YMMV (Your Mileage May Vary)

Another article I wrote for Marathon & Beyond required a great deal of analysis and effort. But the result is one of my favorites. I also received a whole lot of feedback about it, mostly good. There was even a reference to it in Runner's World. So from the January/February 2002 M & B, here is YMMV.

YMMV (Your Mileage May Vary)

The Quest to Determine the Most Relevant Training Elements Begins with Examination of One Runner’s Data

by Dan Horvath

Auto manufacturers are careful to qualify their gas mileage claims with the phrase, “Your mileage may vary”. And so it is with distance running as well. What works for one person may or may not work for another. There will be varying degrees of correlation between one runner’s training regimen and his or her results and those of another.

One of the problems in determining how best to train for and run marathons is that there are so many varied opinions about what works and what doesn’t. Every coach and writer has his or her own biased views about what you ought to do to maximize your training and realize your potential. Unfortunately, much of this is not backed up by hard evidence. In many cases, the data required for adequate analysis simply does not exist.

Over the years, I’ve been collecting an extensive amount of data about my own training and racing. Although this data applies only to myself, and your mileage may vary, some analysis will still be useful to others. At best, you may be able to apply some of the deductions to your own training. And at the very least, you will be able to find out what has worked or not worked for this one individual.

Based on the analysis of the data, we will be able to examine the following:

• Correlation between overall training factors (such as total mileage, overall pace and mile repeat pace) and marathon performance

• Correlation between various long run factors (such as long run pace and distance) and marathon performance

• Correlation between various other miscellaneous factors (such as weight and age) and marathon performance

These elements were chosen for two reasons: first, they may actually provide useful information, and second, I actually have data relating to them.

The Data

In order to determine which training elements have the most influence on a given race, we must first select the marathons, and then the related training in which to study. Here are the criteria for choosing which of my 50+ marathons to include in the analysis:

• There must be data available relating to all of the factors noted.

• The marathon itself must have been a serious effort, not just a “long training run”.

• I must have not become injured or sick just prior to, or during the race.

• I must not have “hit the wall” badly during the race. This is because these efforts are generally precipitated from starting out too fast, and then finishing with a much slower time than I would have had, had I simply run a steady but slower pace.

In other words, I was interested in the races where I would and should expect a certain result, based as much as possible on the training itself. I came up with 30 races that fit the criteria.

It so happens that for many of these marathons, the expected result was a time in the neighborhood of three hours (for more on how to break 3 hours, see “Six Fifty-Two” in the November/December 1999 issue of Marathon and Beyond). Some of these efforts were successful and some not. For the purposes of this study, however, that objective shouldn’t matter. The correlation between the various training elements and the marathon performance ought to apply regardless of the time goal.

Several environmental factors, such as weather conditions, and what one ate the night before the race, may cause variation in marathon performances. Using larger quantities of data help to mitigate these in such a way that they begin to lose significance, however. They will not be taken into account for this study.

Most marathon training programs span a time period of 12 to 18 weeks. Although a longer build-up period is generally preferable, often the first several weeks are devoted to base building. The specific factors to be studied here assume that the subject already has a solid base, so the period chosen for the data is generally one of 12 weeks prior to the marathon.

Information, Correlation and Causation

Although I’ve collected all of this data, I haven’t always done so great a job of processing it into useful information. Sure, I do look back once in a while to determine some of the training that lead to a successful race effort. But these looks back are not very qualitative or quantitative. A detailed analysis at the data is the goal of this exercise. The transformation of the available data into useful information involves correlating the training data with the marathon performance data by use of statistical methods.

Probability and Statistics were not among my strongest courses back in my college days. Those days, in turn, happen to be a long time ago in a galaxy far, far away. I am fortunate, however, to have two daughters in college who were able to help their dad out with his “homework”. You won’t need to be a statistician to understand the basic principals here, however.

To show correlation, each set of training data is plotted against the marathon performance data. In addition, a correlation coefficient, R, is computed. R is a number than describes the relationship between two sets of data. I’m sure there’s a reason why it’s called R, but please don’t ask me what that reason is. To make the correlations more meaningful, R is squared and then multiplied by 100 in order to arrive at the percentage of variation in one factor explained by the other factor.

To put it more concisely, for each set of training data versus the marathon performance data, a scatter diagram will be created, and the percentage of variation will be calculated; the larger the percentage, the stronger the correlation. The actual formula for R, as well as that for any other method of measuring correlation, is left as an exercise for the reader. For my part, I used Microsoft Excel.

There is one other thing. Each daughter independently informed me that, “correlation does not imply causation”. Causation, I said, is the whole idea of training. On the other hand, it is conceivable that a strong correlation between, say, mile repeat pace and marathon pace may simply be due to being in shape (or young, or not being injured, etc.), and not due to one thing causing the other. More on this later.

