Here is the revised version of my paper "Robust Timing of Markdowns", written with Michael Dziecichowicz and Daniela Caro. It considers a product sold over a selling horizon at various decreasing prices by a retailer and applies robust optimization to the uncertain demand arrival rates. A key feature of this paper is that, because the time during which the product is put on sale at a given price is a decision variable, the argument of the (concave) budget of uncertainty is itself a decision variable, introducing non-convexities in the formulation. We show how to solve the problem in a tractable manner. We also show that, under a mild condition, there is one optimal sale time (two distinct selling prices) in the selling season when the budget of uncertainty is concave. We also implement a dynamic policy and demonstrate its potential on numerical experiments, and extend the setup to the case of multiple products.
The National Academies Press recently published a report on "The Postdoctoral Experience Revisited", which was also summarized in Science magazine. (The committee chairman was interviewed in the magazine here.) This work updates a 2000 report called "Enhancing the Postdoctoral Experience for Scientists and Engineers".
The sources of concern have not changed much over time (my favorite excerpt of the report is in the summary: "Is it really necessary to remain in training until their mid-30s before being qualified for his or her chosen career trach? Are these highly qualified PhDs researchers receiving the recognition and remuneration that they deserve? Is there an appropriate balance between the number of postdoctoral researchers that are trained and the number of jobs that require postdoctoral training?"), but the number of PhDs has increased quite dramatically since the 2000 report.
The following particularly caught my attention: "One important finding is that the postdoctoral experience differs considerably among types of institutions. Compared to postdoctoral researchers working at universities, postdoctoral researchers who work at national labs or in industry are typically paid much more, remain for shorter periods, and are often offered fulltime jobs at the end of their appointment. Likewise, postdoctoral researchers who are on fellowships or traineeships have higher salaries, better mentoring, and more control over their research than those who are working under a principal investigator's research grant. The majority of postdoctoral researchers are working under research grants."
The Science summary packs a punch too: "Only a minority of the postdocs working in university labs have opportunities to receive high-quality training from eminent senior researchers, develop their own research ideas, gain experience in lab management and grant writing, acquire contacts and a publication record and, ultimately, move into a tenure-track position at a research institution." Ouch.
Another quote: "The current document observes that, especially since the 2000 report, a number of positive steps have occurred, such as establishment of postdoctoral offices at many universities and increased use of individual development plans to help postdocs clarify their career options." I don't have (and have never had) post-docs in my group, but I noticed how, a few years back, the NSF started requiring the PI of grant proposals to submit a mentoring plan for any post-docs the grant would fund. In retrospect, it seems awful to think that this was introduced because some (many?) post-docs were receiving no mentoring whatsoever from their faculty member, in spite of the faculty in this case being traditionally called "mentor" rather than advisor or supervisor. (Oh, the irony.)
I was initially taken aback by the following quote (by the committee chairman) in Science magazine: "We cannot, in my opinion, train too many STEM PhDs." That's a bit easy to say when you think of the meager salary that said post-docs receive. But it turns out that the committee also recommends a substantial increase in post-doc salaries, and that the full quote in the interview was: "We cannot, in my opinion, train too many STEM PhDs. It’s an ideal preparation for a technologically sophisticated, rapidly changing world. But we can have too many people just defaulting into postdoc positions who don’t need to do so—and we do."
The report makes far-ranging recommendations to improve the post-doctoral experience. Here are a few:
- A limit of 5 years (cumulative) in duration for postdoctoral appointments,
- A use of the title "postdoctoral research" limited only to those people receiving advanced training in research, following by a transition to a permanent position (at the same institution or elsewhere) with a different title and appropriate salary.
- An increase in the post-doc starting salary provided by the NIH's National Research Service Award to $50,000, adjusted annually for inflation and regionally for cost of living and other factors.
- An environment that encourages post-docs to seek advice from multiple advisors.
- A bifurcation point between academic and non-academic careers that occurs before (not after) the post-doc experience.
Let's hope that, if we have to wait 14 more years for an update to this update, it will report substantial progress and improvement of the post-doc experience.
The National Academies Press recently published the report of a workshop on "The Arc of the Academic Research Career: Issues and Implications for U.S. Science and Engineering Leadership" (you can download the report as PDF for free from their website or read it online here). The report first highlights the current unstable arc in academic reserch careers before discussing the economics of early careers, the timing of tenure, incentives for retirement and "the other academe", especially given the rise of 2-year community colleges as a place for undergraduate education.
