we know the mcmillin predictor, but now this bases your performance off mpw.
http://projects.fivethirtyeight.com/marathon-calculator/
pretty close for me. so you low mileage triathletes running slow… run more damnit!
I plugged in all my info from when I ran my last marathon, and it spit out a 2;50.29. It was waaaay off, I actually ran a 2;36. And I of course know why, it does not take into account cross training endurance, and there was no where to plug in the one 3+ hour ride a week I did while pointing for this race. Or the years of endurance training in swimming and cycling that is just locked up in your body waiting for that 20 mile wall!!
My fastest marathon time when I was averaging 50+ miles a week running in my mid 20s (running was all I was doing) was less than 5 minutes faster than several marathons i did in my early 30s when I was tri training and averaging less than 25 miles per week of running.
I used the predictor with stats from my most recent marathon training last year (my first marathon in a decade) and the predicted time was more than 40 minutes faster than what I ran. I was definitely on the low mileage plan (avg. 18 miles per week for the 8 weeks prior to the marathon) but I was still able to run decent times for 5K due to the sprint tri training/racing I’ve done over the past 8 or so years. Maybe I could have been closer to the predicted time with the same nos. if I was 26 or 36 instead of 56. 
The cross training certainly didn’t help much in my marathon last year.
A geeky side note: Pete Riegel, the guy who developed the algorithm cited in the article, may have measured the 5K course where I got my time for the predictor. We both live in Columbus, OH.
Interesting… Last year I ran the Buenos Aires Half Marathon in 1:23:53 and 5 weeks later the Buenos Aires Marathon in 3:06:59, I was 53yo and running 44mpw on average for the last 16 weeks, predicted time with this data is 3:07:01…