Note to journalists -- you don't have to believe everything they tell you (business model edition)

File this one with the flack-to-hack series, that part of the decline in journalism that can be attributed to increased willingness to print whatever the PR person hands you.

You've probably heard about this:

While the world has no shortage of pie-in-the-sky renderings for floating cities, a new proposal by Silicon Valley start-up Blueseed is perhaps the first to invoke entrepreneurial spirit and the American Dream in its soapbox pitch. Looking for a way to give foreign-born tech entrepreneurs access to the enterprising atmosphere of Silicon Valley, the founders of Blueseed are hoping to revamp an old cruise ship to create a community, of sorts, 12 nautical miles off California's coast in international waters, so international techies could live within commuting distance without a work visa. With a simple business tourism visa, denizens of the vessel could take a 30-minute boat ride to the mainland once or twice a week. Of course, creating an inhabitable community on water—one that people would live on six months or a year at a time—requires some interior design finagling, including incorporating lots of light and open space.

The big issues here have to do with labor, immigration and industrial policy, but this also provides a small but useful example of gee-whiz, ddulite journalism. I will give the writer, Amy Schellenbaum, credit for using the phrase 'pie-in-the-sky' in her opening sentence but other than the occasional qualifier, that's pretty much the last trace of incredulity you'll find.

For example:

Blueseed still needs $18M (of the $27M required, total) before any building can take place, but if this actually gets rolling, Mart imagines he'll charge organizations anywhere between $1,200 a month to house their employees in a shared cabin to $3,000 a month per person. Just what everyone wants to do: shack up with their coworkers!

I looked at a few cruise ships on Wikipedia and where cost was given, they were all in the hundreds of millions (and sometimes much higher). Even done on the cheap with an old ship, between buying the damned thing and doing the kind of major refit described here, we've got to be talking nine figures.

Now, I have to admit, my experience with start-ups is very limited, so it's very likely that I'm missing some important aspect of funding in this casse. Perhaps this proposal is more credible than it looks, but as presented here, these numbers seem awfully screwy, and it's worrisome that the reporter either doesn't feel the need to explain the screwiness or, worse yet, doesn't even notice it.

Quote of the day

From Norman Geisler’s book on Thomas Aquinas via John D. Cook:

No, I do not agree with everything he [Aquinas] ever wrote. On the other hand, neither do I agree with everything I ever wrote.

More on statistics in research

One last good point out of the whole excel error in the 2010 Reinhart and Rogoff paper:

This raises another issue. Programming is getting easier and easier, but it’s hard to do well. Economics these days depends heavily on programming. It seems to problematic to me that we rely on economists to also be programmers; surely there are people who are good economists but mediocre programmers (especially since the best programmers don’t become economists). If you crawl through a random sample of econometric papers and try to reproduce their results, I’m sure you will find bucketloads of errors, whether the analysis was done in R, Stata, SAS, or Excel. But people only find them when the stakes are high, as with the Reinhart and Rogoff paper, which has been cited all around the globe (not necessarily with their approval) as an argument for austerity.

It is the most generally applicable advice, in that it might well have prevented many of the other errors.  But it is not a one directional attribution of responsibility.  A careful analyst should have done sensitivity analysis for things like outliers and found the critical sensitivity of the results to the inclusion of New Zealand.  Or compared the different weighting schemes.  If you are going to try and understand a small dataset you need to be very thorough at looking at all of the ways that the data itself could be summarized. 

So statisticians also have a burden of care, here, even if one was not directly involved in this analysis to counsel researchers on how to approach these difficult data problems.  In the same sense, it would not be a bad structural change for researchers to consult more with methodologists, who can look at the problem outside of the lens of strong priors and might be more quick to question a surprisingly good result. 

Weekend blogging -- Gained in translation

Specifically, where does translation stop and adaptation start?

This is never a trivial question. When you read Tartuffe in college, the story was by Molière but there's a good chance the words were by someone like Richard Wilbur. If you reacted to something other than the plot -- a turn of phrase, a subtle shading of character -- it is not immediately clear who should get the credit.

Of course, a Wilbur starts from a position of respect of the original work and tries to capture what it's like to experience a work in its original form. What happen's when a translator simply says "to hell with it"?




From Wikipedia:

Die Zwei, the German version of The Persuaders, became a cult hit in Germany. This was largely because the dubbing was substantively altered creating a completely different program.^[11] In France Amicalement vôtre (Yours, Friendly) also became a popular show because it was based on the redubbed German version instead of the English original. 

