Jul 232011

I just read this NYTimes article on overhauling the US Postal Service, which lost $8.5 billion in 2010 (-12.5% margin), with that rate poised to increase annually. The recommendations seemed to address several parts of what the Postal Service calls its Ten-Year Plan, especially in eliminating Saturday deliveries, and tapping into its retirement fund to plug the budget gap. However, in the Form 10K from the most recent year, it’s clear that the decrease in revenue is a structural problem, and taking stopgap measure that reduce delivery costs by a mere 10% or so will not scale with the decrease in mail traffic.

I took a deeper look, especially at the USPS’s Cost Segments and Components Report, their Form 10K, and their Annual Report, and came up with some of my own recommendations. Looking at the total costs broken down by segments, the largest segments are C/S 3 Clerks and Mailhandlers, with $15.4B, C/S 18 Administration and Area Operations with $13.3B, and C/S 5 City Delivery Carriers — Street Activity with $11.4B. However, all direct costs of delivery segments, city and rural, including transportation, vehicle maintenance, and office costs, are around $29.9B. No other cost segment exceeds $4B.

So, first, let’s look at some low-hanging fruit by comparing to the 2010 Annual Report of UPS. UPS’s operating expenses for 2010 had $26B for compensation and benefits. Total. Total administrative (non-COGS) costs are around $939M. Contrast that with the USPS, where the Administrative Support and Miscellaneous component–a single component within the C/S 5 segment, costs the USPS $887M. Something seriously inefficient is going on here. I have not worked at or done any real look inside the workings of the USPS, but I notice three things.

(1) The USPS is way behind in modernization as compared to the private shipping/freight services. The only time I observed mail being processed at the postal service, I witnessed an extremely inefficient system of manual sorting. I once had the pleasure of talking with the local post office’s manager about how various processes works, and was stunned to discover that mail forwarding is implemented by taping a printed note on the appropriate PO Box or sorting bin. Really? And each time I go to the post office, it amazes me that it always seems to take about 5 minutes and hundreds of keystrokes, and rifling through lots of random bins and cabinets to even do the simplest transaction. And in the end, nor bar codes are printed and nothing is digitized to speed up the next step, besides perhaps a receipt of the transaction. Seemingly mocking how inefficient the system is, the cost of Mail Processing is only $753M, and R&D is exactly $0. Seems like an overinvestment in human capital over better solutions. How about investing in a modern infrastructure, thinking about more efficient processes, and instilling a culture of efficiency and continual improvement?

(2) And speaking of culture, what’s with clerks always standing around while the line is 15 people long? I understand that the USPS is generous with their breaks and benefits, but besides the ever assiduous mail carriers, USPS staff in general do not seem to be invested in improving their organization. I have yet to meet an unfriendly employee, but I also have yet to meet one who seems genuinely concerned when the line goes out the door. Even though the USPS is now technically private as a legal monopoly, it is still very tightly controlled by the government, and suffers from a culture of job security over accomplishment, and rotting from unnecessary top-down mandates.

(3) Administrative costs are sky-high. Looking at C/S 2 Supervisor and Technicians, the largest costs are Supervision of Collection and Delivery, Supervision of City Delivery Carriers, and Supervision of Mail Processing. Supervision? What does that mean? Making sure the correct letters go to the right places? Shouldn’t that be automated? Even if can’t be completely automated, isn’t a few billion dollars a bit steep? That’s something like $50-60M per state. Give me $50 and I’ll require all addresses by typed. Everyone has printers, so just make a super easy web site, with widgets to link with LDAP, Google Contacts, LinkedIn, and whatever else, and have it print addresses on envelopes. For those without computers, just market and sell cheap miniature $10 dot matrix printers that just prints envelopes.

So, instead of cutting out whole segments of your market (i.e. Saturday delivery), how reducing unnecessary costs first? Cutting Saturday delivery might still be a good option, but while the Ten-Year Plan calls for an improvement of processes, none of the appeals to congress even makes mention of it.

