Very early on in my running career, someone told me about marathon pace predictors.
The most popular one is McMillan's, which is found here.
The way it works is you put in an actual recent race time and it will calculate what you should be able to run at other distances (assuming proper training).
You can also do it in reverse, putting in a recent marathon time, and seeing how fast you should be able to run a shorter race.
I don't know exactly how the science/numbers behind the predictor works. It obviously doesn't assume that you can run your 5k pace for a marathon, but I don't know exactly what factor it increases your time by to extrapolate from your current time. And of course there are questions about how accurate any prediction is. The old version of the McMillan calculator had a warning (and it might still be there, I just haven't looked) indicating that the longer and the more recent your "actual" race time is, the more accurate your "predicted" time will be. So if you ran a 20 miler at just a bit faster than your marathon pace, you're likely to be able to sustain your marathon pace. While if you run a super fast 5k, it's not necessarily true that you'll be able to perform at the same level for hours on end.
The predictor has always been interesting to me. I can see what distances are my strongest (they predict the fastest marathon), and what distances are ones where I chronically underperform (predicting the slowest marathon).
Thus far, my mile and my 5k PRs predict me very close to my goal marathon time for Berlin this September. The problem with looking at my mile and 5k predictions is that both my mile and 5k PRs are old. My weakest distance is the 15k (I also have very weak 8k and 20 mile times, but those are old or are from training races). But a lot of my PRs, even recent ones, predict me very close to my actual current marathon PR.
I start training for Berlin in about two more weeks. I've run two 5ks very recently, and one half (two days ago), and all three of those predict me in the exact same time range for a marathon -- less than 2 minutes' difference! Which makes me think there's something to be said for the predictor. But what I am especially happy to see is that the 90 second window based on my 3 recent races as a predictor all show me getting a marathon PR by about 4-5 minutes. For me, when I look at the predictor with my current 5k time, it estimates that I can run marathon pace miles nearly 2 minutes slower than my mile PR, and just over a minute per mile slower than my 5k PR mile pace. Looking at actual recent race times instead of my all-time PRs, my current 5k race time has me slowing down just over 1 minute per mile to run the marathon. Do-able I hope.
While a PR of 4-5 minutes is not the 11 minute PR that I want to see in Berlin, it would be one step closer to that ultimate 11 minute goal. So all in all, my spring focus on speed training seems to have gone pretty well. I have gotten my race times closer to where they were before my 2012 season of running with a weight vest. And at least now (according to the predictors, for what they're worth), I've set myself up as best as I can to train for a PR at the marathon.