Key Points:
On average, most people are accurate at predicting repetitions in reserve (RIR), usually being off by ~1 repetition.
RIR accuracy is usually higher during lower repetitions sets (≤12 reps) and when prediction is performed later in a set.
Training status or sex do not seem to influence RIR prediction accuracy.
Interestingly, coaches estimating RIR via video footage were also off in their predictions by roughly one rep, with their predictions also being more accurate when performed closer to the end of any given set.
Introduction
Proximity to failure can influence both strength and hypertrophy gains, and if you are consistently underpredicting or overpredicting how many repetitions you have in reserve on a given set, you may be doing your gains a slight disservice. In the last decade or so, the concept of autoregulation and using repetitions in reserve (RIR) to guide strength and hypertrophy programming has arguably overtaken the more traditional programming approach of fixed one-repetition maximum (1RM) recommendations (eg: 80% of 1RM for 3 sets of 5 reps).
However, training by using a RIR target (ie: a specific proximity to momentary failure) heavily relies on, you guessed it, the ability to accurately predict RIR.
But before we look at whether we’re actually good at predicting RIR and how we can perfect our ability to accurately predict RIR, let’s take a trip down “iron” memory lane.
The year is circa 2013. Programming for strength and hypertrophy is mostly centered around calculations based on one’s 1RM, the infamous Prilepin’s table, and sometimes simply on “traditional” repetition ranges without any additional guidance on proximity to failure. If the program said 3 sets of 3 reps at 80% of 1RM, and that 80% of 1RM happened to be a load that required you to absolutely grind out your sets as if your life depended on it, then that was unfortunately what you were required to do (at least on paper). Similarly, classic programs like Stronglifts 5×5 rarely made reference to any guidance regarding proximity to failure, simply instructing lifters to add weight on the bar every week, and sometimes recommending that some lifters should “start light” during the initial weeks of their training. On the hypertrophy end, things were often even more vague, with set/rep prescriptions often coming with no explicit direction regarding proximity to failure, usually because it was assumed that all sets should be “hard” or “to failure.”
Additionally, the lack of a consensus on the relationship between proximity to failure and strength/hypertrophy gains made things even more confusing. On one hand, you had top-level athletes advocating for going near or to failure multiple times, while others preached staying away from failure despite aiming to maximize adaptations.
Regardless of which school of training philosophy you adhered to, you either had to follow a program that instructed you to lift specific loads regardless of how you felt, or follow a program that gave you rather vague guidance along the lines of “train hard and make sure you get close or to failure” without necessarily having a way to quantify your effort.
When it came to alternative tools for quantifying your effort in resistance training, looking at the scientific literature was also unhelpful. Most resistance training studies either used a fixed %1RM load or instructed participants to perform repetitions to various forms of failure (volitional failure, momentary failure, etc).
Outside of the gym, though, one tool had started to garner attention, particularly in endurance training. That tool was the Borg Rating of Perceived Exertion (RPE) scale, which aimed to quantify the subjective experience of physical effort, pain, and fatigue during exercise.
Borg’s scale was based on the premise that individuals are able to introspectively evaluate their physical state during exercise and that these evaluations can be consistently mapped onto a numerical scale. This allowed both for self-regulation of exercise intensity and for communication of that intensity to others, such as coaches or healthcare providers.
The original Borg scale introduced in 1962 ranged from 6 to 20, where 6 meant “no exertion” and 20 meant “maximal exertion.” The numbers were chosen to roughly correspond with the heart rate of a healthy adult: a rating of 6 corresponds to a heart rate of about 60 beats per minute (resting heart rate for many people) and a rating of 20 corresponds to 200 beats per minute (the maximal heart rate of a young adult). In 1982, Borg introduced a revised scale known as the Borg CR10 Scale, or the Borg Category-Ratio Scale. The CR10 Scale went from 0 (“nothing at all”) to 10 (“extremely strong”), including verbal anchors at each level (e.g., moderate, strong, very strong) to help users better gauge their level of exertion.
Although the traditional Borg RPE scale can be used to assess perceived effort during resistance training, the endurance exercise origins of the scale make it somewhat inappropriate for accurately gauging/guiding proximity to failure, especially during scenarios where increased discomfort (e.g., during high-repetition sets) may lead to high ratings of perceived exertion. A set of 20 repetitions on the leg extension may feel like an 8/10 on the CR10 scale, but that rating may persist for an additional 5-10 repetitions without really allowing the lifter or person observing them (such as a coach or researcher) to really know whether the lifter is close to failure or not.
