At what point in a Dominion game should a player stop buying middle value cards (e.g., Silver or other cards more suited to mid-game play than Province/Colony-buying sprees)?
One of the first things I try and figure out when evaluating a set of kingdom cards is to try and figure out which way I think the game will end. Is it going to be a three-stack game or a province/colony game? Are there cards that push it in one direction or another? Am I going to try and push it one way or another, and what do I know about the tendencies of my fellow gamers (and their first turn purchases)? If I believe that it's going to be a three-stack game, or I will be pushing it that way, there is no mid-game for me. I'm usually going to try and buy as many VP cards as I can while keeping my deck's head above water. Since I've evaluated the tableau and decided that there won't be an extended run on Provinces/Colonies, the bar for success in my deck has changed. Trying to keep a consistent 5 purchase power (for Duchies) is a much different thing than trying to build to a consistent 8 or more, and in such a deck, mid-power cards can always be useful.
If I've decided that it's going to be a game where one of the big VP stacks is going to run out, then presumably I've decided how I'm going to drive me economy. At every step of the way, what matters more to me than anything else is answering two rolling questions: what is the average value of the cards in my deck and how does my purchase change that number. Adding silver to a starting deck changes things from .7 $/c to .82 $/c, pretty good. A Laboratory, added to a starting deck, does a little worse, .76 $/c. But, later on in the game, if you've got a deck that is averaging 1.5 $/c over, say, 25 cards, adding a silver doesn't do much, 1.52 $/c. But a Lab does better, 1.62 $/c. Etc. Obviously, I'm not computing this stuff directly at the table, but you should have a rough idea of where you are at and should only buy cards that can move the needle. Stuff like Silver (or similar cards that don't have +Cards attached to them) is limited, and should be avoided once the average value of your deck is getting close to their inherent value.
One important thing is knowing how many reshuffles are left for you in the game. Different kingdom cards will lend themselves to different deck sizes, which in turn can tell you how much victory cards are going to weigh down your deck. If another player is looking to end the game soon and is ahead, you probably want to scramble to get as many points as you can, even if its from duchies and estates instead of provinces. This is really helpful when deciding when to buy the lower victory cards.
Its usually fairly easy to get an idea of how many turns are going to go by before the game ends by looking at how many cards are left in the various relevant piles and how quickly players are buying them. You should always have a general sense of how many cards are in your deck. Within 5 cards of the total size is good, though I've seen people who can tell you how many cards are in their deck AND discard pile at any point in the game.
Divide by the average number of cards you're drawing per turn to get a rough estimate of how many reshuffles are left. If you have a treasure heavy deck and not many sources of +cards, you may draw an average of 6 or less cards per turn. If you've got a super action chaining deck you may have an average of 9 or more cards per turn. Obviously its very difficult to get a precise estimate of the number of reshuffles left due to its many influencing factors, but you can usually do well enough. Unless you have more than 6 reshuffles left, just buy the dang province. If you have 3-4 reshuffles left, consider getting that duchy. I you have 1-2, get an estate or whatever the hell you can because there's no point in doing anything that doesn't get you direct points.
Obviously this is a very inexact science, and these are just very general rules. There's a ton of factors that affect your decisions on every turn. There's 2 important things in Dominion: