More HN numbers

I’m curious whether the Hacker News post will lead to long-term increased traffic.

Edit: 10/30/2012. After the post regarding Hurricane Sandy, this experiment had to end; traffic from the new post started to matter. These numbers are the total hits to all-things measuredmass, not just the Google post. So, the above is the final plot. 

By October ~26 (that spike) traffic was coming in bursts from various countries, especially Germany. I’m not sure that the >10 hits/day rate would have remained sustainable for more than another few weeks. Time will tell. 

Hacker News

140 countries in a day! 30,000 hits!

Whoa.

I didn’t expect that post to reach the top spot on HN, let alone stay there. It would’ve been neat just to ping around on the front page for a bit before falling to obscurity. I’m glad you’re curious about it too.

Now I’m curious about you : ).

Continue reading

Google Homepage Size Over Time

This afternoon, I looked at the source of the main Google page with my browser. It spans my whole monitor and takes 15 pushes of the scroll-wheel to see it all. Whither minimalism?

Off to the Wayback Machine. One must be careful to avoid including Wayback Machine Javascript and comments when assessing file size. I think I got it right. File sizes were assessed with `ls -l`, and included only the size of the ‘.html’ file, no images or fanciness. Dates were sampled pseudo-regularly, trying to get about two points/year.

6 points corrected. Thanks ‘qxcv’! Archive, curl, and browser data noted. Thanks ‘zzzwat’!

Time for a log plot. The difference between a browser-acquired page and a simple curl request is an order of magnitude.

Fun with Torque

\vec{\tau}=\vec{r}\times\vec{F}  ?

Nope, this kind of Torque – an excellent Android app that interfaces with the OBD II port in your car.

I’d bought an inexpensive “LM327” bluetooth adapter for my car months ago, but never got it running with my laptop. The Android solution worked trivially, and the application works well.

If you want to play with datalogging with Torque and you’re using the free version, you must explicitly select each sensor every time you use the application. Tedious, but free.

I finally went on a real drive with the car (~10-15 mi) and got logging working. There are many sensors to log, but most of them don’t betray fundamental properties of the car quite like these two do.

Lines are not fit. I tweaked the 5th gear line to match the 5th gear data, then used the gear ratios from here to predict the lower four lines. The ratios used were 3.545/2.111/1.448/1.088/0.780. Wikipedia has two values for the 5MT. The other ratio stated for “All 1990-2003 with EJ engine” of 3.785/1.945/1.500/0.994/0.780 is in obvious error when plotted with the same procedure. I submitted an edit to this effect; time will tell if it sticks.

Same plot, with incorrect ratios.

The GPS and my engine computer’s speedometer are neatly locked, but there’s about a 1 mph constant offset between them (easier to see in the residuals). Neat.

What’s the state of the academic Physics job market?

“Look! There are jobs in Physics Today again!” said officemate Matt as he leafed through the new September issue.

I had a quick look, and sure enough, there were way more than last month. Great.

But, there’s a very strong annual cycle to hiring in Physics. I’d only ever observed it qualitatively. Since I will need a physics job relatively soon, I decided to make my understanding more quantitative.

Thanks to our lab’s collective archive of Physics Todays (thank you Anne, Claire, Dan, Michael, Matt, Nikolai, and Scott!), I compiled the plots below.

First, I made a detailed survey of the past year’s hiring.

There are many ideas one may learn from this plot, but perhaps the strongest signals are that hiring is very seasonal, and that the majority (it’s a logarithmic plot!) of the advertised positions were for faculty in the main hiring sequence (advertise Autumn, apply Winter, interview Spring, start late Summer).

Are there more jobs this year than last year?

Guess not.

Compiling detailed breakdowns by job type is tedious, slow, and subjective. Most interesting to me was the total count.

How does the number of advertised jobs now compare with the past?

Systematic uncertainty?

Chatting with Dr. Tolich about the plot brought to light an important systematic concern: Free online advertising is becoming prevalent (SPIRES has a job mill, and there are others (edit to add one: academicjobsonline.org )). The minimum charge for a Physics Today ad is $525. There are rumors that at least one major department discussed not placing a print ad for recent positions on cost concerns. If job seekers find jobs on the web, why pay for print? I’d never heard of the SPIRES listings, so at least some luddites still prefer print.

Today (September 20, 2012), there are “272” jobs on physicsjobs.com, the online branch of Physics Today (but they charge $525 for online listings; a minimum-length print ad is free). SPIRES claims to have 1420 jobs. Academicjobsonline.org has 76 in narrowly-tailored “Physics”. Make of the situation whatever you will.

With more time/interest, I have the archives needed to add other useful months (October, December, January), but the initial curiosity is sated; there are fewer print-advertised jobs than there used to be.

Appendix: Counting

What counts as a job, in the counts above? An advertisement for “several” positions, without quantitative reference to assured positions, counts as one. “Two or more” positions count as two. Blunderbuss advertisements from National Labs count as one (they typically have many positions open, and I’m not convinced that the ~four highlighted jobs reflect the number of open positions).  Only post-graduate positions count.

Counting is slow, and I have a thesis to write, so I haven’t recounted each point. Recounts (was that 87, or 88? Arg.) of a few Novembers suggest a ~±2% variation in counting/subjective judgement.

A word to hiring committees: “Two” or “three” sounds much more concrete to an applicant than “several”. “Several” sounds like the department doesn’t yet have university approval for multiple hires.

Does the Gravitational Constant depend on Latitude?

Newton’s Gravitational Constant has been the subject of controversy for decades. The measurements are very hard. The CODATA value’s uncertainty has been revised upwards and downwards, and may soon go up again. Three recent measurements ( <– three links here) have added tension. All three disagree with one another, and two disagree strongly with the accepted value.

There have been suggestions by theorists that the value of G might vary spatially on Earth’s surface. It’s difficult to reconcile such a notion with other precision tests of gravity, but there’s no substitute for experiment.

The data used for these plots are from the experiments noted here. The latitude and longitude are approximate and use the google maps datum.

Can the new precision experiments’ disagreement be explained by spatial variation? For model-independence I don’t attempt any fits. Plots will have to do.

Image

Image

Should I switch to Ting?

Ting.com is a new virtual wireless company. They resell time on Sprint’s network, and use a pay-as-you-go model of billing. You buy your own mobile device, pay a monthly access fee, and then pay only for whatever service you consume. It’s the way of the future.

Future or not, it could still be overpriced. Hence this plot. Sprint theoretically roams on Verizon, and only Verizon offers the mountain coverage I require, so AT&T’s not plotted.

Image

Complicated, right? Since there are three variables (minutes, texts, data), and one price, the natural realm for this problem is in four dimensions. Note, for giggles, that 44640 minutes (or less) is a month.

I retreated back to the problem I’m personally interested in: Should I switch? Note that this plot does not include the significant (or subsidized) cost of a phone. I use about 500 minutes of voice a month and send more than a hundred texts.  How much data can I use before Ting is no longer the optimal choice?

My current Verizon plan is ~$40/month before fees with no data. Ting beats that handily. The steep light-blue line must be a way that Ting (or their Sprint overlord) plans to make money. Ting is a great deal for small usage, but very poor for lots of usage.