Overall Training Elements

Correlation between marathon pace and overall mileage

A non-running acquaintance overheard a friend and I discussing his most recent marathon. The acquaintance said, “Wow, a marathon! How did you train for it?” My friend, who didn’t want to be bothered explaining the intricacies of his training to someone who wouldn’t understand (or even ultimately care), replied, “I ran a lot.”

I liked his answer. It really sums things up, doesn’t it? If you want to run a marathon, you’ve got to run a lot. If you want to run a fast marathon, you should probably run even more. But how close is the correlation between number of training miles and the resulting marathon pace?

In comparing the total number of miles run during the 12 weeks prior to a marathon to the actual marathon pace, there is a fairly strong negative correlation, and a percentage of variation of 11.47%. This means that for me, more miles do relate to a faster marathon pace.

Correlation between marathon pace and overall pace

There can be no doubt that training is specific. In order to run fast during a race, you must run fast during your training. The question is, how specific is it? How close is the correlation between the overall training pace over the previous 12-week period and the resulting marathon pace?

The correlation analysis yields a percentage of variation of 0.42%. This number is so small that it is almost insignificant. Could it be that it doesn’t matter how fast I train? Clearly, overall training pace is not as important as I had thought.

It occurs to me that perhaps for my more successful efforts I had more variation in my training pace. That is, I may have done my fast runs faster, although my slow runs may have been slower. I may also have done more fast training. The best way I have to measure this sort of thing is to examine the number of races during the marathon build-up period, and my mile repeat pace. The assumption is that I would be doing these runs much faster than my normal training pace. In addition, the number of races gives us some idea of quantity of fast training efforts. See below for the analysis.

Correlation between marathon pace and the number of previous races done during the marathon build-up

Does it help to run other races as part of the marathon training? I often try to fit some shorter races into my schedule. For the data, I examined the total number of races run during the prior 12 weeks. Marathons or other races done as “training runs” are not included, but marathons that were run as races are counted as races. Of course for a serious attempt at a fast marathon, you generally should not be running another serious marathon within the previous 3 months, but there are always exceptions. Your marathons may vary.

With the exception of those few longer ones, the majority of the races were in the 10K range. It would have been nice to analyze these race paces versus the marathon pace, but the distances do still vary; it may be anything from 5K’s to other marathons. Pace calculators, used to compute one distance’s anticipated pace based on a race pace of a different distance, are available elsewhere.

Comparing the total number of prior races run to the marathon pace itself yields a percentage of variation of 15.20%. This means that for me, it appears that it does help to run other races prior to the marathon, and the more, the better.

Contrast this with the philosophy of making the marathon one’s first road race of any kind!

Correlation between marathon pace and average mile repeat pace

For all of the marathons on the list, my preparation has included several sessions of one-mile repeats. These interval sessions are generally done in the form of 1600 meter repeats, with a 400-meter rest in between. Of course they also include warm-up and cool-down periods. I usually do one session of 6 to 9 of these each week during the previous 12 weeks leading up to the race. The exceptions come when I may have substituted other speed work, say 1200 meter repeats or a race or tempo run. I also usually avoid or curtail these sessions during the final two weeks. Due to all these variables, I was not able to analyze the number or frequency of repeats; only their pace.

The pace for the repeats is important, however. I usually try to run them at close to my current 10K pace. How closely correlated is this pace with the resulting marathon pace? The percentage of variation is 34.81%. This indicates a strong correlation: a faster average mile repeat pace definitely does relate to a faster marathon pace.

Long Runs

There is general agreement that the long run is the cornerstone of any marathon training program. Opinions vary widely, however, on how long the long runs should be, and the optimum pace for these runs.

Correlation between marathon pace and distance of long runs

There has been much controversy over the years regarding how long long runs ought to be. Some say that they should not exceed 20 miles, so that one is able to resume mid-week training sooner. Others have opined that in order to prepare for the rigors of the last 10K of the marathon itself, much longer training runs, say 27 to 29 miles, are necessary. A more recent, and I think, sensible, notion has come along that one should run long runs of the same amount of time that one expects to run the marathon. This may indicate a maximum of about 22 miles, depending on the difference between marathon and long run pace.

I calculated the average distance for all runs 18 miles and over during the 12 weeks prior to the marathon. These runs have generally been done on a weekly basis. The comparison to marathon pace yielded a slight surprise: the percentage of variation is only 0.02%, so there is no significant correlation. This means that I haven’t necessarily run faster marathons when my long runs have been longer; the distance of those runs doesn’t appear to matter. Perhaps I’ve been over doing it a bit, however. It may be better to save my legs a bit for my other training.