Emergence of Autoregulation and the RIR-based RPE Scale
In 2016, Zourdos et al were among the first to explore the use of a “novel resistance training-specific rating of perceived exertion scale measuring RIR” in lifters. More specifically, Zourdos et al explored the relationship between the rating of perceived exertion specifically measuring RIR and various intensities of 1RM in both experienced and novice squatters. The protocol included performing a 1RM squat followed by single repetitions at 60%, 75%, and 90% of 1RM, and an 8-repetition set at 70% 1RM with average velocity recorded for these lifts. RPE values corresponding to RIR were reported after each set.
The study found a strong inverse relationship between the RPE values (indicating the lifter’s perceived exertion and estimated repetitions left in the tank) and the actual velocity of the lift across all intensities. This relationship was observed in both experienced and novice squatters, suggesting that as lifters approach their maximal effort (higher RPE), the speed of the lift decreases, indicating fewer RIR. The study detailed RPE values at various intensities (100%, 90%, 75%, 60% of 1RM), showing how RPE tends to increase with intensity in both groups, with experienced squatters generally reporting higher RPEs. Additionally, the study found significant differences in how experienced versus novice lifters perceived their exertion and estimated RIR at these intensities. Notably, experienced lifters reported higher RPEs at maximal lifts, which might indicate a more accurate assessment of their RIR due to their greater familiarity with high-intensity efforts.
The results of the Zourdos et al study demonstrated the use of an RIR-based RPE scale could be a viable method to not only quantify effort during resistance training but also to regulate training load in real-time, offering a practical way to adjust intensity based on the lifter’s perceived capacity to perform additional repetitions.
This study was also among the first to introduce the concept of autoregulation in the context of lifting programming. As expressed by Dr. Eric Helms in the “The Science of Autoregulation” SBS article, “autoregulation, simply put, is just a structured approach for embedding a respect for individual variation within a program”. Autoregulation allows one to adjust the intensity, volume, or other training variables based on recovery state and overall readiness on a day-to-day basis. Unlike traditional training programs that predetermine load for a given exercise, autoregulation recognizes the variability in an athlete’s daily readiness-to-perform due to factors like sleep quality, nutritional status, stress levels, and residual fatigue from previous workouts. In addition to guiding appropriate load selection, autoregulation can also allow for better fatigue management, as closer proximities to failure may lead to greater neuromuscular fatigue and increase recovery time. A recently pre-printed study found that during the course of an eight-week training study, there were no differences for both subjective and objective markers of fatigue between groups training, on average, close to failure (1-3 RIR) or far away from failure (4-6 RIR).
The concept of utilizing the RPE scale based on RIR to guide programming went hand-in-hand with the concept of autoregulation becoming more mainstream and adopted by competitive and recreational lifters worldwide. More specifically, outside of the literature, the RPE scale based on RIR and the concept of autoregulation were popularized by the powerlifting coach Mike Tuchscherer, who is often credited as a significant contributor to the development of the scale itself. Fast forward to today, and the RPE scale based on RIR is among the most widely used tools in the world of strength and hypertrophy, with coaches of all levels sometimes basing their programming solely on either the RIR-based RPE scale or RIR targets and embracing the flexibility of autoregulation.
Given the popularity of autoregulation, some of the questions that then naturally arise are:
How accurate are we at predicting RIR?
What influences our ability to predict RIR?
How can we improve our ability to predict RIR?
Can others predict RIR for us?
Let’s take a closer look at the literature in an attempt to answer each one of the above!
Current evidence on RIR prediction accuracy
When trying to understand whether individuals are able to accurately predict RIR, a recent scoping review and exploratory meta-analysis by Halperin et al is the most comprehensive analysis of the topic. The study aimed to examine the accuracy of predicting repetitions to task failure in resistance exercise by looking at studies with healthy participants who predicted the number of repetitions they could complete to task failure in various resistance exercises before or during an ongoing set performed to task failure. Overall, the authors included 13 publications covering 12 studies with a total of 414 participants.