The report touches on important points, including the rise in post-doctoral appointments and the increase in their length in the sciences. I wish it had specifically mentioned the growing trend in engineering fields for would-be assistant professors to first take a post-doc before joining the ranks of the tenure-track faculty, which illustrates the growing aversion to risk of university administrators, understandably wary of committing a tenure-track position to someone who, when (s)he applies, is still a doctoral student working under the supervision of a faculty member.
While this stage has existed for a long time in the sciences, a whole new stage in one's academic research career is being introduced in engineering. (Thankfully, it is still possible for engineering PhDs to skip it, but perhaps not for much longer.) The report does do a very good job highlighting the different career opportunities for science vs engineering PhDs, and in particular the availability of well-paying industry positions for the latter.
In itself, adding two more years to someone's training in order for that person to better demonstrate his/her originality and independence of thinking may not be a bad thing, but when you look at the science model this originates from, you quickly realize those post-docs often stay in their position for five years or more, perhaps in part due to the excess supply of PhDs, which makes it difficult for them to obtain the position they aspire to. (The NAP also has a report out on "The Postdoctoral Experience Revisited"; read it here.)
There is a lot of talk from the government and in the media about needing more advanced degrees in STEM in order to foster innovation, but when you look at the career trajectory of biology PhD students, for instance (at least the ones I knew when they were M.I.T. graduate students at the same time as I was), they usually spend 1-2 years as a technician in a research lab before applying to PhD programs, where they stay for about 6-7 years, and then spend 5-6 years as post-docs before obtaining either a tenure-track position or a job in a pharmaceutical lab. This means that when they first start earning some decent amount of money (charitably defined as more than what a B.S. in industrial engineering at Lehigh makes on average right after graduation at age 22) they are between 34 and 37.
Add 6-7 years on the tenure track and these students, who presumably were at the very top of their class as undergraduates (at least those who made it to M.I.T.), are well into their 40ies when they know whether they will stay at their current place of employment or get to do it all over again. Another interesting tidbit I learned from a comment on an old blog post of mine is that faculty members in the sciences are, at some institutions, expected to pay their own salary from their own grant money (the university gives them lab space, a prestigious affiliation and, from what I understand, some small sum. I hope that is not the norm.) This doesn't seem a particularly attractive career path to any undergraduate who thinks ahead.
Yet, with too many talented PhD students who serve as TAs for required undergraduate introductory courses and the like, it is hard how to change the incentives in the system. A possibility would be to make it easier for students to change paths without having to drop out at year 6 (or 9) of the PhD process without anything to show for it, or a Master's that, in the sciences, is well-known to be given to doctoral students who decided to leave instead of being a degree students aspire to. In Electrical Engineering at M.I.T., for instance, "Electrical Engineer" is a degree in between the Master's and the PhD, requiring 162 credits + thesis instead of 66 credits + thesis. (In the chapter "The Other Academe", the report also highlights the situation of adjunct faculty and faculty at community colleges, which typically do not require PhDs and often employs ABDs [All But Dissertations] from the nation's top universities.)
Curiously, the report also makes no mention of Professors of Practice, who seem to be these days a growing trend in universties eager to expose their students to seasoned industry professionals. While a degree between M.S. and PhDs would represent a better off-ramp for doctoral students who decided they're no longer interested in research after all, PoPs provide a valuable on-ramp for industry practitioners interested in academic experience. This is not in itself a "research career" (to echo the report's title), and so may seem to be off-topic, but it is an important academic position at research universities, and thus deserves some attention.
Doctoral students who may lose heart with the lengthy tenure-track process would gain the ability to return to the academic world once they have spend some time in industry, hopefully at higher salaries. PoPs also offer the advantage of 3-year employment and (I believe) benefits, without requiring a PhD degree, again because those faculty members are expected to teach and sometimes direct professional Master's programs. Personally I'd love to fewer adjunct, part-time positions and more full-time, 3-year, non-tenure-track positions, not only for seasoned industry professionals but also for more junior employees. (Do take a look at Figure 4.1 p.28 about the "adjunctification" of the U.S. faculty.) In other words, as the arc of the academic research career becomes multiple arcs, there needs to be more on- and off-ramps between the worlds of academia and non-academia.
