I want to draw an analogy with fever. Fever is a "rational" (that is, both effective, and intended) response of the body to infection. The body raises the set point of its thermostat, because the invading bugs are less able to survive the elevated temperature than the organism that they are trying to invade.

But you can still die of a fever.

However, what I have read of depression (of which I have no personal experience) suggests that if their "ecological" hypothesis is true, clinical depression is analogous not to fever in general, but to the end stages of a failed fever defence -- not simply an adaptive response to stress, but the defeat of that response by the magnitude of the stress, and the collateral damage.

The idea is a variation on one that's somewhat popular here: that some conditions usually regarded as mental illnesses (Asperger's for example) are beneficial, even adaptive.

I have a hypothesis that autism-spectrum conditions are less selected against and therefore proliferate as a result of increased background levels of literacy (increasing the amount of information you can acquire without direct interpersonal teaching). I haven't maintained enough interest in this hypothesis to see if there are any natural experiments to look at.

Labeling the phenomena depression frames the debate in a certain way, that's not useful for understanding evolutionary reasons.

If you hit someone strongly on the head the resulting trauma can frequently lead to depression. Is that depression in reaction to getting hid on the head evolutionary advantageous? That depends highly on how you have to structure a brain to not suffer depression when it get's hit.

While Googling around it seems that dogs also have something like depression: http://pets.webmd.com/dogs/features/depression-in-dogs Does depression helps dogs to solve complex problems which involve analysis? Is that what Andrews et al are arguing?

More evolutionary perspectives here:

http://en.wikipedia.org/wiki/Evolutionary_approaches_to_depression

This would seem to be supported by the fact that sad moods make us process information more analytically.

In addition to providing information that can serve as a basis of judgment, feelings influence how people process information, that is, their processing style. A number of different explanations have been offered for this observation, usually highlighting the role of one specific type of feeling (for reviews see Schwarz & Clore, 2007, and the contributions in Martin & Clore, 2001). Feelings-as-information theory provides a unified conceptualization of these influences in the context of a situated cognition framework (Smith & Semin, 2004). It assumes that human cognition stands in the service of action (James, 1890) and that our cognitive processes are responsive to the environment in which we pursue our goals. This responsiveness ranges from the higher accessibility of knowledge relevant to the current situation (e.g., Yeh & Barsalou, 2006) to the choice of processing strategies that meet situational requirements (e.g., Wegner & Vallacher, 1986). When things go smoothly and we face no hurdles in the pursuit of our goals, we are likely to rely on our pre-existing knowledge structures and routines, which served us well in the past. Moreover, we may be willing to take some risk in exploring novel solutions. Once things go wrong, we abandon reliance on our usual routines and focus on the specifics at hand to determine what went wrong and what can be done about it. [...]

The theory further predicts that feelings or environmental cues that signal a “problematic” situation foster an analytic, bottom-up processing style with considerable attention to details, whereas feelings or environmental cues that signal a “benign” situation allow for a less effortful, top-down processing style and the exploration of novel (and potentially risky) solutions (Schwarz, 1990, 2002). This does not imply that people in a happy mood, for example, are unable or unwilling to engage in analytic processing (in contrast to what an earlier version of the theory suggested; Schwarz & Bless, 1991). Instead, it merely implies that happy feelings (and other “benign” signals) do not convey a need to do so; when task demands or current goals require bottom-up processing, happy individuals are able and willing to engage in it. A study that addressed the influence of moods on people’s reliance on scripts (Schank & Abelson, 1977) illustrates this point.

Employing a dual-task paradigm, Bless, Clore, et al. (1996) had participants listen to a tape-recorded restaurant story that contained script consistent and script inconsistent information. While listening to the story, participants also worked on a concentration test that required detailoriented processing; in contrast, the restaurant story could be understood by engaging either in script-driven top-down processing or in data-driven bottom-up processing. Happy participants relied on the script, as indicated by the classic pattern of schema guided memory: they were likely to recognize previously heard script-inconsistent information, but also showed high rates of intrusion errors in form of erroneous recognition of script-consistent information. Neither of these effects was obtained for sad participants, indicating that they were less likely to draw on the script to begin with. Given that top-down processing is less taxing than bottom-up processing, we may further expect that happy participants’ reliance on the script allows them to do better on a secondary task. Confirming this prediction, happy participants outperformed sad participants on the concentration test. In combination, these findings indicate that moods influence the spontaneously adopted processing style under conditions where different processing styles are compatible with the individual's goals and task demands, as was the case for comprehending the restaurant story. Under these conditions, sad individuals are likely to spontaneously adopt a systematic, bottom-up strategy, whereas happy individuals rely on a less effortful top-down strategy. But when task demands (like a concentration test) or explicit instructions (e.g., Bless et al.,1990) require detail-oriented processing, happy individuals are able and willing to engage in the effort.

Numerous findings pertaining to a broad range of feelings (from moods and emotions to bodily experiences and processing fluency) and cognitive tasks (from creative and analytic problem solving to persuasion and stereotyping) are consistent with the predictions of feelings-asinformation theory (for reviews see Schwarz, 2002; Schwarz & Clore, 2007).

There are many facets to depression and to analyse them from a evolutionary perspective, you should evaluate every single one them separately.

Some specific types of depression, like 'burn-out' syndrome, have an obvious cause-effect and is some sort of defense mechanism to protect you from yourself. The metaphor of getting burned is fitting, because it's similar to physical pain.

But evolution doesn't bother if something is beneficial or not. A common mutation can be devastating for the individuals mental health, but as a species we can have evolved to circumvent this weakness by triggering certain behaviors to prevent us from reproducing. The courting behavior in animals is also often a test the behavior of a potential partner; unexpected behavior leads to rejection.

Genes do not evolve into cleaner, more efficient code, but rather into super complex buggy spaghetti-code that somehow works. We may have evolved to be somewhat intelligent, but we are also inherently flawed.

An interesting observation in that regard was the deranged penguin (documentary: Encounters at the End of the World).

https://www.youtube.com/watch?v=x7kdDeGXUjI