The findings revealed that participants generally underpredicted the number of repetitions to task failure by approximately one rep on average (0.95 reps to be exact), indicating a tendency toward underestimation. In other words, when someone thinks they only have two reps left in the tank, they likely still have about three reps in the tank, on average. However, prediction accuracy slightly improved when the predictions were made closer to set failure and when the number of repetitions performed to task failure was lower (≤12 repetitions). Interestingly, the participants’ experience, whether the exercise was an upper or lower body exercise, and the number of sets performed did not significantly influence prediction accuracy. Overall, there was minimal variation in predictive accuracy among participants (with a standard deviation of 1.45 repetitions), suggesting the primary source of error was systematic underprediction. The results of this review show that – although imperfect – most individuals seem to be relatively accurate with their ability to predict RIR, at least in an environment where they are being observed and are aware that they will be working to failure and will need to predict RIR.
One of the study’s limitations is that the designs of the studies included were mostly acute in nature and may not necessarily reflect an individual’s ability to predict RIR in the long term. Other limitations include the potential biases introduced by the variability in task failure definitions and prediction timing across included studies, as well as the potential for anchoring bias, where participants may have unconsciously limited their effort to their predicted repetitions.
Although the above limitations warrant caution when interpreting the results of the study, the following are some of the practical takeaways:
When instructed to gauge their proximity to failure, most people seem to underpredict their RIR by roughly one repetition.
RIR prediction accuracy may not necessarily improve with training status but may improve when performing less than 12 repetitions per set.
When predicting RIR, it’s probably best to do so as a set progresses versus trying to determine your RIR at the start of the set.
Since the Halperin et al review, more data has come out on RIR prediction accuracy both in untrained and trained individuals.
The first study to be published after the Halperin et al review was a study by Remmert et al that investigated the accuracy of RIR predictions on single-joint and multi-joint exercises at various proximities to failure. More specifically, the study involved 58 participants who performed four sets to failure of three exercises (cable biceps curl, cable triceps extension, and seated cable row) at 72.5% of their estimated 1RM. Participants then indicated their perceived RIR at various points during each set until reaching failure, which essentially allowed the researchers to measure the difference between predicted and actual repetitions until failure.
Similarly to the Halperin review, the study found that RIR predictions were more accurate when made closer to failure and improved in accuracy from one set to the next. At 5 RIR, the mean RIR difference was 1.2, and it reduced to 0.464 at 1 RIR. The mean RIR difference for set one was 0.955, which was significantly higher compared to set three where the mean RIR difference dropped to 0.706. Overall, though, the participants were again roughly one rep off from their actual RIR.
Additionally, factors such as sex, training experience, and prior RIR rating experience did not significantly influence the accuracy of RIR predictions.
The same group of researchers published another study on predicting RIR, this time explicitly looking at trained men. The study aimed to evaluate whether the accuracy of intraset RIR predictions in bench press exercises would improve over a six-week training program. The study involved nine trained men who participated in three bench press training sessions per week for six weeks after a one-week familiarization phase. In each session’s final set, participants noted when they thought they had four reps in reserve, and one rep in reserve, before continuing the set to momentary muscular failure. Just like with the previous Remmert et al study, the differences between predicted and actual RIR were recorded to assess prediction accuracy.
On average, subjects misestimated their RIR by about 1.1 reps when they thought they were 4 reps from failure, and by about 0.7 reps when they thought they were 1 rep from failure. The absolute value of RIR difference showed no significant changes over time, further suggesting that the overall accuracy of RIR predictions remained stable throughout the training period with an estimated marginal slope close to zero.
Furthermore, the study revealed that the number of repetitions performed had a significant effect on the accuracy of RIR predictions. Specifically, for every additional repetition performed, the raw RIR difference decreased by about 0.404 repetitions, demonstrating that participants were more accurate in their predictions as they got closer to the actual end of the set.
Another recently published study by Refalo et al sought to assess the accuracy of intraset RIR predictions in resistance-trained individuals, specifically during the bench press. The study involved 24 resistance-trained subjects (12 males and 12 females) who participated in two experimental sessions which were conducted roughly 48 hours apart. During these sessions, participants performed two sets of barbell bench press at 75% of their 1RM until momentary failure. They were asked to predict when they were at 3RIR and 1RIR, and their accuracy was then assessed by comparing the predicted RIR to the actual repetitions performed before reaching failure. Overall, participants were generally accurate in their RIR predictions, with mean absolute RIR difference accuracy of 0.65 ± 0.78 repetitions. No significant differences were noted between the three and one RIR predictions. Additionally, there were no significant differences in RIR accuracy based on gender or resistance training experience.
Interestingly, the Remmert et al studies and the Refalo et al study align with the findings of the Halperin et al review. Namely:
Most people are relatively accurate at predicting RIR, regardless of sex or training experience.
