using the appropriate statistical tests. If you are familiar with GraphPad, check out their thorough documentation which can answer most of your stats questions.
Open New Doors to Different Career Options
Bench top skills make you a good technician. Pick your favorite bench top technique. Are you really good at it? Then chances are high that so is everyone else in your field. Your competition likely has a similar publication record. They probably even took the same basic classes as you. How are you going to stand out in the academic field? Computer skills are, as they say, another tool in your toolbox that may help you stand out from the crowd. Universities are looking for faculty and post-docs that will add something to their team. That something could be working knowledge of a unique (and hopefully useful) computer program.
While almost any skill may give you a competitive edge when applying for jobs in general, certain careers are looking for scientists with specific skills. Clinical trials and high throughput screening create monstrous mounds of information. While the clinicians and technicians are running the experiments someone has to translate the results into information that the decision makers can use. Data management tools like SQL, R Tool, SPSS, or SAS are great ways to practice making sense out of huge amounts of information, are broadly applicable, and are sought after skills in industry. “Big data” is an emerging field that may provide a potentially satisfying career path for PhD level scientists.
If you are interested in lab management as a career post graduate school then experiences with software like Labguru and Quartzy are good places to start. Because this software can link specific reagents to each protocol they can add reproducibility to your current lab’s data too. Software like these allow for better sharing of data between lab members and/or collaborators. Finally, “Open science” tools like the Open Science Data Cloud allow researchers to access public datasets as well as make their own data available to collaborators across the planet.
Data visualization is a trendy new field that revolves around transforming raw numbers into thought provoking graphics. Journalist organizations, both traditional and web-based ones, need people who are familiar with turning complex scientific ideas or large data sets into attractive and informative images. Professional biostatisticians and bioinformaticists also should be able to communicate their complex results to diverse audiences. Plotly has a Python library all about visualizations that might be a good place to start.
While not every job outside of academia will require a PhD and each discipline has its own set of heavily used programs, it would be unreasonable to try and learn them all. However, the simple act of independently learning a new skill demonstrates that you are driven to learn and can readily adapt to whatever technology is presented to you. Data visualization is a trendy new field that revolves around transforming raw numbers into thought provoking graphics. Journalist organizations, both traditional and web-based ones, need people who are familiar with turning complex scientific ideas or large data sets into attractive and informative images. Professional biostatisticians and bioinformaticists also should be able to communicate their complex results to diverse audiences. Plotly has a Python library all about visualizations that might be a good place to start.
While not every job outside of academia will require a PhD and each discipline has its own set of heavily used programs, it would be unreasonable to try and learn them all. However, the simple act of independently learning a new skill demonstrates that you are driven to learn and can readily adapt to whatever technology is presented to you.
How to Get Started
One option is to ask your PI if they or your research group has a website. If they do not, help make one to showcase your lab’s current research and makes it accessible to potential new members browsing the web. You can also make a website for yourself. This can serve as both an online resume and tangible display of your HTML/CSS abilities. A completed website gives you a concrete “finish line” to your independent study. Plus with websites like Squarespace creating a new website is easier than ever.
You could enroll in a computing class at your university or seek out online courses. Codecademy has free lessons on programming basics and website design. Another potential avenue for learning is Software Carpentry.Software Carpentry is a global network of courses and instructors that seek to teach basic computational workflow skills to scientists. Classes like these can help you to start thinking like a programmer and open your mind to potential uses for code in your own research.
These are not your only options. You could make a video of one of your laboratory methodologies for YouTube. Learn a new program to calculate ANOVAs. Start a podcast with members of your lab. Ask someone in another lab if they can show you what software they use. It doesn’t have to be a huge time commitment. You can get a free week-long trial at Lynda.com where they have hundreds of videos on cutting edge software, but you have to start somewhere. Even practicing some of the advanced functions in Excel can go a long way towards being an effective, tech-savvy scientist.
Is this list comprehensive? No. If you learn a new skill are you guaranteed a job? No. But computers will only become a more integral part of quality science. With a toolbox full of computer based skills you will be able to more nimbly assert yourself in the job market and the scientific community. The key is to try something new that you find interesting and maybe you can turn a hobby into a new career.
I know what you are all thinking: which state has the most beautiful college-aged women? A question for the ages to be sure, one that if I ask again in a few years would probably end with me in jail. Total Frat Move has been faithfully performing the Lord’s work and compiling a database answering that very question.
TFM’s ‘Instagram Babe of the Day’ showcases one Instagram account featuring a woman wearing clothes that her father must...just...not know about. Using the tools at Kimono Labs I assembled a list of 299 IBoDs and sorted them by state. Here are the results in a fraternity-approved-salmon colored map:
Florida is the winner and it is not even close at 67 ‘Babes’ of the 299, or 22%. California is the next closest at 49 followed by Arizona at 38, Texas at 21 and then Virginia and New York with 8 and 7 respectively. In fact together the top 3 states beat out the bottom 48 (including DC) combined. The entire country of Australia had 1 which puts it ahead of the six states with zero. America’s friendly neighbor to the North, Canada, came in at 4 beating out 34 states.
