What skills do self-taught programmers commonly lack?

Hmm. I did a quick poll of the programmers I know personally & would consider successful (for a reasonable value of success). One has a firm grasp of assembly (which makes sense because he uses it a lot), a handful have a firm grasp of C (most claim to have a "little" C but from experience with them they couldn't build anything w/o a book for guidance :P).

So I don't agree with the hypothesis.

Look, as far as I'm concerned, unless you had nothing but other peoples' source code and API documentation, you are not 'self-taught.'

I used to cobble together VBA macros for Excel at work, based on code snippets and the built-in help files. At first, although I got things done, I really had not a blinking clue what I was doing. Then after a few weeks I bought Excel VBA for Dummies - which taught me how to do things properly and understand the essence of all the constructs I was using.

I read more books, and I read web tutorials and watched video lectures. They taught me a lot. Then I left work to get a post-graduate-conversion certification in IT. I always said I was doing it for the cert, not the teaching. The course was good but only because it instilled discipline. Really attending a lecture is no different than watching one on YouTube or iTunesU. The teachers don't have time to give you personal tuition. Their notes are not more instructive than the classic CS/software books in publication. The learning materials available elsewhere are decent rivals to professional education courses. The only thing I can see being a huge benefit is doing pair-programming with an expert for a few years, but how many courses offer that? Not to mention, the learning that's supposed to go on at universities is hardly automatic.

So therefore the question is meaningless, unless you mean actually self-taught people, who would probably (if they managed to develop into competent professionals) have some strange quirks. Like self-taught musicians who never learnt common techniques I guess.

Sadly, the most important one is nowhere to be found: good software development practices. I've met a lot of brilliant self- taught programmers, who could read up on parsers and systems programming, but not one read up on how to keep software maintainable or how to manage project risk.

Of course, they default to waterfall. (!!!)

And I wonder if often better. There is value to having the motivation to learn something before being formally trained in it. I was not too impressed with some of my peers in CS, especially those who could not figure out how to open a file in C in their senior year.

I was self-taught, then formally trained, and since self-taught. I would have to say that both have value and that you are missing something without the formal education. "Confidence" is a good reason, but exposure is important, too. I am sometimes surprised when the college math pops up in my work, spanning anywhere from programming drivers PCI cards to web interfaces. Work with a large enough dataset, and maybe you will need to make use of a different sort from the language-included "quicksort" algorithms.

I thought I could my way out of a paper sack before getting the CS degree. I ended up learning a few things, some trivial, some useful and connecting. You could probably pick it all up up in a couple decades, but four years is a nice jumpstart.

Simultaneously, the learning of approaches and techniques does not end there. For example, I sometimes wish my university had made mention of MOP instead of presenting OOP only the C++, Java, and Ada way.

I think there is a difference between people who can accomplish tasks and people who can accomplish those tasks in a way that they can be easily extended and maintained in the future; in one respect, we can call that experience, and in another, we can call that an exposure to enough approaches/patterns that it is easier to select the best for the job. Any jumpstart is a huge help and can aid in removing the bad habits learned from solo experimentation.

I won't discount the that there is a value in being taught by a professional educator. However, the primary sources are there for everyone. The excellent secondary sources are there for everyone. The amount of supplementary material available via the internet is very advantageous to the autodidact. The available interactions via the internet are such that they approach a collegiate peer group (if triaged). All of this taken together and approached in a steady, determined fashion, I believe, can near to perfectly simulate a college CS education (or education in almost anything else for that matter).

Academia and college educations are getting increasingly difficult to idealize as information becomes redistributed and consumed in unsanctioned ways. And sanctioned ways, too. See all flavors of OCW.