On average, most people may be off in their RIR predictions by approximately one repetition.
RIR prediction accuracy improves when performed later in a set and during lower repetition sets (≤12 repetitions).
As a bonus note, studies on powerlifters where they guided their load selection solely based on the RPE scale based on RIR have also found that powerlifters were probably relatively accurate at predicting RIR. I say “probably” as those studies did not actually assess RIR prediction accuracy but instead instructed participants to perform “daily max” single repetitions on the squat, bench press, and deadlift at an RPE of 9-9.5, meaning at an RIR of 0-1. In addition to the participants of the one study reporting an average RPE of 8.9-9.1, their peri-training single-repetition loads were either slightly below, the same, or slightly above their pre-intervention 1RM values, meaning that they were probably pretty close to a “daily max” 9-9.5 RPE single as instructed. Additionally, a classic study by Helms et al explored the relationship between average concentric velocity and RPE based on RIR across three powerlifting movements: the squat, bench press, and deadlift. The researchers sought to determine how these metrics correlate and how they could inform the prescription of exercise intensity in training regimes for powerlifters. The study involved 15 powerlifters who performed a 1RM for each lift and reported their RPE for all sets. Average concentric velocity was recorded for all attempts performed at 80% of estimated 1RM and above. The results showed very strong relationships between the percentage of 1RM and RPE for each lift, with correlation coefficients between 0.88 and 0.91, indicating that lifters’ perceptions of exertion closely matched the actual “intensity” of the lifts. There were also strong to very strong inverse relationships between average concentric velocity and RPE, indicating that as RPE scores increased, the speed of the lifts decreased.
Overall, it does indeed seem like both untrained and trained participants are relatively accurate at predicting RIR, with some relatively minor “terms and conditions” when it comes to improving their accuracy. It’s also important to note that in studies, participants perform exercises to failure and are actively monitored, gaining both a clearer sense of what training to failure really feels like, in addition to receiving feedback on their prediction accuracy. However, in everyday practice, especially for those who have never pushed a particular lift to failure, their sense of what 1 or 2 RIR feels like may not be as precise. Additionally, without a researcher monitoring their performance, a trainee who is self-noting their RIR on their training log may never realize that they are inaccurate in their RIR predictions. While lifters are generally accurate at gauging RIR, accuracy may be a bit lower in real-world contexts (i.e. lifting on your own, in the gym) than it appears to be from the literature. Despite the above, it’s not uncommon to think that we often see people overshoot or undershoot their sets regardless of repetition range, lift, or prescribed RIR. This may come down to availability bias –the cognitive bias that leads to people overestimating the likelihood of an event based on how easily examples come to mind. It may be that it’s easy to remember seeing posts on social media where people joke about overestimating their RIR or coaches complaining about trainees misjudging their RIR versus remembering somebody “just training.” Additionally, although we may think that someone is not accurately predicting their RIR, we did not have any direct evidence looking at whether our predictions as observers are likely to be right or wrong until recently. As a coach, although I will often assume that clients may be overshooting or undershooting some of their sets, I know that the only way for me to see if I was right is to actually have the client predict their RIR at various points of their sets while taking that set to failure, much like many of the studies above.
Interestingly, Emanuel et al recently looked at assessing coaches’ prediction of RIR. The study aimed to assess the accuracy with which coaches can predict the RIR a trainee has before reaching task failure during resistance training. The study involved 259 certified resistance training coaches who watched videos of trainees performing barbell squats and preacher biceps curls at either 70% or 80% of their 1RM until task failure and made RIR predictions at 33%, 66%, and 90% of the set’s completion. This design pretty much mimics how many online coaches assess the performance of trainees these days (i.e., via online video assessment).
Similarly to what we saw with trainees and predicting RIR, the coaches often underpredicted the RIR early in the sets but became more accurate or slightly overpredicted as the sets neared completion. Specifically, the average absolute prediction errors were 4.8 repetitions at the 33% point of the sets, 2.0 at 66%, and 1.2 at 90%, with accuracy improving significantly as sets neared failure. The analysis also showed that coaches were more accurate in predicting RIR for preacher curls compared to squats, and they performed better in sets with heavier loads. Interestingly, and again much like the research on trainees, the experience level of the coaches had a negligible impact on the accuracy of their predictions. The study noted that coaches’ prediction accuracy improved during subsequent sets of an exercise, suggesting a learning effect as they became more familiar with a given trainee’s performance capabilities during the session.
