A rule of thumb for LEDs is: 1 deg. (°C) above T_j = life shortened to 1/2. Don't know it this has changed recently but it was being taught this way a few years ago.

Now I'm thinking: If they test LED lamps (either retrofits or dedicated designs with a heat sink) in the ambient temperature of 20°C (68 deg. F) and use a minimalistic (the cheapest or the least visible) heat sink that keeps the temperature just at the T_j, you will get quite shorter lifespan in the real life.
If a lamp was designed this way and the rated lifespan was 50,000 hrs., it's 12,500 hrs. in 22°C (71.6 deg. F) or 1,562 hrs. in 25°C (77 deg. F) or 49 hrs. in 30°C (86 deg. F)!!!

In reality, it probably won't be so bad (maybe they test them in the 30°C ambient temperature) but it still could be way lower that what's in a semi-opened fixture in summer.

This rule might be correct for the junction, but

1. I dont think it applies to the phosphor, which is probably the most significant way how LED lamps degrade over their lifetime. I would guess that the phosphor is following the more common 10°C / 1/2 life exponent coefficient

2. Any "radiating block of metal" or "convection interface" only provide a known tempoerature difference between the junction and the ambient air - The system have a fairly constant °C/Watt resistance of any heat transfer stage, that sums up to a fairly constant resistance from junction to air. With constant power dissipated you get the temperature difference

Now, Tj for semiconductors is usually 125..150°C. Know what, lets say 100°C for LED and i really dont think it can be any lower than that. Ta is on the order of ~25°C. Then the overall design have to provide <75 °C for the power of the LED

It simply makes no sense to me to make thermal design to the precision of few °C, when the allowed gap between Tj abnd Ta is 75°C. It is not a tight gap where every °C will determine the destiny of the design to be possible or impossible

Actually i dont think that even calculating the resistance of things like the heatsink to this precision level is possible (air humidity affecting convection, dirt settling on the lantern surfaces, lantern left under the sun during day - even in cold ambient air...)

The "Rule of thumb" is 10degC translating to 1/2 life is nothing else than a diffusion rate. It is based on the fact, than really nearly all of the deteriorating mechanisms are some sort of chemical reaction, which needs a supply of reactants to happen (and the diffusion is the most common mode of transport of the reactants), or the diffusion alone (metals diffusing to each other then causing semiconductor bond malfunction).
With LED's there are few main failure modes, quite "competing" over each other for being the limiting one:

- Bondwire failures. These yield to har failure of that individual LED, if that means a hard failure for the complete lamp depends on the lamp design. are of two types: Fatigue cracks when the bond is exposed to a mechanical strain (was case for the hard plastic lens LED's, now it is rather rare). Not directly related to the temperature, in fact one of the very few cases, where that rule does not apply (what matters is the thermal cycling).
Second type is the metal layer on the semiconductor diffusing into the bondwire ball (aluminum chip metalization into gold bond wire), so the metal is then missing on the bond pad. It could be practically eliminated by using intermediate layers (Ni-Pa-Au on top of the AL metalization separating the gold bond wires), but this treatment is rather expensive. Practically keeping the temperature of the bond below 150degC and using sufficient AL thickness uses to suffice to bring the life in the 10's khours with ppm's failure rate, so pretty sufficient for 20..50khour rated products, if the thermal management is not screwed up - it pretty well follows that rule.

- Cover material blackening (that includes the dicoloration of the fill material used to carry the phosphor sand): That is the typical browning of the plastic in front of the high intensity blue LED chip. Get's accelerated by the temperature and light intensity. Although new silicone or EVA based materials tend to be way more robust, they are still discoloring over the lamp life. With the hard plastic lens LED's this was the cause of the main degradation.

- Phosphor degradation: Mostly related to questionable processing quality and purity of the materials. Because what usually decays is not the phosphor itself, but the other materials present in the mix, whether they are really contaminants, or intentional components (glue,...).

- Diode itself degrading: That is usually caused by the defects initially present in the structure, growing by the LED operation. The severity depend on how strict and mainly thorough is the testing of the final chips, which is supposed to reject the dies with present semiconductor defects.
These defects are common in all semiconductor processes, they can not be avoided, the process control may affect just their density, so the scrap ratio. Problem is, when someone wants to boost their yield figures by tolerating some defects in the "to be delivered" bin (and indeed, with a process with higher defect density it would become more tempting for the managers to let the final test ignore some).
Most defects alone may not influence the main functionality directly (that's why it is so tempting for the managers to let them pass the final test), but they tend to worsen over time. And that worsening usually either follows the diffusion rate rule, or is based on thermal cycling.