Another philosophy that I’ve adopted in recent years is to alternate between very long and semi-long long runs. One week I may do a 25-miler, followed by a 20-miler, and so on. I still like this idea, but even so, perhaps I should cut back on those distances a bit.

Correlation between marathon pace and long run pace

Yet another source of controversy over the years involves the pace of long runs. Is it better to run them at or near race pace, or should they be run at a much more leisurely, conversational pace? The arguments go way back.

Over the years my long run pace has varied quite a bit, as I’ve oscillated between the differing philosophies. Generally, however, it has been about one to one and a half minutes per mile slower than the resulting marathon pace.

In comparing the average pace for all of the long runs over during the 12 weeks prior to the marathon versus the actual marathon pace, we see that the percentage of variation is 1.03%. This is not a strong correlation, but it appears that in general, the faster I’ve run my long runs, the faster I’ve run my marathons.

It may be advantageous to vary the pace of long runs. Begin the long runs at a pace of 2 minutes per mile slower than your expected marathon pace, pick up the pace a bit during the middle miles, and then run the final few miles at marathon pace.

Marathon Performance with Training Factors

Other Factors

Some factors have nothing directly to do with training, but may have a strong effect on a marathon performance. Some of these, such as ancestry and predilection to train hard, can’t be measured, but others can. Two that I’m particularly concerned about are my weight and my age. Both appear to be headed in an upward direction, and although I may be able to do something about one of them, I wonder how well they relate to my marathons.

Correlation between marathon pace and average weight

I wasn’t so sure I was going to like this one. A positive correlation would mean that keeping my weight low correlates with faster marathon times. And, like many people, I have some amount of difficulty keeping those pounds off.

The average measurement for all weigh-ins over the 5 weeks prior to the marathon was compared with marathon pace. We’ll have to assume that the scale is consistent. The analysis for the correlation between marathon pace and average weight yields a percentage of variation of 33.49%. This is a significant correlation.

And it is just as I had feared. I need to keep working at fighting that battle of the bulge.

Correlation between marathon pace and age

I was sure I wasn’t going to like this one. A positive correlation here would mean that my race paces are getting slower as my age has increased. And since my marathon times appear to be getting larger over time, and since I had not been able to manage a sub three-hour marathon since 1996, I thought the analysis would show a strong correlation.

This time the result, a variation of 5.44%, was a mildly pleasant surprise. Although the correlation is positive as expected, it isn’t too far from zero, so it is fairly weak. Maybe there still is some hope for me. Nah!

Conclusions

Some of us like to believe in cause and effect. That if we follow a good training schedule, if we just work hard enough, we’ll see a successful conclusion to our quest of running a great marathon. It ain’t, however, always necessarily so. Sometimes we can do everything right, and the race itself comes out wrong. Sometimes we don’t seem to train as hard as those other times, and we still manage to turn out a good race. For me this happens at least once per blue moon… but only during leap year. Part of this is due to environmental factors; part is due to individual differences.

One of my friends chides me about training so hard. He doesn’t appear to work very hard at his own training, and yet he seems to achieve excellent race results. I believe this is evidence of the YMMV thing at work. On a related subject, there is something else I noticed in my own data: diminishing returns. It appears that I don’t need to work very hard in order to run a 3:15 marathon. But to run 3:10 or better, I need to train much harder than one would expect for a 5 or so minute gain. This may be strictly a perception on my part, but is sure seems real.

Compiling the raw data into the table turned out to be more work than anticipated. But it was also more enjoyable than expected. I was able to re-live all those long training runs and speed sessions. Based on the analysis, I’ve learned that my training has actually been remarkably, and surprisingly consistent over the years. Some training elements, however, appear not to matter as much as expected. I had thought that Overall Training Pace, Long Run Distance and Long Run Pace would show stronger correlations than they did. I had expected that there was a strong relationship between mile repeat pace and marathon pace, so that result was not a major surprise. In addition, it appears that running higher overall mileage and doing several prior races relate fairly strongly to marathon performance. As does weight.

This brings us back to causation. Perhaps the strong correlations between marathon pace and weight, or between marathon pace and mile repeat pace are actually due to being in similar shape for both sets of data. The mile repeats do, however, still represent an important training element. It’s just that they’re not the only one. Indeed, it may be beneficial to study some combinations of training elements.

I’m going to carry on with those one-mile repeats, continue to include races in my schedule, and try to keep that weight under control anyway. Your mileage may vary.

Dan Horvath would like to thank his daughters Veronica Horvath and Valerie Horvath, as well as friend David Couper for their assistance.

BIO

Dan Horvath is a software engineering metrics consultant, who has also been known to do a bit of running and writing. He recently completed the most recent of his 59 marathons and ultras: the Mohican Trail 100, his first attempt at that distance. Dan lives in Broadview Heights, Ohio.