The authors noted that the use of video-based observation was one of the study’s limitations, highlighting that it might not be capable of capturing the full dynamics of an in-person training environment. In addition, they noted that the limited exposure to each trainee’s performance does not fully replicate the typical ongoing relationship between a coach and trainee in regular training sessions. However, I’d argue that although the above limitation is true for in-person coaching, it does not necessarily apply to online coaching. The design of the study actually has a relatively high level of ecological validity to how RIR is often judged by online coaches (ie: via video footage).
Can You Avoid Having to Estimate RIR?
Although the ability to predict RIR is important for the majority of trainees, it may not be necessary for people who are strictly training for hypertrophy and enjoy mostly training to failure. If you fall in that category, there is really no need to worry too much about your ability to predict RIR since the majority, if not all, of your sets will be taken to failure. However, asking yourself “how many reps do I think I have left?” toward the end of your sets and seeing whether your prediction is accurate is easy to do and may allow you to be confident in predicting RIR if you ever decide to stop taking all your sets to failure (eg: training to failure may not suit all exercises or you may find yourself wanting to do more volume and needing to take a step back as far as intensity of effort goes).
If you’re a strength training enthusiast or a powerlifter and for some reason you do not want to predict your RIR, then just using the traditional %1RM approach may be just fine. A 2018 study by Helms et al aimed to compare the effectiveness of using the traditional %1RM approach versus using an RPE based on RIR approach for load selection. The aim of the study was essentially to determine which method better enhances strength and muscle hypertrophy when other training variables are matched. The study included 21 trained male participants who were divided into two groups: one used percentage 1RM to determine loads while the other used RPE based on RIR to select loads that would achieve a target RPE range (ranging from 5 to 9 RPE depending on the training week). Both groups followed an eight-week daily undulating periodization program, performing squats and bench presses three times per week. Muscle thickness and strength were measured pre- and post-training. Without diving too deep into the study, Helms et al found the following:
Both groups showed significant increases in muscle thickness and 1RM strength for both squat and bench press.
There were no significant differences between the groups in terms of strength gains and muscle thickness, indicating both loading methods were effective, though the RPE group experienced a slightly (non-significantly) larger increase in squat strength.
So, with a well-constructed training program, you can certainly make great gains without needing to assess your RPE or RIR, but autoregulation may help you make slightly better gains than you’d achieve otherwise.
Practical Applications
The current literature suggests that trainees and coaches are imperfect but relatively accurate at predicting RIR. It does, however, seem like in order to be as close to perfect as possible, there are a few things that one can do in order to increase their RIR prediction accuracy. Those seem to be:
Assume that you may be underpredicting RIR by approximately one repetition. Performing an extra repetition and reassessing your perceived RIR may allow you to get some more feedback regarding your ability to predict RIR.
Take the last set of some exercises to failure while also predicting RIR during said set. Anecdotally, this may help you calibrate your ability to predict RIR for each exercise and allow you to appropriately select loads for subsequent sets if you do indeed end up either under- or over-predicting RIR. Note that I am not advising you to risk getting hurt by going for that extra repetition on squats without a spotter, but rather, to safely take a set to failure when appropriate. It may be that your ability to predict RIR is on point for exercises where rep ranges are very low and loads are high (e.g., sets of 1-3 repetitions), but you often find yourself struggling to feel confident in your RIR predictions for sets of 10-15 repetitions on various exercises. Taking a set of lat pulldowns and chest press to the point where you attempt another repetition and are unable to get it will not hurt your gains or generate an enormous amount of fatigue, but it may give you some valuable feedback regarding your ability to predict RIR. Additionally, it may also allow you to better familiarize yourself with the sensation of approaching failure, something that may differ from exercise to exercise, muscle group trained, etc.
If you do not train to failure, mostly opting for sets below 12 repetitions may also make your ability to predict RIR better. That’s not to say “never go above 12 repetitions because you won’t know whether you’re truly close to failure,” but to experiment with lower repetition ranges if you’re generally finding it difficult to feel confident in your RIR predictions during higher repetition sets.
Actively trying to predict RIR during a set, and specifically when the set is approaching the end, may also be a solid way to improve your RIR accuracy. Consciously trying to predict RIR mimics what some of the participants of the above studies were doing, and may allow you to actually be more on point with your RIR prediction versus assuming that a fixed load/rep range configuration will automatically land you in the “right” RIR.