If I randomly select a programmer from all the programmers who write Clojure, working or not, interviewing or not, they are likely to be a “better” programmer than if I randomly select a programmer from all the programmers who write JavaScript, working or not, interviewing or not.
But of course, that’s not what happens when I interview. I meet one person in the morning with three years experience working on projects of types A, B, and C. In the afternoon I meet another with four years of experience working on projects of types B, C, and D.
If one has Clojure and the other JavaScript, I’m much more likely to be delighted to discuss Clojure than JavaScript, but as far as HIRE/NO HIRE is concened, the thing that matters to me are projects Of type A, B, C, or D, and the candidate’s role/contribution.
If the JavaScript programmer’s experience is solid, I do not want a recruiter in-house or otherwise trying to save me an interview and filtering them out.
The bottom line is, there are like 100x as many JS folks as CJ folks. So let’s say there are 100 great CJ people out of 1,000 CJ people. 10% ham!
But that means there are 100,000 JS people. If even 1% of them are great, that means that there are 1,000 great JS people out there, and I do not want to pass on meeting any of them just because there are also 99,000 not-so-great JS people out there.
In my imaginary distribution, a random CJ person is 10x more likely to be good. But there are 10x as many good JS people as CJ people, and my experience is that THEIR experience is a far better signal than their programming language.
—-
p.s. I am competing with people who only want to hire programmers with shiny signals like ClojureScript or the ability to quote “JavaScript Allongé.” Good! Let them ignore the 1,000 good JS people, easier for me to interview them and make a successful offer.
Now that I think about it, please ignore what I just said. Only interview people who build systems out of Lisp, Prolog, and Maleboge.
“Nova 1220 BASIC, Acius 4th Dimension, Hypercard, MetroWerks CodeWarrior, Lightspeed C, Turbo Pascal, MP/M, DigiTalk Smalltalk/V, J2EE, ...”
What does any of that have to do with each other.
However, the implicit companion assumption that a resume with lots of clojure experience will have those skills is not something I've seen any evidence for. Many Clojure-focused candidates are unproductively dogmatic and are at risk of wasting everyone's time by trying to rearchitect our codebase because it doesn't meet their particular taste. In general, if a candidate is an adherent to a niche programming language they need to be able to explain to me when its use would be appropriate and inappropriate. If their answer is "this language and/or dogma is always better" then you're adorable but you need some more experience before I'm going to hire you.
It's also important to point out that, while my core requirements are the same for both jQuery and Clojurey candidates (ability to function in a large, complex codebase & ability to manage your own code's complexity), the specific skillsets are very different. Most clojure enthusiasts are actively bad front-end engineers. Which is fine! But many positions require engineers to at least be able to dabble in UI work, and if I get the sense that you would find it beneath you to do so, its probably going to mean better luck somewhere else unless I have a very specific position to fill.
Would you care to share those statistics? While we're at it let's also be sure to define "good engineer."
For what it's worth, I'm challenging you as someone who is neither a JavaScript developer nor a Clojure developer. I'm of the opinion that biases like the one the author laments are perpetuated (in part) by this leaky arguments from "statistics" that never appear.
You're then meant to combine their stated experience with their stated credentials and your assessment of their trustworthiness to arrive at some conclusion.
This imprecision may annoy you but it is a common way of speaking English so it's probably not in your interest to fight over the choice of wording.
This is because saying something is true "statistically" is a claim made in both scientific and casual contexts. It's mentally taxing to figure out which context is in play - someone could have come into this thread with actual statistics and said the same phrase, verbatim. Therefore (as I see it), it's realistically plausible that someone could read the original comment and come away believing recruiters have done actual statistical analysis a la TripleByte to arrive at their conclusion.
For example I have not run a study but I am very confident that, statistically speaking, tall people are better at basketball than short people. I am saying that I have an opinion that if you were to try and choose good basketball players, biasing yourself towards the tall people will increase your odds of making a good choice. I use the phrase "statistically speaking" to make it clear I am not saying "All tall people are better at basketball than short people".
In a lot of contexts you will drop the phrase completely because you are talking to other smart people and they will know that the context of your statement is in the realm of populations and odds, but online there are a lot of pedants with poor communication skills so you have be extra clear.
Idiomatically, any way you use the phrase implies great confidence in the corresponding claim. But I have not seen any substantial explanation of why this effect makes sense, which makes me extremely suspicious that the effect exists at all. I don’t particularly care about whether they’ve conducted a peer reviewed study. I care about whether or not their heuristic is backed by anything empirical. “I’ve noticed this...” doesn’t cut it, no matter how fast and loose you want to play with the literal meaning of the word in context.
"I'm so going to win the lottery"
"On average ... no you aren't" / "Statistically speaking ... no you aren't" / "If you work out the odds ... no you aren't" / etc.
If I ask you a question about some gambling game, you are going to think about it in a statistical way. I mean you will think in terms of populations and samples and odds, not that you will go perform experiments and then do a statistical analysis of that evidence in order to come to a scientific conclusion.
I, like you, am riled up by people using puff words to inflate the apparent credibility of claims, often claims that have little or no basis for actual credibility.
