Wednesday, May 24, 2017

A Powerlifter's Guide to VBT Pt 7: RPE to VBT translator

This entry was meant to be posted quite a bit earlier, but I delayed it. In my part of rural Texas, internet service providers crash with tiny weather upset. I was not about to type this shit out on mobile just to get it done in time.


-RPE and VBT are not competing systems.

-Which system you prefer is entirely dependent on many variables. The most important one is what type of feedback you find valuable: internal subjective feedback or external objective feedback.

-The science shows strong correlations for both, but there's some issue with the noise associated with velocity.

-The exertion-velocity relationship can explain the RPE rating system in velocity measurements, helping bridge the two worlds. Or if you're lazy, and accuracy be damned, there's an equation to level the two.

-Even if you don't care much for bridging the gap between RPE and VBT, doing the exertion-velocity relationship can open a world of programming and planning strategies. 

-Why VBT=awesome-sauce and RPE=face palm. We are competing, VBT is winning, and RPE can suck it.


Most people that tend to be into Velocity Based Training (VBT) are also actively using/interested in Rating of Perceived Exertion (RPE). RPE was popularized by Mike Tuchscherer from Reactive Training Systems. Others have outlined the advantages of RPE vs VBT vs other autoregulation systems better than I have, so I'm not going to attempt to do that. Despite how some tend to put it, VBT and RPE are not mutually exclusive practices, can exist in both harmony and in conflict to one another, and each has key takeaways to aid the other. There are caveats to using both systems.

If anyone thinks VBT is innately juxtaposed against RPE, I steal most of my new ideas from RPE and try to find a way to make them work in VBT. In fact, very, very few of the ideas I've expressed so far are original. I cite most of the ideas I'm stealing. So far, it seems like few are attempting to bridge RPE and VBT. Many tend to favor one method and use the other as a footnote barometer. I tend to favor using them on an "it depends" basis. 


Looking back at my previous writing, I have seriously taken for granted that people understand RPE. Let's alleviate that by inadequately explaining it. RPE attempts to quantify subjective feeling. Here's the infographic:
A more thorough breakdown can be found here. That said, I haven't taken the time to compare and contrast Mike T's current RPE methods against his old ones, The Strength Athlete, or anyone else's. 

There's also Reps in Reserve, which operates more like an abbreviated RPE system with no half steps. I associate this more with Eric Helms' and Zourdos' work, but you could attribute both of these methods to APRE, Borg, or Confucius. You could even argue that Renaissance Periodization/Juggernaut uses an RPE method when saying 3/fail, 2/fail, 1/fail (1-3 reps away from failure). 
The common argument for RPE/RIR is that everyone instinctually uses RPE without realizing it. Everyone uses how they feel to autoregulate their training. This can involving decreasing load when they're having an "off day" or other methods of adjusting training according to perception of readiness. While they're not wrong, that's also like saying just because I pay Social Security that I'm pretty much a Communist. That argument functions as a pidgeon-holing effort that I disagree with. I see the purpose of the argument though. 
I could equally argue that since RPE proponents consider perceived and sometimes measured bar speed, they are closet VBT fetishists. I'm not claiming that.
If I were to inappropriately simplify RPE, it's something like this: say a lifter completes an 8 rep max for 6 reps, they just did 6 @8 RPE or 6 with 2 RIR (reps in reserve). I will switch between all three of those interchangeably because I'm pretty bad at self-editing. 


Obviously, VBT is king. VBT is the absolute king of ding-a-lings. No question. It's science.

Not really, but it depends. Let's start conceptually or by "attitude." If you've been using RPE for an extended period of time, then obviously RPE is better for you, at least short term. The opposite is probably true for VBT. You could argue that VBT relies on reference velocities often in the form of tables, but you could also argue that RPE relies on tables. Both of them rely on previous experience. If you introduce a novel movement someone isn't familiar with, their referent minimum velocity threshold (velocity of last rep in the tank) or RPE rating might be meaningless. 

VBT is external and objective. RPE/RIR are internal and subjective. They're not really opposites since they both serve the same function, but in terms of how the feedback is garnered, they are opposites. Different people will find a meaningful use for both/either.

