I don’t agree that you can’t quantify this, but I do agree that this article is fundamentally bogus. The thing that bugs about these kinds of overblown claims is that it poisons the well for deeper analysis since it causes people to write off the idea of using data. IMO, the problem here isn’t the use of data, it’s the unreasonably strong conclusion drawn from very weak data. Ironically, you can even see this in baseball if you look at pre-“sabermetric” analysis of baseball. Many of the stats people used to judge baseball players simply weren’t very good (in that they had low predictive power in terms of both future player performance and team win probability), which gave the anti-stats camp ammunition they could use to dismiss the idea of deeper analysis.

It appears that the author has taken all factors that aren’t sheer numbers and found “wins about replacement” (WAR) for all of those factors combined. Some things that make WAR work for baseball that seem difficult to apply to warfare include:

1. In baseball, there’s a well defined concept of a “win” and regular season wins are equivalent. This is certainly not true in warfare and a single “win” can sometimes reverse the effect of many losses.
2. In baseball, runs scored are independent enough that you can predict wins from runs scored “well enough” and the effect of players is independent enough that you can predict run generation and prevention “well enough”. You can apply this type of analysis for sports where players actions aren’t even close to being independent, but it’s much harder to do so. It’s not clear what “hidden” or “intermediate” variables in warfare can stand in for the equivalent variables in baseball. A naive application using intermediate variables like casualties seems likely to yield almost no value (I could be convinced otherwise, but I’d need to see an argument, not the blind application of a formula). If this analysis were serious and not clickbait, I’d expect the discussion of these variables and the justification that these variables are predictive to take up most of the space.
3. There’s a large sample size. We have tons of data about baseball (and other sports). While we know about “lots” of battles, we don’t really have good data for many of them. Even for battles that have happened in the past decade, casualty estimates can vary by an order of magnitude depending on who’s doing the estimate. Even if everything else were perfect, you’d have a hard time doing sabermetric analysis of baseball if your error bars on the number of runs scored varied by an order of magnitude. Also note that, when comparing across eras in any sport, you have to adjust for how the environment was different in different eras; different adjustments can give very different rankings of players. If any kind of era adjustment was done in this article, it’s not mentioned and the adjustment isn’t justified. For many eras, I suspect we simply don’t have the data to do a data-driven adjustment, so any adjustment would either involve original research or intuition applied to case studies.
4. Replacement level is fixed. There’s a debatable but relatively well defined meaning for replacement level in baseball, but no such thing exists in warfare. As we’ve seen in baseball analysis, moving the replacement level target around can drastically effect findings of who’s “best” and since we have a variable that gives the person doing the analysis total freedom, we need some justification for why replacement level is set to where it is if we want to be convinced the the result isn’t just a post hoc justification. Note that this goes into all of the variables used in (2) and their weights, so in this case the person writing the article had many degrees of freedom that they could tweak to tune the results arbitrarily and there’s no justification for any of “knob settings” in the entire article.

I don’t think there’s anything inherently wrong with running the numbers and seeing what comes out. But the author then claims that this somehow “proves” that some generals are mathematically superior to others. As you say, this is not correct.

Using a methodology that can’t answer the question being asked and then feeding in bogus data is quite common, so I don’t mean to pick on this article in particular for that reason. The thing I find most unfortunate about this is that I’ve seen this article passed around a lot, and AFAICT it’s been passed around precisely because the author makes a lot of claims that they can’t back up – an article titled “a preliminary investigation of ‘wins above replacement’ applied to warfare” probably wouldn’t go viral, wouldn’t have gotten linked to on Marginal Revolution, etc. This article appears to have been widely read not in spite of being clickbait, but because it’s clickbait. It uses the cachet of sabermetrics, as if sabermetrics is about having some formula and a term called “wins above replacement” rather than being about figuring out what we can “know” from data and what the limit of our knowledge are. As an article about who the “best” general is, it’s pure cargo culting.

