Ed Kyle wrote:

Although Apstar 5 happily made it to its station on
orbit (a mission success for Apstar 5), the launch
vehicle unquestionably suffered a serious failure.
This failure cannot be ignored by those who take on
the substantial financial risk of the satellite trade.

I think this is a very reasonable way to look at it. Apstar 5 was a
launch vehicle failure (the vehicle failed meet its advertised minimum
performance specifications) but not a mission failure. If you are
measuring the ability of launch vehicles to meet their specifications,
then this clearly fails on the "fail" side of the fence. If you are
measuring percentage of payloads that perform their designated mission,
that is a completely different story.

It might be interesting to look at individual stages, rather than LV
statistics. In that case, Apstar 5 would count as a success for Zenit
sl stages 1 and 2, but a fail for Block DM. To get your overall chance
of failure for a given LV, you could then combine the stats for its
stages. In some cases (block DM being a possible example) this would
give you a larger sample size for some components.

Of course some failures could be hard to blame on a specific stage.
Perhaps you need another category of non-stage specific things. You
would also have to consider what variants of a given stage you can
actually lump into the same catagory (proton DM vs DMSL).