But I think that "statistically speaking", as it is used by nearly all English speakers, means "speaking about these groups in aggregate", not "speaking as a statistician". It doesn't even mean something about an empirical basis, it means talking about groups. I agree pretty much entirely with freshhawk's analysis.
Now, aside from _word choice_, your skepticism about whether the original poster has any empirical basis at all for their claim... well, fair enough.
I'm deeply skeptical. People like to throw around percentages and references to statistics to justify their biases, but I don't think the commenter has done nearly enough to motivate the conclusion that the modal Clojure developer is more likely to be a "good engineer" than the modal JavaScript developer.
If someone who interviews thousands of people and finds a correlation between those who have some clojure experience and quality, I believe him.
For the simple reason that most programmers don't try very hard and anyone taking the time to even learn a rarely used language is demonstrating a positive quality that most don't.
I stand by my point: using words like "statistically" and "correlation" perpetuate biases which may not have a rational basis. In this entire thread no one has explained why Clojure programmers are more likely to be "good" than JavaScript programmers (nor has "good" been defined!). I could just as plausibly state that someone tried to learn Clojure because they perceive it to be a buzzword, just as JavaScript programmers are often accused of playing buzzword bingo with web development frameworks.
Empirically speaking, we haven't ended up anywhere. No matter how sure you are of this phenomenon, unless you try to make its observation more robust, it will continue to be a microcosm of the tech hiring industry. Heuristics often belie subtle biases that do not actually have a foundation in truth.
Once known as the Python Paradox http://www.paulgraham.com/pypar.html
Nowadays I suppose it would be the Clojure Conundrum or the Haskell Happenstance
Their websites sound like they are the pinnacle of the industry, but on phone/email they sound like a bunch of HS dropouts :(
Heck, I’d even say if you only look at the population of engineers who you have directly observed doing a good job on some interview trivia or code test, you still cannot reliably conclude that draws from that population are statistically significantly more likely to be acceptably good at a job once hired than draws from a different population.
People and processes capable of detecting skilled engineers are extremely rare. Skilled engineers are also somewhat rare.
What we set up as tech hiring processes are mostly just crap-shoot just-so stories we tell ourselves, mixed with a ton of hubris.
No, they're paying them to find engineers that will pass the interview process. It's a subtle, but important distinction since prejudices that exist among engineers will naturally develop in recruiters. It's the recruiting equivalent of teaching to the test.
And like it or not, the kind of signaling the article describes is real. When an engineer says they can write clojure, it indicates a curiosity that ability to program JavaScript does not. Almost no one is forced to learn Clojure because of some job, project or even job prospects. It means you likely cared enough to devote your own time to learning for the sake of learning. And a lot of engineers look at this kind of thing as important and screen for it in interviews.
A recruiter operating any differently would be like a machine learning algorithm that ignores its training data.
The curious thing is, I think the number who regard it as important is likely far higher than the number who actually do it. Which is understandable; once you reach a level of seniority you also have many other demands on your time and may not be able to be a hobbyist anymore on top of day job, family commitments and so on. So maybe once you learnt every new language but now you don’t have the opportunity, or you have decided to “go deep” in just a few, established languages. So the power of the signal is highly variable depending both on where the interviewer and interviewee are in their lives.
Also, knowing JS in the late-90s did indicate a curious and ambitious developer. It only went mainstream later...
P(broken|android) =~ P(android|broken)
We see from Bayes rule that this is false, since P(android) >> P(broken)But when my iPhone breaks, I take it into Apple for service and add it to my collection of “Apple gives great service” data points.
Saying that prejudice is just applied statistics completely mischaracterizes prejudice.
The "Bell Curve" for females of a species generally has a significantly narrower standard deviation for practically any genetically driven characteristic.
The problem is that none of this applies to an individual.
Isn't this limited to traits which are influenced by XY chromosomes? (Usually due to males having only one X and thus being more likely to express recessive genes?)
It's an argument that doesn't play well on message boards, because software developers love to believe that they're part of a cognitive elite "inventing the future" and "eating the world", when in fact really the overwhelming majority of us are performing a bog-standard white collar symbol manipulation job no more challenging than the job done by an accountant or actuary.
Nerds like to pretend that we're all working on the graph isomorphism problem alongside Babai, but really most of our days involve basic wiring of data from one place to another; form fields to database rows to packet fields to page markup. We want to believe we're brain surgeons, but most of the time we're barely even dentists.
But I will also say that various software shops I know of large and small are always desperate to find good talent. I’ve hired quite a bit myself and finding capable talent is very difficult. So our work is not as totally deprecatable as you make it out to be.
If this is a true effect I imagine it's a selection effect. Women who felt luke warm about coding never entered, or left.
Given a woman in tech, you probably have someone who is at the tails, someone really good.
Recruiters use an intuitive approach to conditional probability. Given only the info alone (if someone did JQuery or Clojure) exposes you to two bell curves, with the X-axis being "programming ability". The first bell curve that has all JQuery people is more on "the left" and the one having all Clojurians is more on the right. The Clojure bell curve has shitty engineers and some really good people are part of the JQuery one.
Prejudice is sort of rational. If a recruiter looks at one datapoint, it's unlikely that it is at the tails of the bell curve, so the Clojurian applicant will get more attention. If you are treated unfairly because of the group you "belong to", I am sorry - but this is reality.