One of the big divides at NPP is between those editors who think it is as Kudpung put it "important to catch the creators while they are still online and logged in". And those of us who think that while it is good to quickly warn bad faith editors, good faith editors respond best to being helped, having some slack cut for them and being given a little space; But are driven away by threats to reject their work and delete their article. This divide makes certain changes difficult to agree because we have very different perceptions of the problem, there are some changes that don't involve this and that both sides can support, but making changes at New Page patrol would be much easier if we had some research behind this to show whether very promptly templating articles and warning their creators encouraged or deterred the creators from fixing their articles, and in particular whether doing this very quickly was either more effective at getting articles improved or more efficient at driving newbies away .
I've done quite a bit of trawling through the BLPprod queues, and while I'm not claiming any statistical robustness, my impression is that if articles with a BLPprod tag get rescued it is usually an experienced editor who does so. But I would be open to persuasion on this, as I hope would be those who take the opposite view to me. Speedies are harder to measure in this way because they get deleted so quickly I find it difficult to spot ones where the author has been spurred on to make greater efforts to improve their article by the threat of deletion.
If we do some research it would be important to focus on the relatively recent data, as earlier this year the system on EN wiki was changed to default to emailing editors when they received a talkpage message. One would presume that this would reduce the advantage of catching the editor before they left.
Whether the results showed a pattern that the quicker the templating the less likely the editor was to stay, or the reverse, I believe it would be much easier to improve NPP if we had some research. I remember trying to get something like this into the WSOR program for 2011 and we might be able to use some of the datasets created for that program. NB G3, G7, G10, G11 and G12 tags need to be excluded, or better still the results analysed by deletion code otherwise you risk this being skewed by our effectiveness at targetting attack pages and driving away vandals.