Mike Hall from the Merseyside Skeptics says of the placebo effect:
“Confirmation bias means patients are more likely to notice and report changes that they are expecting and ignore changes they aren’t. The Hawthorne Effect means patients may alter their behaviour, simply because they are aware they are being observed. Then there is the Observer-Expectancy Effect, Recall Bias, the Clever Hans Effect, improved compliance, selection bias, and more and more. All these effects and biases can alter the recorded data, even if they don’t change the condition of the patient. This makes the patients in the placebo arm look as if they’ve improved more than they have.”
There are also some statistical paradoxes such as Simpson's paradox, the Will Roger's effect, and Berkson's paradox that could distort the findings to make the placebo effect appear stronger than it is.
In general, placebos only seem to improve subjective outcomes (i.e. self-reported symptoms such as pain) rather than objective factors (such as blood test results). In some cases (like migraines), improvement in self-reported symptoms might be all that's needed. But in cases where a disease is contagious or could degenerate without further treatment, it could be dangerous for someone to 'feel better' if the underlying problem persists.
There is some evidence to suggest the placebo effect works even when the person knows what they are taking is a placebo. Presumably this is because people have heard of the placebo effect and believe in it? I wonder what would happen if you told someone who had never heard of the placebo effect that you were giving them inactive medicine, would you still observe the placebo effect? Is the placebo effect weaker in people who are skeptical of it, if they know they are getting a placebo?
Rather than being a separate thing with mysterious healing powers, people like Mike argue that the placebo effect is actually just a fuzzy name for a collection of cognitive biases and statistical effects that lead to the appearance of improvements.