The German dubbing was "a unique mixture of street slang and ironic tongue-in-cheek remarks" and that it "even mentioned Lord Sinclair becoming 007 on one or two occasions".^[18] Dialogue frequently broke the fourth wall with lines like "Junge, lass doch die Sprüche, die setzen ja die nächste Folge ab!" (Quit the big talk, lad, or they'll cancel the series) or "Du musst jetzt etwas schneller werden, sonst bist Du nicht synchron" (Talk faster, you aren't in sync any more).
Research from the University of Hamburg notes the only common elements between Die Zwei and The Persuaders! is they use the same imagery. Other than the "the linguistic changes entailed by the process of translation result in radically different characterizations of the protagonists of the series. The language use in the translations is characterized by a greater degree of sexual explicitness and verbal violence as well as an unveiled pro-American attitude, which is not found in the source texts".^[19] 

In 2006 a news story by CBS News on the German dubbing industry mentioned The Persuaders! The report discovered that many German dubbing artists believed that "staying exactly true to the original is not always the highest aim". Rainer Brandt, co-ordinator of the German dubbing of The Persuaders and Tony Curtis' dubbing voice, said "This spirit was invoked by the person who oversaw the adaption and also performed Tony Curtis' role: When a company says they want something to be commercially successful, to make people laugh, I give it a woof. I make them laugh like they would in a Bavarian beer garden." ^[20] 

Other researchers suggest international versions of The Persuaders! were given different translations simply because the original English series would not have made sense to local audiences. For instance the nuanced differences between the accents and manners of Tony Curtis, the American self-made millionaire Danny Wild from the Brooklyn slums, and Roger Moore, the most polished British Lord Sinclair, would be hard to convey to foreign viewers. Argentinian academic Sergio Viaggio commented "how could it have been preserved in Spanish? By turning Curtis into a low class Caracan and Moore into an aristocratic Madrileño? Here not even the approach that works with My Fair Lady would be of any avail; different sociolects of the same vernacular will not do—much less in subtitling, where all differences in accent are irreparably lost".^[21]

War and research

Though it's too early to tell whether the geopolitical lessons are going to take, we did learn some things from the war in Iraq.

From the New Republic:

Neither Boston Medical Center nor Boston Emergency Medical Services have responded to queries about how tourniquets were used after the marathon bombings, so we can't yet confirm their effectiveness. It wouldn't be the first time, though, that killing abroad has saved lives at home: Wartime medical advances have long translated into civilian life (trauma centers full of specialists sprouted up in cities, for examples, after they’d worked to great effect in Vietnam). America's latest conflicts have also improved techniques to repair tissues and nerves that have prevented amputations in operating rooms around the country.

That kind of research funding tends to dry up when the soldiers come home. But hundreds of thousands of people die from traumatic injuries every year, and Jenkins says that huge gaps still remain in our knowledge of how to treat them. The National Trauma Institute has been campaigning for more money to research trauma, which doesn't loom as large in the public consciousness as many diseases do.

This goes beyond tourniquets. As NPR reported a few years ago, techniques from the war have revolutionized emergency medicine:

The medevac choppers land and then taxi over to the gate just outside the emergency room, where gurneys are waiting. Nightfall has brought a bone-chilling wind, and a gang of nurses and orderlies rushes four patients into the warmth of the ER.

It's more than warm inside. In fact it's 100 degrees. It's the first clue that this hospital — the Joint Theater Hospital at Afghanistan's Bagram Air Field — is a little different. Through years of war, combat surgeons have learned that hypothermia is a big risk in patients with significant blood loss. Nine years of conflict in Iraq and Afghanistan have brought some grim benefits: a new wealth of knowledge about treating war wounds.

"At the beginning of this conflict, we were taking the best trauma medicine from the civilian sector, and we brought it to Iraq and Afghanistan," says U.S. Air Force Col. Chris Benjamin, the hospital commander. He says now his doctors tell him it's the other way around.

"Here we are seven, eight years later, taking what we've learned in these conflicts to teach them the advances that we've made with this data collection here in theater," he says.

It seems strange to discuss war in terms of research and data collection, but despite what you constantly hear, most big problems are solved by throwing large sums of money at them and one of the most effective ways of convincing a government to start throwing money is to get it into a war. I'm not saying that the means justify the ends or that we couldn't find a way to get better results without the horrific costs.