Clearly not all states are equally gifted. Warmer states seem to be dominating and the upper portion of the Midwest is just a deserted wasteland. But Wyoming only has 6 people per square mile. I didn’t expect there to be bronze bombshells patrolling the streets of Cheyenne. But even adjusting for the number of female college students per state does not even the playing field. Here is a graph of the total number of IBoDs plotted against female college students by state (data are in thousands, from 2008; available at the US Census. Thanks Obama).
Florida and Arizona are really just doing a great job. Keep it up gals. Texas and especially New York are under-performing their potential. Both have fewer TFM ‘Babes’ than would be expected based on their student population. Hawaii, with its low student population, actually has the highest rate of IBoDs. Around 1 in every 8500 college girls is on TFM. Florida’s rate is the only close competition at 1:9000 followed by a distant Arizona.
Babes of the Day is only one metric though. Is it really accurate in predicting the most beautiful states? I looked up some other rankings and chose two (the first two that came up on Google). The first was from the internet’s bastion of journalism, Buzzfeed, and the other was from Comcast Xfinity (even they have lists?). Both used a combination of arbitrary numbers to reach their equally dubious conclusions about the most attractive states. Here are the top 5 for each in descending order. (I switched back to using total number of IBoDs for this one because it gives a better list. Rigor is not the goal of this analysis)
No surprise, California shows up on each list. We also see Florida and Virginia show up on one of the other lists. The other two lists have Hawaii which would have been there if we had taken into account total number of college girls. So overall 3 of our top 5 ended up on the other lists which gives us a 60%. Plus 10% extra credit for Hawaii almost being on there gives us 70%. And C’s get degrees.
There is no way to differentiate between correlation and causation. Either warm weather attracts beautiful women, makes women beautiful or coerces them to send their names to TFMintern@grandex.co. Who cares though? I’m not complaining.
How old are you? How old does your computer think you are? That is the question Corom Thompson and Santosh Balasubramanian over at Microsoft have been thinking about. With Big Data it probably is not too tough for a computer to crunch the websites you visit and the purchases you make online to spit out a reasonable guess for your age. That’s how targeted advertising works. But what if computers only had pictures of you? Behold how-old.net. Just like that carnival “game,” can a computer guess your age by just how you look?
Um…I guess not. But this was only one trial, so I decided to try again…seventy-five more times. All with pictures of myself. This quickly devolved into quite a narcissistic project. This involved delving into the photos on my Facebook profile which proved to be both a hilarious and horrific endeavor. After shamefully mining my own photographs I ran a simple regression to get a rough idea of how the algorithm was doing.
Just for fun I decided to break the images into pictures where I was wearing glasses and pictures where I was not. If the algorithm is reading facial features glasses could be affecting the results. This divided the images almost exactly in half.
While exploring the wild world of Twitter I decided to investigate the following link:
“Uh oh, I probably should have cashed out my Robinhood account today” was my immediate thought before the page even loaded. What followed was a short blurb about presidents and then a long list of each State of the Union (SOTU) and the corresponding percentage change in the Dow the following day starting from 1961 copied from The Wall Street Journal. A list of numbers? That is just begging to be played with.
Here is a graph of the average percent change for each president. The bars are Standard Error from the Mean. (Guess what the colors mean):
Ouch. Looks like on average President Obama’s updates have been a net loss. President Bush’s speeches on the other hand have been a boon for our economy. And we should have had President Ford giving SOTU speeches every day. He had the greatest average effect on the stock market (or at least the Dow) with a whopping +0.61%. It is possible that was a big effect in those days but now we see that much volatility on a below-average Tuesday.
The overall average for the speeches was -0.05%. Instinctually this makes sense. The SOTU is a chance for the President to remind us of all the shitty things happening right now and what wacky plans they have for fixing them. This is understandably going to inject uncertainty into the economy. Republicans have a slight tendency to bring the Dow up, +0.003%. The Democrats, -0.112%, not so much. If you think about what Republican and Democrat presidents might highlight (Tax breaks vs. Social Spending) this is also intuitive. Basically it becomes giving wealthier people (read: people who invest in stocks) money or taking it away and giving it to those most in need (i.e. people who would rather spend their money on rent or Taco Bell than a share of Google stock).
I also ran some regressions to see if the parties of the House and Senate mattered (they don’t) and to see if it mattered whether the House, Senate, and President were in the same party (also a no). What’s my prediction you ask? I bet Barack will stick to his guns and we will see another drop tomorrow, especially with all this talk about raising capital gains taxes, but probably not one that is out of the ordinary in magnitude. With a Republican House and Senate the likelihood of any changes in the capital gains tax impossible anyway. So the Dow should chill out.