While the report also shares a few words about the future of tenure, very little is made of new models, e.g., at Olin College, where tenure simply doesn't exist. Obviously this is not a model that can easily be implemented elsewhere, but my personal opinion is that a model where promotion to associate professor would come with guaranteed 10-year employment and tenure would only come with promotion to full professor would make far greater sense in aligning faculty's incentives and universities' interests. In particular, the downside of the current model is that you end up with career associate professors or terminal associate professors, i.e., associate professors who never got promoted to full, and I don't see how that is a good outcome.
After my previous post on professional networks, here is another great HBR article, this time by Herminia Ibarra on "How to stay stuck in the wrong career" (December 2002). She argues that common-sense wisdom on how to engineer a career change is the precise reason why so many people stay stuck in jobs they no longer find fulfilling.
Here is what NOT to do, according to her, is "plan and implement": "First, determine with as much clarity and certainty as possible what you really want to do. Next, use that knowledge to identify jobs or fields in which your passions can be coupled with your skills and experience. Seek advice from the people who know you best and from professionals in tune with the market. Then simply implement the resulting action steps."
Ibarra provides compelling arguments as to why this is flawed, including the fact that we don't really have a "one true self" that is up to us to discover if we think hard enough, that intense introspection may get us stuck in the realm of daydreams, that the people around us whom we ask for advice may have a vested interest in us not changing or may not be able to imagine us in other roles, and more.
Instead, Ibarra explains that change happens when people try a new working identity, first by getting involved in side projects that allow them to try on this new identity and then by using the feedback to, hopefully, implement a complete change. The key is to do rather than to think through the problem. She says: "Doing comes first, knowing second." She describes success stories resulting from the "test and learn" model of change and provides guidance regarding how to craft experiments (side projects) to help professionals figure out what to do next and how to shift connections in one's professional network.
Read more here.
I recently came across a Harvard Business Review article that I'd read back in 2005 and find as relevant today as it was 9 years ago. It is called "How to build your network". The authors, Brian Uzzi and Shannon Dunlap, explain that "Strong personal networks don't just happen at the watercooler. They have to be carefully constructed." It is easy to believe that someone you have talked to for 5 minutes at a conference or professional event will make the perfect addition to your LinkedIn network, but in practice, people will rarely get out of their way to help others they don't really know unless those are strongly recommended by a third party they trust.
The authors make the observation that personal networks tend to be "highly clustered", i.e., "an individual's friends are likely to be friends with one another as well." They add: "Most corporate networks are made up of several clusters but with few links between them. Brokers are especially powerful because they connect the separate clusters." The article provides further details on these brokers and on ways to diagnose one's network, based on identifying your key contacts and remembering who first introduced you to them. (Hint: introducing yourself to your key contacts more than 2/3 of the time is viewed as a bad thing.) To forge stronger network ties, the authors recommend shared activities, whether membership to the same nonprofit board or outdoors activities such as running.
Here is a useful nugget of advice as parting words: "To build a network rich in social capital, cultivate powerful brokers who aren't in positions of formal authority - the places where everyone else looks."
Read more here.
It was announced this week that Megan Smith, who was named U.S. Chief Technology Officer in September and graduated from M.I.T. in 1986 and 1988 with a B.S. and M.S. in Mechanical Engineering, respectively, will be M.I.T.'s 2015 Commencement Speaker. Before becoming the nation's CTO, Smith served as Vice President of New Product Development at Google and was part of the leadership team at Google[x]. She has also served two terms as a member of the M.I.T. Corporation. The news release explains: "As U.S. CTO, Smith guides the Obama administration’s technology policy and innovation initiatives to advance our nation, with the goal of bringing the benefits of advanced information, data, networked communications technologies, and talented innovators to every sector of the economy."
In the words of Chancellor Eric Grimson: "As an MIT alumna, she understands the passion that drives our students; as the leader of Google[x], she understands the role that innovation and entrepreneurship can play in changing the world; and as the U.S. chief technology officer, she understands how technology can be used for social good. These are all themes that are of great importance to our graduates, and I am sure her remarks will be an inspiration to them."
