A couple of weeks ago, I was going through a cabinet where I store books that I have owned for as long as 35 years. I have
the complete works of two of my favorite authors, Hemingway and Faulkner, that I read 25-30 years ago and have decided to
reread this summer. While selecting the volume with which I will start....Hemingway's, "The Sun Also Rises"....I came upon
the first book about running that I bought when I started running 20 years ago. It's the first edition of Runner's World's
"The Complete Runner" which was published in 1973. I had forgotten that I have it. The content of it is quite different than
that of the latest (1997) edition. Some of the articles are reprints that had previously appeared in RW's magazine. Others
were written specifically for the book. Most are as valid today as when they were written. A few are a bit archaic, however.
For instance, in several articles about training there is no mention of threshold training or hillwork....two staples in the
regimen of today's serious runner.
One article in particular caught my attention because it relates to a subject that we discussed in depth on this forum
a couple of weeks ago....predicting a marathon time. The article documents what appears to have been one of the first attempts
to develop equations to predict a marathon time using personal characteristics and training data. It was written by Paul Slovic,
a psychologist at the Oregon Research Institute, and was published in the October, 1973 edition of RW. Slovic surveyed participants
in the Trails End marathon that was run at Seaside, Oregon on February
24, 1973. He collected an extensive amount of data about the training of the participants. He then used the reported data,
along with their finishing times in the marathon, to develop a set of eight equations to predict marathon times using a combination
I found the article to be very interesting. I'll get into the equations a little later. However, I would first like to
talk about a few things that I gleaned from the article that, I think, illustrate the changes that have occurred in marathons
and marathoners in the last 30 years.
(1) Slovic said that the Trails End marathon was "....a particularly attractive setting for such a study for several reasons:
It draws one of the largest fields of participants in the US, and the runners cover the entire range of experience and ability,
from national and international class to novice." There were 541 participants in the race, with 441 finishers! And
this was one of the largest marathons in the US
in 1973!! A marathon of that size today would have trouble surviving. And it certainly wouldn't draw national and international
class participants. It is a microcosm of the 1970's marathoning community that would be hard to find today.
(2) Of the 441 finishers, 178 men and 6 women returned Slovic's questionnaire, which had been included in all participants'
race packets. That's 42% of the finishers. If we assume that they are representative of the entire field, then note the ratio
of men to women....30 to 1!! Today, it isn't uncommon to find marathons where more than half the finishers are women.
(3) The average finishing time of the 184 responders was 3:28. The median time was 3:24. Approximately 75% finished under
4-hours. Compare those data to a current marathon. The times are about an hour slower today! A time of 3:24 would put a runner
in the top 5-10% of many (maybe most) marathons today, instead of the top 50% of yesteryear!!
(4) The average longest training run among the 184 responders was 18 miles. The median longest run was 20 miles. It gets
even more interesting when we look at the average longest run as a function of finishing time. (Slovic summarized the data
into eight 15-minute finishing time categories, such as 3:01-3:15; 3:16-3:30; 3:31-3:45; etc. The fastest category was 2:20-2:45
and the slowest was "over 4:30".) Only those in the three fastest categories of 3:01-3:15 and faster (75 responders) averaged
a longest training run of 20 miles or longer. The other 109 responders, who finished as fast as 3:16, averaged 19 miles or
less as a longest training run. Also, only the three fastest categories averaged 2 or more runs of 20 miles or longer. The
other five categories averaged from 0.2 to 1.7 runs of 20 miles or longer....remember, that includes runners as fast as 3:16-3:30!
Today, few marathoners, even first timers, show up at the starting line with less than a couple of 20+ milers under their
(5) There was a very direct correlation between finishing time and both training mileage and peak mileage week. Also, it
appears that training mileage was a bit greater than that of today's typical marathoner, although probably not much greater
for the faster categories. There were just a lot more runners in the faster categories....and the slowest category didn't
go nearly as far back in time or depth as in today's marathons. The fastest category (2:20-2:45) averaged a peak week of 92
miles with an overall weekly average during the 8-week period before the marathon of 68 miles. The average peak week in the
five categories over 3-hours ranged from 64-42 miles and overall weekly averages for those categories during the 8 weeks before
the marathon ranged from 42 to 25 miles, with a consistent decrease in both areas as finishing times got slower. (Apparently,
one super-stud in the 3:31-3:45 category who had run 24 marathons skewed upward all the mileage numbers, as well as other
variables, in that time category.)
I found the article to be interesting, although not very helpful in answering today's marathon prediction questions on
this forum. It really might be of more benefit in describing how, in general, marathoners of 30 years ago trained as compared
to today's. Of course, we already know that they raced better.....just as Hemmingway and Faulkner were far better writers
than those of today. ;-)
The article does offer a rare insight into the training of the typical marathoner of 30 years ago, relative to performance.