If you're bad at gauging internal, subjective intensity, then RPE won't work for you. It can be frustrating to get ahold of and in the moment if feels like your navigating with a broken compass. If that sounds familiar, then troubleshoot the situation or move on. Maybe something more objectively based is more appropriate. VBT and AMRAP sets become better options in that case. Each of those has drawbacks though. I'm not going to cover AMRAPs as an autoregulation method much - but GZCL and 5/3/1 are good examples of that.

So how could external, objective feedback be worse than something like RPE? This probably would serve better with illustration. Here's an as-many-reps-as-possible set (AMRAP):

The first rep looks like an honest attempt. I would argue this rep could be a good indicator of intensity - around 80-85% of that day's 1RM. The last rep is equally a good read, and would be a good indicator of minimum velocity threshold (MVT). Again, MVT has been a good indicator of the last rep in the tank, velocity at 1RM, or what we could call RPE 10. 

The issue is reps 3 to 5. Rep 3 takes a massive nose dive, rep 4 attempts to recover that velocity, but it isn't really regained until rep 5. At rep 5, it seems to take the same incremental loss in velocity it had in reps 1-2. There's a certain amount of noise to every measurement. The problem leveled against VBT is often that it is noisy, that its measurement varies too widely from rep to rep. We try to reduce this by choosing appropriate devices (as we've shown, with the device comparisons). You'll subconsciously try to do this if you rely solely on VBT. You'll start forcing yourself to perform sets in a way that data acquisition is more optimal - often in ways conducive to the training process, but possibly in ways that hinder the training process. The upside of VBT is you often have to think less to augment your training. If your measurements are noisy though, you have to put on your coaching cap more often and determine what's a real measurement and what's an artifact of performance. Sometimes you want to work out and not pretend to be a half-decent data scientist.


RPE is king of ding-a-lings. VBT=Communism. 

Not really, but with the limited research we have comparing the two, RPE comes out on top in a few respects. Firstly, velocity is not as good at predicting 1RM based on the levels of confidence. In the tables I've used, there's a point estimate as well as the upper and lower ranges of the estimate based on 95% and 90% confidence intervals. These intervals can be fairly wide, so the point estimate should be taken with a grain of salt.

VBT needs to be individualized and varies by exercise. The general guideline is that 0.30 m/s is MVT (1RM, 10RPE, 0 RIR) velocity for normies in the squat, and 0.15 m/s in the bench and deadlift. Zourdos et al found that experienced squatters hit MVT at a slower velocity than noobs (0.24 m/s vs 0.34). Similarly, Helms et al found the same trend, with squats clocking in at 0.23 m/s, bench at 0.10 m/s, and deadlift at 0.14 m/s. But when it comes down to the nitty-gritty, here's what that looks like in the grand context:

A value of 1 or -1 is a perfect relationship. 

You have to think of this in the context of their use though. It's far easier to gauge a 1RM attempt as a 10RPE than it is completing a 10RM for 7 reps and identifying it as a 7RPE. Much the same, it's easier to determine reps in reserve closer to failure with velocity as the referent measure than it is further from failure. For intermediate lifters, both systems tend to excel closer to failure. Both studies seem to demonstrate the relationships of velocity and RPE at 1RM, but the science is chugging along for loads short of 1RM and the information is still forthcoming. Additionally, there is something to be said about framing a user's understanding in the context of VBT. If you were to anchor velocity training in the same context that you anchor RPE, and furthermore allowed a good period of evaluation of both systems, you would likely see some changes in the results. Maybe even VBT comes out worse, or maybe that RPE is strongly associated with Sovereign wing nuts.

Anchoring is essentially how you "build" your personal RPE tables. With a 1RM and an AMRAP at a given percentage (let's say 85%), you can split the difference and generate a table of best-guesses of %1RM for every given prescription of number of reps and the corresponding RPE. See the link provided earlier in the article.

Much like you really have to force yourself to be honest when using RPE, you have to move with intention with VBT - or attempt to move every rep as fast as possible while maintaining technique. At this point, you could circle back to the common assertion by both parties, that either system performs well enough, requires at least intermediate experience to gain full utility, and either could be more appropriate than the other for different individuals.

Future directions for this could be studies showing whether or not one could accurately predict 1RM based on sub-RPE10 ratings for multiple reps. Likewise, the same sort of study could be used in the VBT realm, showing whether you can predict daily 1RM based on regular fluctuations in velocity at a standardized load. Chris Duffin mentions this in a YouTube round-table about VBT and his experience using OpenBarbell before a meet. Additionally, there is no science in the works to determine reps in reserve to velocity, something termed the exertion-velocity relationship.