You can read Smith's official White House bio here, including: "She has served on the boards of MIT, MIT Media Lab, MIT Technology Review, and Vital Voices; as a member of the USAID Advisory Committee on Voluntary Foreign Aid; and as an advisor to the Joan Ganz Cooney Center and the Malala Fund, which she co-founded. She holds a bachelor's and master's degrees in mechanical engineering from MIT, where she completed her master's thesis work at the MIT Media Lab."
A short profile of Smith is available on the MIT Alumni Association's website. Some Twitter accounts that may be of interest if you want to know more about the White House's technology policy are @WhiteHouseOSTP and @USCTO.
I very much look forward to her Commencement speech.
Here are a few articles from the Fall 2014 issue of SSIR that I found interesting.
Case Study: From Petitions to Decisions. (Article.) Describes Change.org's business model, including a switch to bring petitions to the center of its platform and a current push to position the site as a "venue for connecting those who want to change the world with those who run the world." Change.org focused starting in 2010 (with a Boulder, Co. petition to "suspend an ordinance [targeting homeless people] that made it illegal for people to sleep outside at night") on petitions that were actionable rather than rhetoric-oriented. It is now also creating an online venue on the site for the targets of petitions such as corporations or elected officials to interact with the people signing the petitions through Decision Maker pages. The site's vision is now to "establish Change.org as a place where multiple stakeholders can craft solutions through extended debate and negotiation."
A key element I didn't know was the reliance of Change.org on sponsored campaigns for its revenue generation. Sponsored petitions are promoted to "issue-aligned" users, i.e., users who have signed similar, non-sponsored petitions. "Because millions of users collectively sign thousands of petitions every month, Change.org can draw on a huge amount of data to predict who is likely to sign which petition." By default, users who sign a sponsored petition are kept updated of its progress and their email address is provided to the sponsor. The monetary value for Change.org and the sponsor is that Change.org isn't selling a list of names, it is selling "pre-qualified leads." From the sponsor's side, "sponsors don't pay for anyone who views their petition but doesn't sign it. They don't pay for anyone who signs it, consents to additional communication, but already appears on their own mailing list. Advertisers specify how many email addresses they want to acquire through the campaign and in what amount of time." Other sources of revenue for Change.org include letting users promote a petition to other users.
The article also details the tension arising from Change.org's "claim to be a platform with broad appeal and its affiliation with traditional progressive causes" and the shift to an open advertising policy.
Global Problem Solving without the Globaloney. (Article.) "Problems that seemed global in scope could have been more effectively solved at the regional, national or even local level." (This is what the author calls the devolution principle.) Most interesting, I found, was the distance-directedness principle. The focus is on gravity models, which - in international economics - "link interactions between countries to the product of their economic masses, divided by some composite measure of distance." Distance, however, "is not simply measured in miles" but includes multiple dimensions such as Cultural, Administrative (political), Geographic and Economic, leading to the acronym CAGE.
Useful words of wisdom include: "Where a business originates affects what countries it should expand to - and that answer usually isn't 'everywhere'. In 2004, of all US companies that had foreign operations, the largest fraction operated in just one foreign country, the median number in two, and 95 percent in fewer than two dozen." Another valuable principle is the distinctive-competence principle. The author explains: "The distinctive-competence principle extends the where, what and how questions, to ask whether a particular social enterprise is best positioned to pursue a particular global problem-solving opportunity."
The article, written by a globalization expert but targeting non-profits and the social sector, will certainly provide a mine of advice to the interested reader.
A New Vision for Funding Science. (Article.) Discusses the idea-to-impact gap and provides examples of pioneering approaches regarding the funding of lab-to-field investment. The authors advocate the emergence of philanthropic investing to "occup[y] the space between research grants and for-profit risk capital," blending philanthropic and financial perspectives.
Since my previous post on two recent papers of mine attracted quite a few viewers, maybe you'll find the following one insteresting as well: "Robust Binary Optimization using a Safe Tractable Approximation", this time with former PhD student Ruken Duzgun, now at Marriott International.
Abstract: We present a robust optimization approach to 0-1 linear programming with uncertain objective coefficients based on a safe tractable approximation of chance constraints, when only the first two moments and the support of the random parameters is known. We obtain nonlinear problems with only one additional (continuous) variable. The resulting robust optimization problems can be interpreted as nominal problems with modified coefficients. We compare our approach with Bertsimas and Sim (2003). In numerical experiments, we obtain solutions of similar quality in faster time.
Let us know what you think!