I found it particularly interesting that he seems to have been training at a level generally comparable to that of today....with
somewhat less emphasis on 20+ mile runs and, perhaps, a bit more total mileage....yet was racing considerably faster. I think
that further supports the opinions that I reached in a previous thread about why the American marathoner has declined.....today's
marathoner brings less baseline development from his/her youth to the sport than those of 15 or more years ago.
Now, to the race prediction equations presented in the article. Slovic used statistical analysis techniques to develop
a set of eight equations that use various combinations of personal characteristics and training data to predict a marathon
time. Rather than try to repeat and explain them here, I will refer you to the original article, which I put on my homepage.
You can find it at http://mysite.verizon.net/jim2wr/id54.html. Now, you didn't think I was going to tell you where to find it until after you read my preceding comments, did you?
So, how valid are Slovic's eight equations? To get a checkpoint, I applied them to my personal data from the year (1989)
in which I ran my marathon PR of 3:22:27. The first four equations use the fastest mile run in the past year. However, I never
ran a mile race or time trial during my first running life. The fastest timed mile that I ran was a 5:50 first mile on a flat
5k course and it was two years before I peaked in 1989. So, for this exercise, I have assumed that if I had run a mile race
or time trial during the year before my PR marathon, it would have been about 5:35. The following are my results from the
Fastest mile - 5:35 (estimated)
Previous completed marathon - yes
Miles run in previous 8 weeks - 408 miles
Longest run - 22 miles
Number of 20+ mile runs - 2
Max mileage week - 62 miles
Ponderal index - 13.3
Age - 51
Equation results (deviation from actual race performance)
Equation 1 - 3:23:19 (00:52)
Equation 2 - 3:24:38 (02:11)
Equation 3 - 3:24:44 (02:17)
Equation 4 - 3:29:13 (06:46)
Equation 5 - 3:21:53 (00:34)
Equation 6 - 3:22:05 (00:22)
Equation 7 - 3:28:44 (06:17)
Equation 8 - 3:35:47 (13:20)
Five of the eight equations predict a time that is impressively close to my actual time....they varied by 00:22 to 2:17
from the actual time that I ran. Even the three equations that missed by more than that were only 6:17, 6:46, and 13:20 from
my actual time....and all three were on the conservative side. Thus, if I had based a race plan on any of them, I should have
been able to make up some of that conservatism in the late stages of the race.
Equations 5-8 are based on real, hard data, whereas equations 1-4 include an estimate of what my fastest mile time might
have been at that time. Interestingly, the equations that don't include an assumption produced both the two closest predictions
and 2 of the three worst predictions relative to my actual time. However, I think that anomaly is clarified in my analysis
of the results of these equations, as follows:
(1) The four equations (1,2, 5 and 6) that come closest to predicting my actual time all include total mileage over the
previous 8 weeks as a variable.
(2) A comparison of the results of the two best predictors of my actual time (Equations 6 and 5) to Equations 7 and 8,
which were the farthest and 3rd farthest from my actual time, indicates that total mileage in a training program is more significant
for me than the maximum mileage week, which supports (1) above.
(3) A comparison of the results of the two worst predictors (Equations 4 and 8), both of which use the variable of the
number of 20+ mile runs, to Equations 3 and 7, both of which include the length of the longest run, indicates that the length
of the longest run is more important than the number of them. Now, that surprises me. However, I also note that all four equations
predict a slower time than the remaining four equations that all incorporate total mileage as a variable and confirms (1)
above. So, this factor is really a sub-analysis of the worst of the worst, which relegates it to a secondary level in fine
tuning a training program.
My conclusion is that total mileage is more important to me than the length of the longest run or the number of 20+ mile
runs. In other words, the totality of training is more important than the peaks. That is a theme that I have preached on these
forums for 7 years. Now I know why. ;-)
Regardless of the absolute veracity of these equations, considering that they are based on a very limited survey base,
my point is that equations such as these can be useful as tests against a few actual marathon performances, which just might
enable one to zero in on the set of variables that best predicts his/her marathon performances. In turn, that can help one
to identify the elements of training that are most significant to himself/herself and fine tune his/her training program.
I am impressed by the results of this early attempt to develop a method to systematically predict a marathon time. I think
that Slovic was on the right track by incorporating several variables into a family of prediction algorithms. They might not
be as convenient to use as simply plugging a race time into a calculator. However, his approach embraces many of the variables
that make us all different from each other and cause many race calculator "inaccuracies." More importantly, it can provide
constructive feedback to a training regimen. I just wish that there had been more follow-up to his study to refine his findings....and
that, in his questionnaire, he had elected to solicit best 10k time during the previous year, instead of best mile time. ;-)