Note that the BLS definition of “Computer programmer” appears to be very low level:

“…They turn the program designs created by software developers and engineers into instructions that a computer can follow.”

It’s likely that many of us here who would call ourselves programmers might fit into another statistical bucket for the BLS, like “software developers” - median $97k, or even “Computer and Information Research Scientists”, median $108k. Honestly, even those seem low and so I assume they’re wrapping together some jobs that I wouldn’t consider equivalent.

In general it seems really hard to tell how much of these stories of high compensation to believe without just asking your peers, something I’m always reluctant to do. Certainly having some idea of what sort of field these offers are being made for is useful - Dan’s article was helpful in pointing out that people in “hot fields” get gobs more money.

It’s also not always totally clear what being “senior” means. I think I had “senior” in my title once, but I think it means different things at different places. :)

How much of that $250K is actual salary and then how much of it is “value of benefits”?

Zero. A mediocre compensation package for a senior engineer today is $150k salary, $100k/yr of equity that’s not quite as good as cash (but pretty close) and bonuses.

Sure, tools in general, but probably not for formal methods, which is what’s being discussed in that section. Sorry if that was unclear :).

What kind of evidence do you want? Google is pretty careful about drilling “communicate with care” into employees from day one. There are lots of ways they do this, but the takeaway is that most legally sensitive issues end up being discussed one on one, in person if possible, and over video chat or phone if not possible.

Having worked at Google in the past, I can think of a number of pretty bad stories, and in every case where I heard about what HR did, HR was very careful to act in a way that mitigated the risk of legal liability as much as possible. It’s extraordinarily unlikely that there’s an email lying around that’s a smoking gun that says something like “Welp, director Y clearly harassed person X, now what?”.

On the other side of things, I know multiple people who have avoided talking publicly about their stories because of the severe damage to their career they were afraid would be caused by the backlash. There’s a lot of downside to speaking out about stuff like this and not much upside.

Assume for a minute (or an hour, or a day), that it happened just the way she tells it. So, what do you do after the internal reporting process failed and you kept it quiet? Also, harassment cases are rarely the one where a colleague flips out the camera and starts filming.

“Dang, my smoke alarm is going off. That’s a pretty serious allegation, and what if it’s a false alarm? Better stay in bed until I see some REAL evidence of a fire.”

While I wouldn’t dismiss these stories out of hand (after all, it is pretty hard to prove of hand remarks that have passed before you realize what’s happened), I agree with your stance that we should approach them with critical caution.

While sexual harassment is a real problem, and no doubt exists in the technology industry (and pretty much everywhere else), there have been many cases (outside of tech) where women falsely accused employers, coworkers or acquaintances of sexual misbehaviour of various natures.

History also teaches us that if you give a certain issue public attention (such as these kinds of acts have been enjoying in recent months), many stories, both true and false, will come out of the woodwork.

I’m not saying this one in specific is a false story though.

My advice here is pretty standard, not just for writing, but for any field, but I think it’s standard because it works. Get good feedback and practice incorporating it. Don’t try to fix everything at once; add skills one at a time.

In my experience, it takes a while to find people who will give you good feedback for free. Most people don’t come up with substantive writing critiques, but if you ask for advice on a lot of drafts, you’ll eventually find a few people who come up with meaningful critiques and not just spelling and grammar corrections.

And of course you don’t have to only get free advice. If you’re a programmer, you can certainly afford to hire a professional editor to help you. I don’t want to post what’s basically an advertisement for an editor on lobsters, but if you’re interested in this, contact me privately (see my profile) and I can recommend a good editor.

My process is to get good feedback about something (often a blog post), incorporate the feedback in the current thing, and then try to write something from scratch that incorporates one specific thing from that feedback. I then get feedback on that second thing to see if I was able to improve that one thing. When I’m able to consistently write something that doesn’t get the same critical feedback, I move on to another problem in my writing and try to eliminate it.

