You left me dear readers, basking in the onset of summer, and already there were indications that after my marathon training in the winter / early spring and the focus on speedwork in the late spring, summer was going to be much lazier… And indeed, in typical Sisyphean fashion I let my fitness roll down the proverbial hill in July, and again (despite a resurgence in training in August) during my late summer holidays in September; and before I got to threaten my Parkrun PB!

But these were by no means wasted months: Despite the guilt associated with not training, eating and drinking a lot and watching my waistline expand (we’ll come back to this guilt thing), it is good (mentally as well as physically) to intersperse periods of hard training with some easier or even rest periods: And while I may have taken the piss a bit with the rest, at least we had a lovely family summer out of it!

The aforementioned guilt, but more than that, the fact that I had enjoyed the comfortable running sessions which form such a great part of the 80 : 20 running approach that I have been following this year, meant that I did much more running than I’d normally do while on holiday: These were meant to be nice, short, comfortable runs, but the heat and the hills made them a bit more strenuous than originally planned!

And finally I got more intrigued by the concept of “fitness” which Strava recently introduced (many of us originally came across it as a Training Peaks feature), and from which the chart above is taken. It is a nice, pictorial representation of your calculated fitness (measured as the weighted average of your training volume over the past 6 wks), and how that changes over time. It gives an indication of how hard you have been training, but how useful is it really? I have been wondered about this since June:

So I’ve added to the “input data” (i.e. what goes into my training) that I have at my disposal (apart from the plethora already presented to us through our running watches, Strava etc such as Ground Contact Time and Balance, Suffer Score, Weight, Fat Percentage, Heart Rate…), but I am also trying to identify some relevant output data (i.e. that show how effective my training is: weight, fat percentage, 5k race time, Garmin estimate of VO2 Max, Strava Fitness score…) to understand what really matters, what it actually tells me and what I can do differently as a result…

For example, is there any correlation between Strava’s Fitness or Freshness and my Parkrun time? Or my weight? I feel that if I can somehow understand the relationship between the effort (quantity but also quality) I put into my training to the outcome this effort has, I will be able to make better decisions about how to train to achieve the twin goals of increased performance and enjoyment.

So I decided to find out. Demi and Philip had a longer summer holiday than I did, and with more time on my hands than was good for me (I averaged more miles in the two weeks that I was home alone than I ran in the peak marathon training), and with my geeky streak in overdrive, I set to work!

For fitness I used Training Peaks’ fitness measure rather than Strava’s, as this was easier to calculate for past exercises, and I wanted as long a range as practicable to compare it over (I decided to go as back as November 2016, when I first found out I had a place in the 2017 London marathon). The formula to calculate it is in the public domain, all you need is the Training Stress Score (TSS) for each workout, which a free Training Peaks account will provide you with (essentially this is a measure of the training load of a workout: it contributes to your fitness long term, but also fatigue in the short term). This is calculated as the Intensity Factor (IF) of a training session times the duration of that session, for each exercise.

Perhaps unsurprisingly, it turns out that there is indeed a correlation (correlation value of 0.73) between the Training Peaks fitness measure and Parkrun Performance: Nothing surprising there, the more I train, the faster I run. The correlation was strong, but at 0.73 there were obviously other factors playing a part: Again, perfectly reasonable, as even on the same course each race is different: environmental conditions, where you start in the pack (too far back and you are slowed down at the start, too far to the front and you are in danger of being dragged into an early pace that you can’t sustain), how rested and hydrated you are, how you slept the night before etc., etc.

The correlation was however much stronger between my fitness and weight (0.95): knowing myself this again makes sense: my diet doesn’t tend to fluctuate that much, even between the peak of marathon training and rest periods (this is not a good thing, but there you go), so it makes sense that when I don’t train much I put on weight, and when I train more I lose it. Also, my periods of rest tend to be holidays or when relatives visit, and in those times I both train less and eat more, thereby compounding the effect.

Finally, there was also a strong correlation of 0.87 between the Strava measured fitness and the one I had calculated, which was encouraging.

So that’s well and good, but what could I do with that information: It reinforced the importance of training volume (we already talked about balance here), and with some analysis, it helped me determine exactly how much training was needed over a certain period (say a month) for me to start seeing improvements in my Parkrun time and weight. I saw, for example, that my best running followed periods of training when my TSS was 1,100 or over per month.

So the first useful outcome was that it told me the volume of training I needed to fit into any one month. Remember that TSS is calculated as duration times intensity, but as duration tends to vary more than the Intensity Factor (think long runs compared to short, sharp speedwork sessions), TSS is more sensitive to variations of training duration than to training intensity (this is not to distract from the usefulness of a healthy mix of workouts, it is solely a point on what has the greatest impact on TSS).

The next question was how do I go about and achieve that training volume (measured always in TSS) in a month? My TSS calculations would retrospectively tell me if I had achieved it, but how should I construct my training programme to ensure I am getting enough training in a month? While at the same time adhering to the 80 : 20 principle of training intensity (80% of training should be at low intensity and 20% at medium or high), and ensuring I pick the right workouts for my objectives (think race-specific, building on my weaknesses, etc)? Would three runs per week be sufficient, or would I need five? And what workouts should I pick?

