"Evidence on the Effectiveness of Comprehensive Error Correction in Second Language Writing" by Catherine G. Van Beuningen, Nivja H. De Jong, and Folkert Kuiken was just pre-published online in Language Learning. (The dissertation on which it is based is available here.) Though I have no special expertise in this area, I believe it is the most relevant and well-designed empirical study I have seen related to this question. While accepting the limitations of this paper, I think it provides clear evidence that direct comprehensive corrective feedback should probably be the default position. In other words teachers should correct, rather than simply mark, and they should do so for all errors (except spelling, which in this case benefitted more from indirect feedback).
[Update, April 2015. In a recent meta analysis, This study was found to be an outlier. "The Z scores showed that the effect size of this study was five standard deviations above the overall effect size. When the outlier was excluded, the overall effect size slid to g = .54 (CI = .35~.72) from g = 0.68 (CI = .35~.74)."
Kang, Eunyoung & Zhaohong Han. 2015. The efficacy of written corrective feedback in improving L2 written accuracy : A meta-analysis. The Modern Language Journal 99(1). 1–18. doi:10.1111/modl.12189.]
That is not to say that the default should never be overridden, but that teachers should think twice about doing so. The article doesn't address affective factors, for example, and certain students may react negatively to overmuch correction.
And we do have to take into account the limitations, which are well addressed in the paper. It was a relatively short treatment, based on only a single feedback cycle, and shorter treatments tend to produce stronger results than longer treatments (Li, 2010). On the other hand, it was in a second language environment, where Li found smaller effect sizes than in foreign language settings.
Overall, though, I think it's a compelling paper, but this is the position I've held for years, so I'm probably guilty of confirmation bias. Here's an example of the type of feedback I provide.
Student's original text:
The animators are the soul foundry. They are also the engineer of the human civilization. In animation industry, animation software is just vivid and lifelike tool to create a virtual world such as Maya, Zbrush, 3D Max and so on. However, through the understanding of things the animators build significant things adding their own creativity. Ultimately they reveal people are the crystallization of beauty. Learning 3D animation is a good approach which includes presenting vivid expression, purifying people’s mind, and training human creativity to everyone.
The animators are the soul foundry
. They are alsoand the engineers of the human civilization. In animation industry, aAnimation software is just vivid and lifelikeis the tool they use to create a vivid and lifelike virtual world and Uultimately theyreveal people areas the crystallization of beauty. such as Maya, Zbrush, 3D Max and so on. However, tThrough their understanding of thingsbiology, physics, and human nature, the animators build these significant things, adding their own creativity. Ultimately they reveal people are the crystallization of beauty.Learning 3D animation, then, is much more than learning to draw with computers. It is a good approach which includesmeans for presenting vivid expression, purifying people’s minds, and training human creativity toin everyone.
The animators are the soul foundry and the engineers of the human civilization. Animation software is the tool they use to create a vivid and lifelike virtual world and ultimately reveal people as the crystallization of beauty. Through their understanding of biology, physics, and human nature, the animators build these significant things, adding their own creativity. Learning 3D animation, then, is much more than learning to draw with computers. It is a means for presenting vivid expression, purifying people’s minds, and training human creativity in everyone.