Thx for these links. I’ll need some time for a deeper reading, but after a few hours the first bits of my take home message are probably settled to: there’s no theorical guarantee that pursuing DP (or, more precisely, pursuing any of DP numerous variants) lead to worse result, except that’s what everyone report in practice, so if that’s a pure technical issue it’ll probably be not trivial to fix it.
I like your note that DP generalizability might not be causal generalizability: that’s both a contender for explaining why these seemingly technical difficulties arise and a potentially key thought for improving this strategy.
Thx for these links. I’ll need some time for a deeper reading, but after a few hours the first bits of my take home message are probably settled to: there’s no theorical guarantee that pursuing DP (or, more precisely, pursuing any of DP numerous variants) lead to worse result, except that’s what everyone report in practice, so if that’s a pure technical issue it’ll probably be not trivial to fix it.
I like your note that DP generalizability might not be causal generalizability: that’s both a contender for explaining why these seemingly technical difficulties arise and a potentially key thought for improving this strategy.