Having calculated the TSS for each of my activities since November 2016, I was able to assign a “planned TSS” for each workout in my plan, using an average historical value for that same (or very similar) exercise. For example, I knew that a 12 mile long run was equivalent to a TSS score of 150 – 160, while a 40′ long run would score 57. By contrast a Hill Rep session would only score 40 (due to its shorter duration, and rest between reps), but that did not mean I should only pick my exercises by their TSS score: rather I should make sure that when I constructed a balanced training plan based on my objectives, strengths and weaknesses, the sum of the TSS of all exercises in a month was over 1,100 (and ideally, this would need to increase from one month to the next, as my plan progressed). If this wasn’t the case, I could go back and tweak my plan, adding some extra mileage here, an extra day there, to ensure that it did.

That immediately gave me a view of how my fitness should increase over the next training period (which I set from mid-Sept to Xmas), based on the plan I had constructed:

And suddenly, fitness wasn’t just a backward looking measure, but a forward looking control and forecasting tool as well: So by logging the actual training volume achieved for each exercise, I am able to track my progress (fitness curve to date), compare it with my planned progress (how I had expected my fitness curve to look, blue in the chart below) and calculate a forecast fitness (which is basically how my fitness will change from where I am now, if I follow the rest of my training plan to the letter, red below):

And, having already established a link between fitness and Parkrun time and weight (and then getting Excel to work out the equation for it for me), it also allowed me to very roughly predict how my Parkrun time will improve as my training progresses: Now I accept that this is an indication only (the formula itself is imperfect, and we’ve already discussed that race day performance is a factor of more than just training), but the beauty of it is that it is a dynamic measure: So while the vast majority of training plans will give you a target finish time for the race you are training for, what I have come up with automatically updates your target based on whether you are over- or under- performing on your training!

In other words, now you can see in real-time the impact on your expected race-day fitness (and by extension race performance) of putting a few extra miles in, skipping a session, a week or a combination of all the weird and wonderful alterations that happen to our training plans when they meet reality! Not bad for a homemade little spreadsheet, eh? I am not aware of any other training plan (and I have tried a few!) which automatically adjusts your race targets based on your actual training performance!

And with Parkruns available throughout the country every week to check whether your actual performance matches what was forecast at that stage of your training, you have a great framework to see how your training is really progressing!

Now, this is still unproven, and I appreciate that there are a number of weaknesses with what I have come up with:

1. I have created this model based solely on my own training and race data for the past 11 months: Just because the model seems to work for me over that period, doesn’t prove that it has wider application.
2. Even if it does, the various variables need to be customised to each runner, following a period of observation. This means that it takes some preparation to establish each individual’s planned TSS values for each type of workout, and a period of validation to establish the formula that describes the relationship between input (fitness or form) and output (race performance or weight).
3. While the formulae use fitness (in other words accumulated TSS) as a predictor of race performance, the large implied assumption is that the training plan followed is optimised for the target race, and not for maximising TSS: E.g. I have already mentioned that the best way to maximise TSS (and thereby artificially improve the expected race performance) is to pack your plan with ever longer runs: But such a training plan is not optimised for 5k performance and so actual performance is likely to fall much lower than expected: In other words, the usual guidelines of creating a good training plan still hold, e.g. build towards race-specific workouts, include a rounded selection of workouts, make sure you address any weaknesses you have identified in your performance, train in cycles, include easier “rest” weeks etc.
4. This model is built around Parkruns, for the obvious reason that they provide an easily repeatable race which I could use to compare weekly changes in my fitness, weight etc with. But if it turns out that this model works, then a next step would be to extrapolate predicted performances in longer races, using one of the many race time predictors which exist on the internet, e.g. this one from Runners’ World – accepting however that such an extrapolation will further reduce prediction accuracy (another calculation which may not be very accurate, some people have a bias towards shorter or longer distances, etc).
5. My local Parkrun recently changed its course slightly (taking away congestion at the start and a hill at the end), which caused an immediate hiccup in the correlations: the course has become easier, so my forecast times are a bit off, as they were calculated on a different course. Hopefully, when I have run the new course enough times, I will be able to adjust the formulae and correct that.
6. All these are estimates, based on information available to date, and assuming everything else remains the same; so if, for example, I make a dramatic change in my diet, the relationship between training volume and weight will no longer be the same and will need to be recalculated.
7. Training doesn’t guarantee a race result: It is certainly a prerequisite for one, but anything can happen on the day!

As I said above, this has all been calculated around me. If anyone else would like to give it a go, please contact me and I can help you go through the process, share templates I have created etc. I would be fascinated to see whether this model holds across runners!

But with its limitations in mind, and recognising that it is still a work in progress, I am excited by its potential and the fact that it gives me a real time view of how my training is going, and what I should expect out of it: In the past I would stress about missing a week or so, knowing that it would have some impact on my objectives, but not knowing how big. Now I can see exactly what that impact is (depending on what week of the plan you are in), what impact it might have on my expected race time, and whether any extra training I have put in may have mitigated it.

And the Sisyphus in me feels that this has allowed him to map the path up the hill to fitness, and may help him get the rock all the way to the top this time!