What I am saying is that, in the world we've got, for researchers, like Keynesians, the steps we think society should take (like spending significant amounts of money on trauma research) often only came as a byproduct of war.

Trying to wean myself off "hibernate"

I assume I'm not the only blogger with this bad habit: I'll come across something that's either worth sharing or worthy of comment or a good jumping-off point for a discussion, but I won't have the time (or perhaps the will) to write out the post right then so I leave the tab up and hibernate the computer rather than shutting it down.

Here are the tabs I currently have open. Hopefully, some will eventually grow into posts of there own. If not, maybe you'll find something of interest in the list.

Dana Goldstein has an article at the Smithsonian on the influence of American industry on education. 

ThisTV is showing The Sweet Smell of Success and I can't help thinking that it's time for a sequel.

I've been thinking about the problem of pivoting something as complex and amorphous as a political party, or maybe thinking about the problem of how to think about the problem, either way this and this (both from TPM) highlight the difficulty.

It is generally both foolish and dangerous for a reformer to say nothing could be worse than what we have now (worse is almost always a possibility), but when the speaker is a college president in the state with arguably the best university system in the country with probably the best university system in the world, it indicates just how detached from reality the debate has become. (via More or less bunk.)

On a related note, I've been meaning to discuss the Hawthorne effect in educational research but now I have no idea what to call it.

I've never been a big Fitzgerald fan (more of a Count No 'Count man, myself), but if I were, this would worry the hell out of me (in 3D):

Maybe I can sucker Joseph into writing a few posts on the wrong kind of government assistance for business.

This Felix Salmon piece on Ripple nicely compliments all of the Bitcoin coverage.

On a related note, though it's not really an insult to say a rich person has more money than sense, some rich people apparently have less sense than you have money.

Finally, it seems that a stock based on a gourmet snack fad may not hold its value.

Two views of Excel

Part of the Reinhart-Rogoff fall-out (see here for Joseph's take) has been a discussion of the role of Excel and similar programs in analytic work. Andrew Gelman has a post up on the subject that includes this quote from an unnamed statistics professor:

It’s somewhat surprising to see Very Serious Researchers (apologies to Paul Krugman) using Excel. Some years ago, I was consulting on a trademark infringement case and was trying (unsuccessfully) to replicate another expert’s regression analysis. It wasn’t until I had the brainstorm to use Excel that I was able to reproduce his results – it may be better now, but at the time, Excel could propagate round-off error and catastrophically cancel like no other software!

Followed by this assessment by Gelman himself:

Microsoft has lots of top researchers so it’s hard for me to understand how Excel can remain so crappy. I mean, sure, I understand in some general way that they have a large user base, it’s hard to maintain backward compatibility, there’s feature creep, and, besides all that, lots of people have different preferences in data analysis than I do. But still, it’s such a joke. Word has problems too, but I can see how these problems arise from its desirable features. The disaster that is Excel seems like more of a mystery.

In contrast, here's James Kwak discussing Excel (prompted by news that an embarrassingly simple spreadsheet error contributed to the London Whale fiasco and, yes, the similarities have been noted):

Microsoft Excel is one of the greatest, most powerful, most important software applications of all time.** [** But, like many other Microsoft products, it was not particularly innovative: it was a rip-off of Lotus 1-2-3, which was a major improvement on VisiCalc.] Many in the industry will no doubt object. But it provides enormous capacity to do quantitative analysis, letting you do anything from statistical analyses of databases with hundreds of thousands of records to complex estimation tools with user-friendly front ends. And unlike traditional statistical programs, it provides an intuitive interface that lets you see what happens to the data as you manipulate them.

As a consequence, Excel is everywhere you look in the business world—especially in areas where people are adding up numbers a lot, like marketing, business development, sales, and, yes, finance. For all the talk about end-to-end financial suites like SAP, Oracle, and Peoplesoft, at the end of the day people do financial analysis by extracting data from those back-end systems and shoving it around in Excel spreadsheets. I have seen internal accountants calculate revenue from deals in Excel. I have a probably untestable hypothesis that, were you to come up with some measure of units of software output, Excel would be the most-used program in the business world.

