Statistically, how much attack damage does the AC boost from a shield prevent on average? I've seen claims it results in a 10% reduction in damage taken, is that accurate?

10% is the naive answer

In actual games it is around 15-25%.

The +2 bonus to AC is 10% of the d20 roll, but it is more complicated than that:

• If the enemy can only hit you on a natural 20, adding a shield does not do anything


• If without a shield the enemy hits you 50% of the time, it becomes 40% with one. The difference is about 20%¹.


• If the enemy only misses on 1, dropping the shield does nothing

Calculation

Assume for simplicity that the attacker does 10 hp damage per hit, and criticals increase damage by 5 HP (50%).

The DPR against a shieldless opponent is $$frac{(21 - rollNeeded) * 10 + 5}{20} {}$$

The DPR against a shielded opponent is $$frac{(21 - rollNeeded - 2) * 10 + 5}{20} {}$$
rollNeeded = (your AC) - (opponent's to hit). It is always between 2 and 20.

Table

Substract your typical enemy's to hit from your AC, and find how much a shield would help.

For example you want to decide between dual wielding and sword-and-board for a new Fighter. As starting equipment you can get Chain mail (AC 16), and you expect many Goblins (+4 to hit). You get 12 -> 21.05% less damage received with a shield!

Graph

1) Criticals complicate things

About 14%-30% in normal play

I've produced odds for the scenario where the Attacker is a Great-Weapon-Fighter using a Greataxe with a +3 Damage modifier, or 1d12+3[GWF] for a normal hit, or 2d12+3[GWF] for a critical hit. Different weapons and damage modifiers and features will skew these results, but these should represent a relatively "normal" scenario for most modes of play.

Across most of the game, a shield will reduce this Attacker's DPR by a flat 1.034, only meaningfully varying at the highest ACs, when Critical Hits represent the only significant chance to hit.

My experience with the game has taught me that at the usual range of Armor Class values that an average party will encounter (both for themselves and for enemy creatures), the "to-hit" range usually lands between 7 and 15. In this range, we can see that at the low-end, a shield reduces the Attacker's expected DPR (per attack) by about 13.592%, and at the higher end, it reduces DPR by about 29.824%.

begin{array}{|l|r|r|r|}
hline
text{Name} & text{Normal} & text{+Shield} & text{[%]}\ hline
text{To-hit = 1} & 10.183 & 9.667 & text{-5.067%}\ hline
text{To-hit = 2} & 10.183 & 9.150 & text{-10.144%}\ hline
text{To-hit = 3} & 9.667 & 8.633 & text{-10.696%}\ hline
text{To-hit = 4} & 9.150 & 8.117 & text{-11.290%}\ hline
text{To-hit = 5} & 8.633 & 7.600 & text{-11.966%}\ hline
text{To-hit = 6} & 8.117 & 7.083 & text{-12.739%}\ hline
text{To-hit = 7} & 7.600 & 6.567 & text{-13.592%}\ hline
text{To-hit = 8} & 7.083 & 6.050 & text{-14.584%}\ hline
text{To-hit = 9} & 6.567 & 5.533 & text{-15.745%}\ hline
text{To-hit = 10} & 6.050 & 5.017 & text{-17.074%}\ hline
text{To-hit = 11} & 5.533 & 4.500 & text{-18.670%}\ hline
text{To-hit = 12} & 5.017 & 3.983 & text{-20.610%}\ hline
text{To-hit = 13} & 4.500 & 3.467 & text{-22.956%}\ hline
text{To-hit = 14} & 3.983 & 2.950 & text{-25.935%}\ hline
text{To-hit = 15} & 3.467 & 2.433 & text{-29.824%}\ hline
text{To-hit = 16} & 2.950 & 1.917 & text{-35.017%}\ hline
text{To-hit = 17} & 2.433 & 1.400 & text{-42.458%}\ hline
text{To-hit = 18} & 1.917 & 0.883 & text{-53.938%}\ hline
text{To-hit = 19} & 1.400 & 0.883 & text{-36.929%}\ hline
text{To-hit = 20} & 0.883 & 0.883 & text{-0.000%}\ hline
end{array}

I'm producing these values just as a confirmation against Andras' values: I believe the small difference between our results are just the consequences of using different damage dice, and that based on my own results, their values are probably correct for their scenario.