I don't think you can definitively exclude any of the items you've mentioned, however some things are much more likely than others.
Cold (or fractured) solder joints tend to get worse with time because of corrosion. This is frequently seen in some BGA packages that are driven hard so experience a lot of thermal expansion and contraction, for example in some high-end computer monitors.
Electrolytic capacitors age and increase in ESR with time.
Semiconductors have internal joints that can become intermittent and that often is related to temperature (LEDs that are driven hard in cheap consumer products in particular exhibit this regularly, but it can happen in ICs).
Semiconductors that are pushed to internal current densities that are excessive can fail over time, especially if exposed to sustained high ambient temperatures.
Even stressed resistors such as relatively high value types exposed to hundreds of volts can drift upward in value over time and cause starting issues for off-line SMPS circuits.
Optocouplers age and are temperature sensitive, so they would tend to fail at certain temperatures well before they are really most sincerely dead.
MOVs wear out after repeated transients and eventually tend to fail shorted.
Semiconductors are affected by radiation (other parts are too) and tend to degrade with exposure.
Environmental conditions can cause failures due to corrosion, water leakage causing tracking etc.
There's a common theme in what I've written above-
Things age. Some things age much faster than others.
Designs that 'push' the parts (deliberately as in the above-mentioned LEDs or because of bad design) may age much faster.
Stressed parts (electrically, thermally or mechanically) may tend to fail over time (or instantly, but that's not what we're talking about).
Environment can be a huge factor.
For useful failure analysis you really need to get to the bottom of what exactly has changed enough to cause the intermittent starting issue.