But while Excel the program is reasonably robust, the spreadsheets that people create with Excel are incredibly fragile. There is no way to trace where your data come from, there’s no audit trail (so you can overtype numbers and not know it), and there’s no easy way to test spreadsheets, for starters. The biggest problem is that anyone can create Excel spreadsheets—badly. Because it’s so easy to use, the creation of even important spreadsheets is not restricted to people who understand programming and do it in a methodical, well-documented way.*** [*** PowerPoint has an oft-noted, parallel problem: It’s so easy to use that people with no sense of narrative, visual design, or proportion are out there creating presentations and inflicting them on all of us. ]

To the extent there's a conflict here, I'm with Kwak on this one. For all their problems, I'm still a big fan of Excel and similar programs (such as the OpenOffice version I have on my laptop). They are indispensable in business for just the reasons Kwak lists.

Furthermore, I'd like to see them have a similar role in secondary and even primary education. I've long found spreadsheets to be a great tool for introducing basic concepts about math, computing and data visualization. You can see some sample lessons over at You Do the Math (one on functions and iteration, another on geometry and Monte Carlo techniques).

Swords and Wizardry Appreciation Day: Wow

I was amazed by the number and quality of Google+ responses to the Swords and Wizardry Appreciation Day.  It is pretty amazing the actual depth that this segment of the OSR actually has. 

One thing that really stood out is how easy it is to houserule S&W.  It is not just that they have the rules available in editable text format but that the rules themselves are very robust to tinkering.  This was one feature of more modern RPGs that I found difficult -- if you changed the assumptions of the game designers then the system showed problems.

The classic example is magic items in a D&D 3.X/Pathfinder world.  It is well known that the system presumes a certain level of magical gear for the players.  It also assumes that it is easy to make magical items.  Without these features, characters may well not operate as expected (the Monk 1/Druid X is much more formidable, for example). 

But with S&W, the alteration of a single rule rarely has repercussions across the system.  Sure, you can break it if you try to.  But the game plays well with many tweaks -- some of which are actually discussed in the rulebook (ideas like giving +1 to hit to two weapon fighters, or no memorized  wizard spells above level 6 but rituals instead).   This allows for a lot of freedom for groups to tinker and try to find the optimal balance. 

Swords and Wizardry Appreciation day

I was always surprised to be a Swords and Wizardry fan.  In a lot of ways I was a bigger fan of the original basic and expert (B/X) D&D books released after the first run of Dungeons and Dragons (and contemporaneous with Advanced Dungeons and Dragons). 

But the adventures sucked me in: both Grimmsgate and MCMLXII caught the sense of wonder of the early adventures and the lack of systematic approaches to everything that came latter.  Modern descendants of the original D&D (both the Wizards of the Coast and Paizo versions, both high quality products) lacked the sense of wonder that can from not fully understanding the world.  When everyone knows what an Ogre is and can do then there are few surprises left. 

The latest addition, in terms of rules, is Swords and Wizardry complete, which manages to have all of the key elements of the rules in 131 pages or so.  Sure some of the elements are not described in great detail (especially the spells) but that is a virtue.  I do think it was a mistake to include the spells above levels 5 or 6 . . . but the designer even admits to this in a sidebar.

Overall it was a refreshing vision of the game in an earlier era and one that I quite liked.  It was well timed to come out with the re-release of TSR D&D (both as premium reprints and as PDF files).  I have quibbles (no elf Druid despite an illustration to the contrary) but I think the vision and the production quality make up for these issues.  It is true that the system will work best for those who already understand D&D, but that really was how the hobby came to be (with the rule books often being somewhat inscrutable). 

Overall, I have really enjoyed this game system and I hope that Frog God Games continues to do high quality products like these ones (especially the modules). 

Tonight only -- free teleseminar on getting rich by writing for television

OK, not really, but a friend sent me this and I thought I'd pass it on because:

1. It sounded interesting;

2. It ties in (albeit weakly) with some recent discussions of distance learning;

3. Ken Levine (MASH, Cheers, Frasier) is one of the best bloggers on the subject of television and the business of writing.

Sorry about the late notice.

Tonight at 9 PM Eastern/ 6 PM Pacific I’ll be holding another FREE teleseminar. The topic is writing partners and my partner, David Isaacs will be joining me. We’ll answer all your questions. If you’re interested in participating and not already on the alert list here’s where you go to sign up. You must sign up by noon PDT. Did I mention it's FREE?

