DougM:
You make a good point. The tempo paces I offered are a starting line, just as Daniels or Vigil or Canovas formulae are reference marks (with small possibilities for error).
Here is something for us to consider, variations in the middle or any statistical schema are small. The ends of schema are where more variations occur. The line of predictability between the 5k and one-hour race pace (found to be LT) is quite predictable because the time range is small. We have a centralized region of predictability which gives us statistical confidence of validity and reliability. On the other hand, if we correlated the mile to the marathon, we would have extremes on the range and thereby less validity and reliability.
If we asked a bunch of milers to race a marathon, we might not be very accurate at predicting who will place first, second, third, etc. if the milers were in a small range of ability. If we had five milers with p.r.s of 4:16, 4:18, 4:20, 4:22, and 4:24, we might have many possible finishes in the full marathon race. The 4:24 miler might even finish first in the marathon while the 4:16 might be last. But, if we take 5k runners with times of say 14:40, 14:50, 15:00, 15:10, and 15:20, we would have far greater predictability in the place at which said runners would finish in a marathon. Better yet would be 10k runners or half-marathoners, obviously.
Our starting point, just like those of other authors like Daniels, etc., in which we use .93 for LT, .90 for half marathon, .87 for marathon pace tempo runs of short, medium, or long duration, may indeed be both predictably accurate (especially at the LT which is the one-hour race pace predicted intensity) and useful for training distance runners. The key would be to be realistic with ourselves about what are inherrent strengths are (power vs endurance) and are training status. A miler may be running minimal mileage and not be able to hold the half marathon predicted tempo run very well because the 5k was predicted from their mile time which led to the predicted LT time. If a person actually runs a 5k or 10k (just divide 10k time by 2.08 to get predicted 5k time...and yes, it is fairly accurate, just look at Daniels charts or Vigils, or Canovas or play around with elite runners times and correlate their performances with the 2.08 factor) and then derives the tempo paces based on the correlates mentioned, then we can say with some confidence that the paces run will be reasonably accurate; assuredly enough to qualify training more effectively than going by gut instinct.
The problem faced by all distance runners is that racing in longer events can't be done on a frequent basis without interupting long-term progress. So, in order to solve this problem, we have to find good estimates of paces to be run to meet physiolgic components of race performance. We can't go out every week and run a hard one-hour race just so that we know what LT pace is, for example, nor can we run a half-marathon or full marathon. We can, however, use reasonably accurate formulae to predict those paces and use the predictions to guide our training. Hit and miss is what we are avoiding. Long term improvement for most of us requires a bit of a scientific, systematic approach; hence, the formulae I offer.