In addition to increasing efficiency, the USPS has struggled as of late to truly innovate. They are legally assured to be the only service to have a mailbox for every single household living in the United States, something I’m sure many delivery services would kill to have. The private delivery services have all standardized shipping labels and shipping procedures as they attempted to reduce cost and increase reliability, but for some reason, that didn’t catch on with USPS. You can use printed shipping labels, but because written addresses are still permitted, most of the human energy goes into sorting those badly written ones. Require people to print addresses. Then the post office would be the perfect player to move into the address management space. Not one company even remotely has the address data that the USPS does. So use it. Allow people to opt into releasing some of their high-level non-PII mail data, and use that for collaborative filtering, especially for local networks. Again, no company even remotely has that kind of data. Privacy advocates might freak out a little, but of all things I trust the USPS to get right, it’s privacy, so having an opt-in system for a system where I can see which cable service or charities are most popular in my neighborhood is worth it as a consumer, and, for the USPS selling licenses to the data, should be worth it as well.

Jul 142011

The topic of grade inflation always interested me, since there are so many variables to consider. The New York Times Economix blog has a short piece on a recent study.

From reading the article and the insightful comments as well, it raises many points I have considered before, but raises several new ones.

The important issues to consider when looking at something like grade distribution data can be broken down to a few categories.

Evaluation Metrics

First and foremost when looking at statistical evidence that an event exists is to see whether what they’re measuring even makes sense. In this case, the primary metric–the distribution of grades–is very well documented and indisputable. We are getting significantly more A’s than ever before.

However, some of the metrics used to give reasons why this could be happening are not so indisputable. It cites an earlier study that conclusively found a correlation between private schools and a 0.3 higher GPA as proof that private schools systematically have grade inflation, when controlled for selectivity. It also states that students have been spending less time doing works, as a proxy for how much students are actually accomplishing for the same grade.

Both of these lines of reasoning are flawed.

There could be many reasons why private schools have higher grades than public schools–better teaching environment, financial pressure to issue higher grades, more rigorous selection criteria for selecting students (since the study mostly looked at SAT and acceptance rate for selectivity), and better study environments. Just as in the global warming debate where it’s crazy to take seriously the talk of a wide-ranging unspoken understanding between individual climate scientists to push forth a collective agenda, it’s seems unlikely that there is a collectively shared attitude among private school professors to push grades up any more than those in public schools.  Granted, separating out for-profit schools may yield more interesting results, but having met plenty of professors in both public and private colleges, it seems absurd to imagine that the ones in private institutions are somehow being controlled by unscrupulous administrations, and would be willing to play along with those administrations.

And the use of time spent on working? It measures fact that I can write papers three times faster with a computer than with paper and pencil. I measures the fact that I no longer spend hours organizing note cards and arranging sources when researching my paper. It measure the fact that when I am doing research, or just looking up something I don’t understand, it takes me minutes to search on an online site or database, rather than an hour to find the right books in the library. I strongly believe we are becoming more productive.


I do agree on a few key premises–that competition for students, teaching evaluations, and the reliance of grades in hiring in certain industries all contribute greatly to grade inflation.

The competition for top students is real, and is affected by grades. Since Princeton implemented their bone-headed grade deflating policy, there has been dissent among current and potential students in how it affects them.

Professors also gain tenure, awards, and recognition from teaching evaluations. The importance of these varies from college to college, but more and more schools are implementing mandatory online course evaluation systems that students much complete after every term. This constant feedback almost undoubtedly leads to better teaching overall, but because of the potential rewards attached to good evaluations, and the close correlation between grades and evaluation, professors do have a motivation to hand out higher grades.

After graduating, when looking for jobs, many top employers use GPA as a metric for hiring employees. Many of these companies frame this requirement as evidence that they care deeply about critical thinking and academic pursuits, but many times, the way it is used is a mistake. Companies who say they only hire those with 3.7 or higher GPAs are encouraging grade inflation as well as using GPA inappropriately as a proxy for skill and experience in their field.

Segmentation of Data

Going back to the issue of private and public schools, private schools often have significantly higher tuitions as compared to a public school. The super-elite private schools now all generally have nearly 100% financial aid for families needing it, but the vast majority of private schools do not have the multi-billion dollar endowments and the hundreds of millions in annual donations that makes that kind of financial aid possible. Hence, using acceptance rate to compare selectivity of colleges is not valid, since there are fewer applicants to private colleges. Using objective scores like the SAT are also invalid. Many public schools admit based heavily on objective criteria–SAT scores, GPA, etc. But many private institutions do far more–heavily considering teacher recommendations, evidence of a motivated person, essays, leadership qualities, interviews, and more. Success in college is not the rote study of facts that is often what numbers like the SAT and GPA represents, so these other qualities that are reviewed more often by private schools may lead to selecting higher quality students given the same SAT/GPAs.