From unsupported hypothesis to conventional wisdom

I don't want to pick on Andrew Delblanco, who turned in an unusually well-balanced piece on MOOCs in a recent issue of the New Republic, but this passage bothered me quite a bit:

But the most persuasive account of the relentless rise in cost was made nearly 50 years ago by the economist William Baumol and his student William Bowen, who later became president of Princeton. A few months ago, Bowen delivered two lectures in which he revisited his theory of the “cost disease.” “In labor-intensive industries,” he explained, “such as the performing arts and education, there is less opportunity than in other sectors to increase productivity by, for example, substituting capital for labor.” Technological advances have allowed the auto industry, for instance, to produce more cars while using fewer workers. Professors, meanwhile, still do things more or less as they have for centuries: talking to, questioning, and evaluating students (ideally in relatively small groups). As the Ohio University economist Richard Vedder likes to joke, “With the possible exception of prostitution . . . teaching is the only profession that has had no productivity advance in the 2,400 years since Socrates.”

(for more along similar lines)

I'm not an economist, I haven't gone through Baumol and Bowen's research, and my college teaching experience is more than a decade out of date, so I'm certainly missing some salient points here, but if the hypothesis is true, I'd expect to see the following:

1. There should be a shortage of teachers

2. The percentage increase in wages for teaching should be greater than the percentage increase in overall tuition.

3. The share of tuition going to instructional costs should increase substantially relative to costs such as administration.

Rather than seeing all of these, it's not immediately obvious that we're seeing any. College teaching jobs are not easy to get, non-instructional costs seem to be more than holding their own and if you look at people who are paid solely to teach, their wages are increasing more slowly than tuition.

We have to be careful about treating informal observations as data, but with that caveat in mind, there are lots of reasons to question and not much evidence to support the hypothesis that a lack of technology-driven productivity gains by instructors are causing the sharp growth in tuition. Nonetheless, the idea has gone from interesting but unlikely theory to generally accepted fact.

Here's where I blame it all on Steven Levitt.

OK, not really, but this is an example of a journalistic fad that owes a lot to Freakonomics: propping up an argument with a semi-relevant economic allusion. This isn't the same as analyzing a problem using economic concepts (for example, trying to explain health care inflation as a principal agent problem). Instead, what we have here is the idea that explaining the concept is the same as arguing that it applies.

It is, of course, possible that cost disease really is driving the growth of tuition. This is an enormously complicated problem and complicated problems often look very different when examined in detail. Viewed from a distance, though, the idea that a lack of instructional productivity gains are driving the growth simply doesn't jibe with what we're seeing.

Lesson for Epidemiology

I am amazed that Mark was not all over this statistical modeling issue.  Matthew Yglesias posts the response from the authors here. 

A couple of quick thoughts.  First, no matter what you think of the authors of the original paper, I am inclined to applaud them for sharing both the source data and the statistical code used to analyze it.  It opened to them a comprehensive methodological critique but that could not have happened if they had not shared the source data and code.  The next result, from a scientific point of view, is a huge win (and I think this outweighs any actual errors, presuming that they were not intentional). 

Second, the whole issue of association versus causality is one that we should think more about.  It can be easy to slip in the fallacy of equivocation: asserting associational thinking when a study is attacked but interpreting it in a very causal manner in a policy discussion.  We have this problem all of the time.  We state associations like "people who eat red meat have a higher rate of disease" but then turn that into a recommendation of "eat less red meat". 

Now clearly one has to make an actual decision at some point and, in a lot of cases, the guidance can cause no harm.  But there are complex cases where this could matter a lot.  Consider the inverse association between smoking and Parkinson's disease.  Imagine (based on no data -- just a though experiment) that we extended this association to show slower progression in patients who continued to smoke.  That might make one think that cigarette smoking might be recommended to Parkinson's disease patients.  But if we are wrong (and the association is confounded or acting as a marker for some other trait) we will increase the rates of lung cancer and serious heart disease for no therapeutic benefit. 

Sometimes a randomized trial is an option.  But when it is not (as many exposures, like height or air quality, are hard to randomize) then there is a very delicate decision-making process.  It is very instructive and important to watch how other fields of study deal with these types of issues. 

Edit: It seems everyone has been looking at this question.  Some additional very good takes (over and above the Mark Thoma, Matt Yglesias and Mike Konczal linked above) include:

Andrew Gelman
Noah Smith
Paul Krugman
Tyler Cowen
Jonathan Chait

I continue to hope Mark Palko can be convinced to weigh in as well. 