AMRAPs are great at getting a submaximal measure of MVT. If you run a VBT-1RM estimation without a known MVT, you're guessing. There's two ways to get that: do an AMRAP or do a 1RM test. Obviously, the second option renders the estimation useless. Also, it shouldn't be surprising if your estimate is off. 

AMRAPs are also great to determine exertion. Exertion could very simply be explained as the proximity to failure. Assuming a MVT of 0.11 m/s, a rep at 0.19 m/s requires more exertion than a rep at 0.26 m/s. But can we take that seriously when we stop short of failure? Especially with how much noise can be in that measurement? Maybe.
Two AMRAPs pictured left, then the velocities averaged across according to the reps in reserve. Notice the change in R2

Here's two sets according to their reps in reserve. One was a 4RM, another was a 6RM. Because this is ordered by RIR, it was essentially performed in reverse (RIR 5 in the 1st graph was the first rep, RIR 3 in the 2nd graph was the first rep). You can see by the data points that there is some noise. According to the regression, the first AMRAP has a higher coefficient of determination at 0.89 than the second at 0.79. A perfect fit would be 1. When we average across, that coefficient of determination kicks up to 0.93. 

A recent exertion-velocity table that helps demonstrate
AMRAP utility in VBT
The above table shows an example of how we can smooth the noise inherent in VBT across a set. Bring your attention to AMRAP set 2, particularly in the middle. The last rep (0.25 m/s) is an honest attempt, and you can see that the first can be taken seriously to (0.36 m/s). The middle two make no sense. However, when we average two sets across, we can smooth that noise out. If we plot this averaged mean velocity across two AMRAPs and correlate it to RIR, we can interpolate how many reps we have left in the tank based on final velocity. The generalized midpoints between each rep become half-steps. A half-step might be noise, so maybe it's more useful to round it, but it could be interpreted as RPE.

Another graph of successive rep velocity decrements
from a published article, not quite as noisy as mine. 

I keep talking about noise. So what's the smallest worthwhile difference to figure out noise? It likely depends person to person. I would guesstimate it's something in the neighborhood of 5% of velocity, generally speaking, but I generally operate off of +/- 0.03 m/s. I came to this conclusion based off of individual experience, so don't hold it against me too seriously. It's a ballpark.
You'd be right to point out that there is some noise in RIR 1-3. The incremental drops in velocity aren't that large, and therefore are questionable. By using the regression, we can use the slope to get an optimal read on exertion. Much like RPE though, you'll have more reliable measures the closer you get to 10 RPE or 0 RIR. Much like we don't worry about anything below 6 RPE to begin with, you do see noise in velocity at 6 RPE or lower. 
If you were to run this against a regression and estimate RPE according to velocities, you sort of force a fit, but it does ball park things for you. The main issue is when you're going for large sets in excess of 10. Larger sets are going to be more noise than they are data, so I try to keep the nRM attempts under 10 reps. It's worth pointing out that half steps in RPE might be truncated due to the size of the incremental velocity losses. Guess what fills in the gaps in this case? Subjective, internal feedback. Just like you can use VBT to spotcheck RPE, you can do the opposite.
Sounds like a bunch of hand-wavy mathemagic. Possibly. If you ask me for a peer-reviewed article that supports this assertion and I can't help you. The best I do is refer you to people that are more qualified than me. To be fair, in a previous article I did mention this as an outstanding issue that researchers need to resolve. 


I've tried to do my due diligence here. A quick search through RTS's forums will find mostly inquiries about how devices compare. One reply in particular from November of 2014 has the following equation from Mike T:
This was back in 2014 though and I'm not sure if Mike has since disregarded this. I also can't verify whether or not this matches up with my experience because I haven't used it extensively.

[Edit 8/8/2017: It would appear from an article in July from IJSSP that generalized equations don't work, however this could be due to the fact they didn't account for MVT. In a follow up article, I advocate using initial velocity to determine XRM and regulating exertion using that relationship. This appears to be back by the research in that same article.

Edit 9/03/2017: According to unpublished work by Garcia-Ramos, a single repetition to gauge daily 1RM tends to overestimate. A better method is to use 2 points to generate the estimate. That's sort of what many practitioners did anyways without so much evidence and with a bit more intuition. TBH, VBT is just gonna be one of those things where the practitioners are going to be one step ahead of the science at all times.]