I don’t claim that my writing is good, but when I look at how my writing has changed over the past year and a half, it’s certainly improved a lot. For a given level of effort (30 second email, 15 minute blog post, 3 hour blog post done in 3 sittings, etc.), my writing is a lot better. I’d say that my dashed off 15 minute blog posts are as “good” as my serious 2-4 revision blog posts were a year and a half ago, when I started this blogging/writing improvement experiment. Considering the time investment (a few hours a month), I’d say that improving my writing using this method is one of the highest ROI things I’ve done lately.

Does anybody have any advice on how to become a effectively become a better writer?

Yes.

Yes, you want to write as much as possible, that’s a given,

That is a given of course. Read too. Read as much as you can on as much as you can. It really is a shame that our modern lives are so strongly aligned towards any other form of recreation than reading, given reading’s relatively lengthy time requirements to get much out of it. Beware the shysters who claim they can “speed read” at some exorbitant number of words per minute. The act of reading requires more than merely pushing and popping groupings of words into and out of your brain’s mental queue as fast as possible.

but can you get meaningful feedback from someone who’s much better than you at it?

Absolutely. Editors exists for more reasons than merely finding typos and other grammatical errors.

Is there quality coaching of any sort that you can get?

There are several communities of critiques, readers/writers, reviewers, editors, teachers and so on. In my experience, their eagerness to help is typically proportional to your own reciprocity in helping them in turn. That being said, finding an audience can still be difficult given our desire to become ever more efficient schedulers of our limited (and in some cases decreasing) free time.

I agree with the posted essay. Writing can bring great clarity and you can grep it.

A few observations from good essays / blogs

1. Put your personality into it.
2. Give a strong purpose to each essay / blog.

I consider zenhabbits.net to be a good representative of the above said points.
Words like good and better are however, subjective.

Hope that helps :)

Thanks for the correction! It’s really appreciated.

I credited you as “tedu” in the acknowledgments. Let me know if you’d prefer to not be mentioned, or be mentioned with a different name/alias.

Thank you so much! Deleting all of my existing web fonts fixed the problem. I wonder why Chrome can’t/doesn’t checksum fonts, and why the problem didn’t affect Firefox.

I find cache eviction policies really interesting. I find it interesting how they’re somewhat like generational GC, and whether some of the algorithms from caches can be applied to GC (thinking SRRIP and DRRIP). I guess the only difference is you can evict something you’ll use again, this is not the case with GC.

Your eviction policy post was interesting! I’m not much of an expert here, but the first things I thought of were evaluations of disk buffer caches and network processor ASICs.

I don’t have any actual direct recommendations, but all the things I found in my quick googling just now were posted in SIGMETRICS, so that’s where I’d start.

There’s also multiprocessor cache coherence directories, although replacement policies there get complicated. ( I liked this paper about SCD (Scalable Coherence Directory) at HPCA ‘12 from Daniel Sanchez and Christos Kozyrakis - has some interesting info about how design constraints of coherence directories impacts replacement, noting that their preferred cache design ends up selecting replacement candidates in a de-facto random fashion. I haven’t thought about coherence directories in a few years and that paper was pretty easy to follow - they did a nice job on the background (section 2).

Yes, it turns out there are a lot of studies on this. After hearing some particularly strong claims about this sort of thing recently, I did a literature review and wrote up notes on each study, just for my own benefit.

The notes are way too long to post as a comment here (the notes are over 6000 words, i.e., longer than the average Steve Yegge blog post), and they’re actually pretty boring. Let me write a short summary. If anyone’s interested in the full results, I can clean them up and turn them into a blog post.

1. Most older controlled studies have a problem in their experimental methodology. Also, they often cover things most people probably don’t care about much anymore, e.g., is ANSI C safer than K&R C?

2. Some newer controlled studies are ok, but because of their setup (they compare exactly one thing to exactly one other thing) and the general lack of similar but not identical studies, it’s hard to tell how generalizable the results are. Also, the effect size these studies find is almost always very small, much smaller than the difference in productivity between programmers (who are usually selected from a population that you’d expect to have substantially smaller variance in productivity than the population of all programmers).