Simply amazing tale of corporate governence

This was just amazing:

At Iris International, a medical diagnostics company based in Chatsworth, Calif., shareholders rejected all nine directors in May 2011. In keeping with the company’s policy, they submitted their resignations. And then they voted not to accept them. The nine stayed on the board. (The company was acquired in late 2012 by the Danaher Corporation.)

Assuming that this is a factually correct version of events and does not omit a key detail of some kind (always a risk when a story seems this outrageous), this is a remarkable separation of the owners of the company from their own board of directors.  In theory, could any board-member be replaced under these circumstances?  

A defense of the envelope method

Emily Oster has a provocative piece in Slate arguing against the envelope method of budgeting.  The crux of the argument appears to be:

The principle in question here is that “money is fungible.” In reality, all dollars are the same. There is no such thing as a gas dollar, a grocery dollar, a “fun” dollar. The Ramseys of the world argue you should just apportion them into the envelopes depending on what you think you might spend. But doing it this way is not actually the best way to dole out your money; you are not optimizing. Your money is just not working as hard as it should.

The system works great, as long as nothing ever changes. But the minute that some price changes, you’re in trouble. Here’s an extreme example. Imagine you drive to work, and your “gas money” envelope contains enough money to get you to work for the month, and maybe a little cushion. But then gas gets more expensive. You may first react by switching to a worse grade—say, from premium to regular (research shows many people do this). But if gas prices go up even more you simply will run out of money in the gas envelope. And then you won’t be able to get to work. Strictly following the envelope system here would be much, much worse than “cheating”: It’s certainly bad for your household budget if you miss work.

This presumes that the envelope system is static and the allocation of money in the envelopes never changes as prices change.  In other words that you do your budget once and never alter it.  It is clear that the strong form of this argument is actually a straw man -- would anybody let themselves be evicted because their rent went up but they could not bear to alter the values on the envelopes?  In general, this system will fail int he face of massive price and income volatility, but then nothing works well when you cannot plan for future income and expenses.

In my experience, the actual reason for the envelope system is one of two cases:

1. It reduces the complexity of managing a detailed household budget
2. You are really so poor that you have no ability to reallocate money

In my youth, I knew a lot of poor people (less so now, which is not necessarily a change for the better) and, in fact, was very poor for a time.  When your budget is tight enough it may really be the case that you cannot overspend on gas for the car without being unable to pay rent and/or eat and/or pay the electric bill.  People at this level of poverty do not necessarily have credit cards..

One of the features of being middle class is having flexibility.  Consider the plane ticket example that Dr. Oster presents: that makes total sense if you could (in theory) allow a credit card to increase.  But if the cost of the item would bring your bank account below zero and you don't have credit (or it has ruinous terms, like a payday loan or a 30% credit card) then you simply can't spend the money at all.

In practice this lack of flexibility can be very tough.  There are a lot of items that, bought in bulk, are cheap.  But bulk items are hard to transport on the bus (especially if it is crowded) and cost more (and you often cannot borrow against the future).  There are a surprising number of costs to being very poor (as opposed to being very frugal).

So, yes, it is possible for the envelope system to be too inflexible and that is not ideal.  But it is very, very useful when you have $110 after rent and utilities to make the rest of the month work.

Risk Aversion

What is the average interest rate these days?  According to one source it is puny:

Throughout the entire industry, however, the national average bank interest rate for savings accounts is only .21 percent

So why might be it rational to overpay your taxes by a small amount?  Consider this comment from Megan McArdle:

A tax refund is not free money you get from the government. It’s free money you gave to the government—by overpaying your taxes all year. You think the interest rate on your savings account is paltry? At least it's not zero, which is the amount Uncle Sam pays you for your yearlong interest free loan. (If you underpay the government, on the other hand, they charge you not only interest, but also penalties.)

On one hand her advice is completely sensible.  It makes a lot of sense to reduce a very large tax refund.  On the other hand, you need to save a lot of money for the 0.21% interest in that savings account to outweigh the interest and penalties.  Since even interest is 3% (15 times the savings rate), I think it makes sense to aim for a small refund rather than to make errors on both sides of the distribution (since underpaying is much more expensive than the lost revenue from overpaying).  This logic may not hold if interest rates and/or inflation pick up in the future.