Reviewing what we've done so far leading up to this article... that's a load-velocity test (which is just a fancy data-geek warm up) and two AMRAPs. The load-velocity (L-V) table becomes a %1RM-velocity (%1RM-V) relationship. The two AMRAPs give you MVT, turning to L-V into a 1RM estimate. Then the two AMRAPs averaged across gives you the exertion-velocity (E-V) relationship. This E-V relationship is transformed just like the previous relationships to give us the RPE-velocity relationship (RPE-V). Essentially three rounds of data collection to give us five different relationships. And you have to do that for every lift you plan to autoregulate. 

So if you have 15 movements to be used in a macrocycle, that's 15 movements to map out. That initial investment of time is pretty large though. In theory, you should have your competition movements repeat every macrocycle so the data could roll over with very little update. So that's 12 movements, which is still an incredible amount of investment. This is also why I introduce movements into follow on macrocycles slowly. Rather than having only 3 movements carry over, I usually have three-quarters of my movements from my last macrocycle repeat into the next one, so roughly just 4 movements. Exercise selection is a whole different argument, and I'm neither a huge believer in variety for variety's sake or hyper-specific training. Or you could use deload weeks between cycles much like I do and make them testing days since the volume is fairly low. You could cap intensity for load-velocity testing at 80%, and I doubt a single at 80% is going to be a huge hindrance to your recovery.

But really, one of the great advantages with this is you may only need this one relationship, depending on how you use it. Using just E-V, you can set a rep goal, like 8 reps with 2 RIR (or /2 fail, or a 10RM for 8 reps, or 8 RPE - pick your semantics). You can then get within the 2 reps based on the linear relationship between E-V and RIR. For heavier lifts where fewer reps are possible, it's even easier to determine what the opening velocity should be. But now it just sounds like you're doing exactly what RPE advocates always say: base your autoregulation off RPE, but adjust with external feedback like VBT.

If you're going for straight up training wheels to get you from VBT to RPE, this is probably the way to go. If you also wanted to run an RPE program as a VBT ideologue, you could transliterate the E-V data into RPE. 


Previously I mentioned using INOL to gauge the number of reps per lift in a session. This was generally my attempt to control for training stimulus regardless of intensity used. While I think INOL is quite effective in regulating intensity and volume in proportion, they really don't say much towards your proximity to failure. Three sets of five reps at 60% is regarded the same as one set of fifteen reps. Because of this, it's obvious that INOL somewhat ignores proximity to failure, or exertion.

New reader, what this? INOL attempts to gauge training stress/stimulus through an equation. It takes into account the number of reps and intensity of the lift. This gives you an arbitrary number that you can use to determine if a "volume day" is as taxing as a "heavy day." This is covered and linked to in previous articles.

Furthermore, the blending of INOL and VBT is somewhat dubious at best. INOL is more theory than substance, and any modicum of its substance is relative to percent based training, not VBT per se. This detail may be beside the point, especially if your autoregulation isn't changing load by more than 7.5%. To confound the issue even more, INOL is supposed to be a planning tool, not an autoregulation tool. Maybe you can figure out some sort of system with hand-wavey reverse look ups, but maybe you're over-complicating it. 

Regardless of intensity, I see additive value in regarding each set in the perspective of proximity to failure. RPE does exactly that, where 1x @8 RPE is essentially one rep with two left in the tank, or more concisely a 3RM done for 1 rep. Why does a rating system matter though? Why does it matter to talk in "proximity to failure" terms or RPE terms?

Mike T's recent summit at powerlifting university 2017 shined some light on this that made me take it more seriously. Much like INOL, Mike had created a system of coefficients that weighted the training stress of each set. For example, @9-10 RPE (no reps left in the tank) was 1.33, @8-9 was 1, @7-8 was 0.8, and so on. You can add all weights across a movement within a session or within a week and compare it to some ballpark guidelines. Whereas INOL weights reps according to intensity regardless of proximity to failure, Mike T's training stress index weights exertion by sets. Granted, if you're measuring by a different unit, you should create bounds for those units.

