I've been meaning for a while to write a guide for friends/family about this. I thing that first you really have to have an understanding of why this is happening, what the goals (hidden and obvious) are for those engaging in the spying, and determine where you stand on the subject before you can't make any sort of plan for implementing the level of privacy you desire. From there the entire discussion is about capabilities and methods. I will forgo the first points in the hope that the hacker mentality still thrives at least somewhat on /.
First, there was metadata,
Metadata combined with modern algorithms and big data can give it's owner just about everything on you. Here is what I consider metadata
(this assumes every point compromised except local, imagine NSL's etc)
IP - Your ISP will always know this. Circumvention includes tor, i2p, other anonymizing technologies. VPN does not secure your metadata. Wardriving. Rooted boxes.
MAC - Much less of an issue, can be spoofed easily. Usually not know outside of edge network devices or ISP.
Time - Heavily used but not well understood. Correlation of login times to compromised activity elsewhere holds up pretty good in court. The longer they've been watching you, the more dangerous to security this is.
Other machine identifiers (agent strings, cookies, DNS, etc) - mostly a software (and knowledge) issue. Have to be able to prevent DNS leakage, spoof agent strings, keep machine clean of cookies (including harder to find/remove cookie types like flash) If you are on windows... this is your most likely failure point.
Then, there was low hanging fruit.
Low hanging fruit: cloud services (webmail providers, social networking, cloud apps, cloud storage/computing, voip/txt chat protocols, etc) If you use these services you must expect them to be compromised and not private. You can choose to not use these services, or compartmentalize use of them (which is my preferred method). Data poisoning becomes more relevant here. Now, you can attempt to be anonymous while using them (say tails(tor) for facebook), but the data is still compromised. But if they can't tie my identity to X, why does it matter. Two reasons: one, because if you are using a service like that, all it takes is one slip up to tie everything to you, and two, because there are other ways beyond even time-data correlation to do so (writing analysis for example)
So, assuming you have figured out how to be relatively anonymous and encrypt your data (ssh, tcplay, dm-crypt, gpg) You self host as many services as possible, and directly connect to people/sites you "trust". You have in intelligence terms "gone dark" or "dropped off". I'm going to ignore the issue of DPI for the moment.
This is where the majority of people who care about privacy want to be. They want to be just enough of a hard target that it's not easy to grab up their info. This is what the 90's cryptowars were about. The ability to go dark.
The problem with this state is twofold: First, your data can still be retroactively inspected. So that AES-256 you think is nice and secure is finally cracked by the NSA (if it isn't already). Then they run it on gobbled up data from the past, and suddenly your encryption is worth jack. (save discussion of storage feasibility for another time, some of the math has already been done over on Schneiers blog)
Second, once you become a target for other reasons, they will resort to other methods. First with off-site but close compromise. Usually ISP. Then escalated to remote compromise (trojans, keyloggers, etc through 0-days or backdoors) If for some reason you are still safe at this point, commence black bag operation. While you are at work, they break into your house and plant a physical keylogger, audio bug, copy HDD, install trojan (MBR not encrypted? evil maid!) or any other number of growing possibilities. This boils down to your physical security. Think your ADT alarm system works? Think again (well, this depends on who you pissed off, normal FBI team probably thwarted, special FBI team or JSOC/OSI/CIA/NSA etc? No problem. Cameras and a self-managed security system in parallel with a more obvious one like ADT might work here. Of course, by the time you reach this point... you have much bigger issues and are likely to be harassed consistently or suicided/plane crashed eventually.
Bottom line, security and usability have an inverse relationship, and you have to decide what level works for you. Just getting to hard-target mode should be fine for most people. Work in R&D and make frequent trips OCONUS though? You better step up your game.
Or you could just stop using technology and go live in the mountains.