3. People rarely study what PL enthusiasts would consider “modern” languages, like Scala or Haskell, let alone languages like Idris. I don’t like that, but it’s understandable, since it’s probably easier to get funding to study languages that more people use.

4. There are a number of studies that mine data from open source repositories. They come up with some correlations, but none of them do anything to try to determine causation (e.g., instrumental variable analysis). Many of the correlations are quite odd, and suggest there are interesting confounding variables to discover, but people haven’t really dug into those.

For example, one study found the following in terms of safety/reliability of languages: (Perl, Ruby) > (Erlang, Java) > (Python, PHP, C), where > indicates more reliable and membership in tuple indicates approximately as reliable. Those weren’t the only languages studied; those particular results just jumped out at me. The authors of the study then proceeded to aggregate data across languages to determine the reliability of statically typed v. dynamically typed languages, weakly typed v. strongly typed, and so on and so forth.

That analysis would be interesting if the underlying data was sound, but it doesn’t pass the sniff test for me. Why are languages that are relatively similar (Perl/Ruby/Python) on opposite ends of the scale, whereas some languages that are very different end up with similar results? The answer to that is probably a sociological question that the authors of the study didn’t seem interested in.

This seems to echo the results of the controlled studies, in that the effect a language has on reliability must be relatively small for it to get swamped by whatever confounding variables exist.

5. There are lots of case studies, but they’re naturally uncontrolled and tend to be qualitative. Considering how little the other studies say, the case studies make for some of the most interesting reading, but they’re certainly not going to settle the debate.

Anecdotally, SML’s type system is not particularly friendly to programmers (it’ll beat you over the head over and over again until you satisfy its exacting requirements) but extremely supportive of correctness; Java’s type system isn’t especially friendly to anyone due to its inflexibility, inexpressiveness, and extraordinary amount of required boilerplate; and Python’s type system (or lack thereof) is incredibly friendly to programmers (“sure, you can do that! Why not! Treat this int as a dict, I don’t care!”) but allows all sorts of bugs ranging from the blatantly obvious to the deviously subtle to make it into running code.

I’ve seen a couple of studies of programmer productivity in various languages/environments (not, IIRC, looking specifically at type systems), but none well-founded or conclusive enough for me to remember or recommend them.

Here’s some that I’ve found:

• An Empirical Comparison of Static and Dynamic Type Systems on API Usage in the Presence of an IDE: Java vs. Groovy with Eclipse
• A Quantitative Analysis of Whether Unit Testing Obviates Static Type Checking For Error Detection
• Chapter 14 of Making Software
• An Experiment About Static and Dynamic Type Systems
• Static vs. dynamic type systems: an empirical study about the relationship between type casts and development time

I have the same problem. I wonder if we could improve the situation with comment-discovery mechanisms that are less obsessively focused on recency. As it stands today, if you’re commenting on an article more than a day old, it’s unlikely your comment will be read by enough people to get a reply. (This is far better than HN, where the time horizons are even shorter, and if you take too long thinking about your comment you get punished with a /deadlink when you go to post it, but it’s still bad.)

It might be useful, for example, if I could see a feed of the latest comments by people I think are interesting enough to follow — like a Planet aggregator, but for comments; or if it were easier to link to or even transclude relevant comments from the past. (For example, at the moment, the Your Threads page only lists the last few comments you’ve made; if you want to link to a comment even you yourself made a few months back, good luck finding your own comment. Maybe you can Google it. And finding other people’s related comments is even more difficult. Meanwhile lobste.rs has comment score data that it could totally use in a comment search engine, and Google can’t.)

Usenet did pretty well for some years with only a feed of the latest “comments”. You might even say that Usenet’s second-biggest problem was how well it did at fostering ongoing conversations. (Its biggest problem was vulnerability to spam.)

I would have liked to see the @dl comments before, and in fact I’m replying to one from two months ago (which is actually related to this topic) now, but without better discovery mechanisms I probably won’t see them. And I won’t see if anybody replies to my replies, either.

Another problem is that the user pages display the average score per article/comment; this score goes down if you post a comment that nobody reads, because it will always be scored 1. If you display scores on a web site, people will try to game them, and the way to win this game is to post very few comments — ideally short jokes or related links on recently-posted articles, and never questions betraying ignorance, corrections of errors, lengthy analysis, especially on articles that weren’t just posted, and especially when it’s not a top-level comment or a reply to a top-level comment; deep in a thread your chances are nil. (This comment’s score doesn’t fit my theory, though.) Compare how badly @dl (6.47) is beating @kragen (1.70), largely by virtue of posting an order of magnitude fewer comments. (Also @dl is probably smarter, more cogent, and better-looking than I am, but I think the score difference is primarily a result of the comment count.)

Maybe we should consider how to adopt some techniques from the popular sites that are at the opposite extreme: StackExchange and Wikipedia. SE automatically does some kind of full-text indexing thing to find “related questions” and answers, and interesting questions and answers continue gathering upvote karma (and sometimes further responses) for years, not hours. Wikipedia articles continue to garner interaction forever, even if that interaction is in the form of somebody correcting your spelling or punctuation (or, worse, deleting your contribution); and although there’s a “Recent Changes” page, the only people who read it are the vandalism patrol. Also, both SE and Wikipedia have social structures and software that reward participation, rather than punish it.

(Edited extensively.)

There’s a high volume of new stories, which means that things have a very limited amount of time to hit the front page.

That is exactly what /recent was designed to fix (which I just enabled again right before you posted your comment), since that was one of my complaints about HN when creating Lobsters.

The underlying problem here is that we’re constructing conversation environments where fast-twitch reactions drown out slow, considered ones. There have been attempts to construct conversation environments where that’s not true, like the Rotisserie system (description in a book) and Wikipedia, but they have been only partly successful; this is perhaps partly because the matching law of operant conditioning means that rapid rewards, even small ones, are much more effective than slow rewards, partly because web sites that you interact with only occasionally don’t have a way to get your attention, and partly because the rapid back-and-forth of rapid reactions simply allows groups that interact frequently to be creative in ways that rarely-interacting groups aren’t. (Consider how many current popular memes come from 4chan.)

I saw Fig 8, and I think it’s informative, but note that it’s normalizing for known factors, which includes factors that are partially caused by discrimination. If the original article had claimed that 2/3 of the gender pay gap was understood, I would have no problem with that.

But the title claims that the gap doesn’t exist (that’s the part that’s literally the opposite), and the text is a combination of claiming that it doesn’t exist and that it isn’t a problem because it’s understood. However, as the authors of the AAUW study go through and account for various factors, they explicitly note that some of them are partially due to discrimination. To make a claim that there is no gap is false. To make the claim that the gap is not a problem because we’ve understood that part of the gap is because of discrimination and don’t understand what 1/3 of the gap is from is not technically wrong, but it strikes me as a strange claim to advance. My interpretation is that the original article makes much stronger claims than that.

Sorry for side-tracking here:

What is Engineering Technology, should it be include with Engineering? Why are CS and IT lumped together? Why are Bio/PhysSci/Science Technology(same question as E.T.)/Math/AgSci all lumped together?

I don’t know if these categorizations were yours or the original studys.

I’m sorry to hear that, Dan. I’ve upvoted you. Definitely appreciate the actual statistics vs. linkbait crap other people post online.

I am in complete agreement.

I agree with those that believe showing the names of those who’ve voted will add some culpability and responsibility. There was a time when I believed that Lobste.rs could be mature about that sort of thing. Recently, my faith has been adjusted and I think it would end up in meta-hell (or worse.)

Thanks for taking this seriously. One of the worst things about the programming reddit is that people use downvote to indicate dislike, and that things quickly fall off the front page if downvoted.

Link bait gets enough upvotes to survive the torrent of downvotes that hits pretty much every new article, but technical articles rarely do, which heavily biases the top articles towards rants and away from posts with real technical content.