For a single movement, Mike put forth 2.5 for an easy session, 3.5 for a moderate session, and 4.5 for a hard session. Much like INOL, it operates under the "single movement" clause, whereby the training stress of bench vs that of your squat aren't lumped together. It's not clear to me whether something like touch and go bench and feet up bench are arbitrarily separated, but you could argue it either way. For a training week, good planned bounds would be 14, 20, and 26, for the same rating scheme (easy, moderate, hard).

As I said before, this is all in reference to Mike's seminar on 2017 powerlifting university summit. I'd use the pictures or talk more in depth about it, but given that it was paid-for material and it's not my idea to share, I'd rather just give you the jist and refer you to buy access.

Unlike session rating though, this is separated by movement pattern with all squat type movements lumped together, etc. I may have misunderstood this, to be fair. Much like the INOL theory though, these guidelines could vary from individual to individual. This makes sense though given it's a proxy for work capacity, and work capacity is more of a moving target than it is a fixed point.

I like this application because everyone talks about work capacity when there's seems to be little agreed upon measure, normative values, etc. Work capacity is essentially a working theory. Mike Israetel talks about maximum recoverable volume (MRV) and maximum adaptable volume (MAV), which sounds much like work capacity as well. Mike T's training stress index also meshes pretty well with MRV terms of understanding how we manage lifting schemes.


Another good reason to consider RPE methods of the E-V table is the inherent fallacy of velocity loss. I've said before that velocity loss can be misleading. At very high percentages of 1RM, the threshold of 20% velocity loss (or 40% even) is below your minimum velocity threshold, or your last rep in the tank. Following the velocity loss rules to a T guarantees one of two things: 1) blunt force adaptation or 2) getting stapled.

Velocity loss is a VBT concept that strength adaptations occur from training where only 20% velocity is lost. Training with a loss of 40% of opening velocity is better from hypertrophy. It's sort of a minomer since you can perform both 85% of your 1RM to a quarter of velocity loss or 60% of 1RM to a quarter of velocity loss - with one of those better contributing to strength development. What is really hard to do is to get 20% velocity loss at sets at 90% and above - even if you can bang out 5 reps at 90%.

Even if you think E-V tables are too much additional work (when really they can be the only VBT data you record, ignoring all others), you still have to remember your MVT for these situations and have a general sense of how much velocity you usually bleed off in your final reps. If your final rep or 1RM moves at 0.24 m/s and you hit a rep for 0.27 m/s, there might not be another rep in the tank for that set. The next velocity decrement could be below your ability to power through. Alternatively, 0.27 m/s could just be noise - and you should plan around that. The only method you have to rebut this without this information is by going by subjective opinions on whether you can complete the next rep - which sounds exactly like what RPE is trying to accomplish. 

Furthermore, E-V tables allow another programming strategy that ignores percent of 1RM. This might be useful for movements that have no real 1RM as well - like your rear foot elevated split squat or other supporting exercises. 1RM for those lifts may never be tested, may be implausible/unreliable to test, or pointless. Rather than using percentages, you can undulate by number of repetitions if you know your E-V. If you plan to hit sets of 8 with 2 reps before failure, you can use E-V to determine appropriate starting velocity.

Or you can do what I do: not really care about the autoregulation, hit a fixed number of reps and sets, and make sure it's heavy enough to be hard. It's an accessory - not existentialist theory. Just do the damn work and be done with it.

While I've so far been an advocate of autoregulated volume and intensity, you can use this to truncate your autoregulated variables or control for number of lifts (NL). 


I hope after reading through this you've understood that VBT is king of all of the dingle-dangles. Not really, but since this is a VBT-Powerlifting blog, it's sort of becoming of me to suggest it.

Others have shown a strong inverse relationship with velocity and RPE. RPE and AMRAPs are the most common autoregulation techniques. VBT can be used to augment RPE, and is most commonly used that way. Likewise, AMRAPs can be used in conjunction with VBT as well. These three methods tie together on multiple levels, providing more forms of feedback. Brandon Senn explains this process really well in his article about the Autoregulation Book of Methods.

Which system or combination of systems is best is really user dependent. Even a staunch VBT advocate should be able to see some redeeming qualities in RPE. More importantly, autoregulation training methods are more similar than they are dissimilar in their application. Even if it's not your preference to use one system to drive the training process, another system will help analyze the process. The very obvious advantage that RPE or AMRAPs have over VBT is the bar for entry. Neither have price tag implicit to the process. I still do not feel required to win you over, but I hope I've alienated and emasculated all sides in this obvious shit post.