It’s also true (perhaps unfortunately) that the vast majority of top colleges in the United States are private. With some of the top students from the country in its courses, it is often the case that almost everyone will do very well in a difficult course. Should we try to average out their grades to a B? As a Stanford Computer Science graduate, and having seen the curriculum and evaluation of other Computer Science programs across the nation, I am actually appalled that many programs would even give a B to a student who does well in some of their courses. The same student would most likely straight-out fail the equivalent Stanford course. Yet, Stanford students are being differentiated into buckets of grades, sometimes based on something as small as a single mediocre problem set out of ten, because we see the dire warnings of grade inflation.

Many people also mention that there is a grading difference between humanities and sciences/engineering. I have observed this as well. I’ve thought of many reasons why this is the case. The most important one–and this will make a lot of people mad–is that in general, humanities courses are just not as rigorous as science and engineering courses. I know plenty of engineers taking graduate-level history and English courses for fun, but I have only once or heard of someone doing it in the other direction. Humanities degrees are also much less useful in the marketplace–this is directly from statistics; not making a value judgement here–so departments are constantly trying to convince students to take their courses as their departments face a constant existential crisis. This leads to far more fluff courses than in engineering, where enrollments are generally very high, and professors are overwhelmed by just teaching the core courses.

I’m not trying to be dismissive of humanities–I actually think that more scientists and engineers ought to take more humanities courses to diversify their view of the world–but from what I’ve seen, there is a gap between the rigor of engineering courses, and the rigor of humanities courses in general.

Quality of Students and Teaching

College admissions have become more competitive as more people have been deciding to get a college degree. A greater supply of college applicants may supply the colleges with better students. If grades are supposed to reflect the true abilities of students, and not a normalized ranking, should the GPA not rise?

There has also been a resurgence in promoting the importance of teaching in schools, as we have realized how bad our education system, and how difficult it is to battle for the bright students. I’m an optimist, so I’d like to think that this focus on good teaching has had an effect in the form of more learning.

Most Probable Theories

Amazingly, there are some very plausible theories for grade inflation that are just not mentioned much. Perhaps it is because these are not as provocative or sensational as accusing the private schools of operating a cabal to put footstools under their GPAs, but in any case, it seems to me like a combination of these almost certainly should explain why we have grade inflation.

The most important reason is just simple escalation, like nuclear warfare. If most schools have a 3.3 GPA, but a few select, “better,” schools are giving out an average of a 3.5 GPA (for any reason, whether it’s because of better students, attempt to get money, etc), the schools with the 3.3 GPA will want to shout to the world that they are relevant–just as relevant as those better schools. They invest money in new centers, new facilities, better researchers, but they would really like to (1) attract better students, and (2) have their students go on to do more interesting things once they graduate so that the school can boast in its marketing material that they have X number of students earning over $100,000 a year just a few years out of college. Whether implicitly or explicitly, grades go up to assist in this. So slowly, the 3.3’s catch of to the 3.5’s. But of course, it doesn’t stop there. There is always a temptation to up the grades just a little.

Once the grades reach a level, it is imbued with meaning. These days, B+ in most places probably means “average.” This is quite unlike the 1940s, when a C was imbued with the connotation of “average.” These days, an A no longer means great, but in most places merely means “perfectly satisfactory.” The only way to say that a student went beyond the call of duty is an A+, which many schools have started assigning a 4.3 or 4.5 GPA, since there’s not much room to navigate from 4. Once these attitudes are set, it’s difficult to reverse them. Reverting back to a C-centered scale, and giving good students B+’s would feel like telling good students that they are average, even though the scale is totally arbitrary.

Conclusion: This is Silly/Let’s have A+++’s

The debate over grade inflation and its causes seems fraught with political agendas and an inexplicable blind eye to probable causes. The debate seems centered not on how it actually affects society (I don’t buy that it does that much), but rather on who’s to blame, and what evil motivations people have.

At the end of the day, the grading scale is arbitrary. Just as in real monetary inflation, as long as the rate is not too high, the current value of $10 or an A is still on a relatively well-defined scale. It’s only when we start comparing values across history does it truly matter. And, if you are comparing someone’s A they received today with an A received 20 years ago, whether for a job application, to assess someone’s character, or anything else, I think that you have more to worry about than grade inflation–namely, your evaluation methods might need some work. Many today are also scoffing a the abundance of A+’s and 4.0+ GPAs. I fullheartedly applaud the addition of GPAs past 4.0. Inflation works because the scale of money is still linear, and the amounts are unbounded. $20 is twice that of $10 no matter what decade I’m in. Grades unfortunately don’t work like that, largely because there’s this artificial cap at 4.0/A. So, let’s increase the range, and then actually be able to reward and differentiate students for exceeding expectations now that everyone is getting B’s and A’s.

I’d love to see transcripts with lots of A+++’s and A+++++’s in the future. Maybe the grading scale will just be transformed into an A1-A7 scale, where the number is how many plusses after the A the student received.

Jul 082011

Between my main blog and my teaching blogs, I’m on-track to hit my bandwidth cap again this month. So, I’ve finally taken flight into the world of cloud storage. I’m using the “Amazon S3 for WordPress” plugin to seamlessly connect my S3 account to my blogs. You, the reader, shouldn’t notice any differences, but this will save me around $25 a month. Good times!

Jun 232011

I’ve randomly run into this song half a dozen times in the past few days–playing in Walgreen’s (!), at a BBQ at Stanford, from one of the AI counselor’s computers, and a few other places. That combined with the fact that I don’t have a single Arcade Fire song featured here led me to feature this song.

The lyrics here really got to me this morning as I was making the 8-mile, 30-minute drive to campus in typical Bay Area traffic. I do have dreams of impulsively, unreservedly packing a pack and just disappearing off in Big Sur for a few weeks. Or months.

Fortunately, I like what I’m doing now even more–teaching AI to really, really cool, smart, and motivated high schoolers, and working part time at Palantir Health. Even the 4-5 hours of sleep a night feels fine. I haven’t felt this happy with my daily work in quite a long time.

Jun 182011

I recently began seriously trying to learn Javascript and JQuery. Up til now, I knew just enough to tweak a script or make some WordPress behavior changes when needed, but if I were to do a complete project in Javascript from start to finish, it would probably look like a mess of PHP. (ooh, zing!)

In case anyone out there is in a similar position, with a reasonably solid coding background, but looking to beef up Javascript, I found the following resources really helpful:

  • Javascript: The Good Parts by Douglas Crockford — Most JS developers have probably heard of this book, and for good reason. This book really helped clarify how the language should be used properly to avoid many pitfalls I saw myself running into while playing around.
  • In fact, Douglas Crockford is kind of the man when it comes to Javascript. His entire web site is full of goodies.
  • http://jsfiddle.net/ — Great IDE for playing with the language, different platforms, and of course the HTML and CSS on the page. Hooks up everything you need to quickly play around with ideas quickly.
  • The Javascript Guide on the Mozilla Development Network–very complete, well-written, and well-organized.
Jun 122011

I went to the Dispatch last night at the Greek Theater in Berkeley. It was one of my favorite bands who I’ve waited ten years to see (with no promise that they’d ever do a reunion tour post break-up) playing in perhaps my favorite musical venue. Seeing them was like like being shown that a unicorn actually existed. They blew me away with their artistry, rotating instruments and improvising and feeding off of each other. It couldn’t be more obvious that they were loving what they were doing, just being there and making music and playing their instruments with each other.

The only downside of the night was when a crowdsurfer (and my, were there many of them) fell head-first into the ground about ten feet away. He got up dizzy, but was still visually tracking and responsive and oriented. I ended up just telling his friend to take him to a first aid tent and check his pulse every so often to make sure it didn’t go up, and watch for nausea or irritability. Hope he still had a good night after that incident.

Here’s them performing last night:
Dispatch at the Greek Theater in Berkeley, CA

Get: Silent Steeples mp3/CD (Reissue)

May 312011

A few weeks ago, I was walking past the Lorry Lokey labs and the Mudd Chemistry Building on Stanford campus, when I noticed how interesting the bamboo that lined the exteriors of the buildings looked in the evening light.

If you just let your eyes drift, the bamboo is so thick that it creates this relaxing background.

And if you look closer, and near the ground, where it’s dark, the bit of light that filters in gives the bamboo shoots some incredible highlights.