When are lifters strongest?

Abstract

We study 898,009 raw full-power results of 341,216 lifters aged 14 to 70. Taking all ages at once, the median Dots score peaks at 24 for women and 27 for men. But young and old lifters in that snapshot are mostly different people: 88% of men competing at 50 to 54 have no recorded meet before 30. Following the same lifters over time with a fixed-effects model moves the peak to 35 for women and 31 for men. Part of that is not age: within a career, age and calendar year advance together, and among 24- to 30-year-olds the sport gains about 3.6 Dots a year for women and 1.6 Dots for men. Age, period and cohort cannot be separated without an assumption. Removing the full trend puts the peak back at 25 and 27; the truth most likely lies in between.

In plain words

27–31
is the age at which men are strongest, measured in Dots, depending on how much of the sport’s yearly progress you count as their own. For women the range is 25–35.
88%
of men competing at 50 to 54 have no recorded meet before 30, in any equipment. A chart of all ages compares different people, not one lifter growing older.
1.0 to 1.4%
per year is what men lose from their total between 40 and 60, the same lifters followed over time. Strength fades slowly.

Data

The data is the public OpenPowerlifting archive. We keep every valid raw full-power result (squat, bench, deadlift) from a sanctioned meet with a known age between 14 and 70, leaving out high-school divisions. A lifter entered in two divisions of the same meet appears once, with the higher total.

Strength is measured as the Dots score, the total divided by a curve in bodyweight, so that lifters of different size can be compared. For the calculator at the end we also use the total itself.

The models in Sections 4 and 5 use only lifters with at least two meets, since a single meet says nothing about change.

A snapshot of all ages

The simplest answer takes every result, groups it by age and asks where the typical lifter is strongest.

Median Dots by age, all lifters at once

Figure 1. Median Dots score by age. Shaded: the middle half of lifters (25th to 75th percentile). Ages with at least 200 results.

Third Attempt by Daniel Vahl · Data: OpenPowerlifting, current dump

The median rises steeply through the teens, peaks at 24 for women and 27 for men and declines slowly afterwards. At 50 the median is 51 Dots below the peak for women and 47 Dots for men.

The trouble is who is in the picture. The 50-year-olds in Figure 1 are not the 25-year-olds of the same chart grown older: 97% of women and 88% of men competing at 50 to 54 have no recorded meet before 30. Most of them only started competing later in life. A snapshot compares groups of people, and the groups differ in more than age.

Same lifters over time

The alternative is to follow each lifter through their own career and ask how their score changes as they get older. Everything that is fixed about a person, talent, sex, the age they started, how seriously they train, then drops out of the comparison.

The math: the fixed-effects model

Let $D_{it}$ be the Dots score of lifter $i$ at meet $t$ , $A_{it}$ their age in full years and $K_{it}$ the number of the meet among their results in this sample, capped at 6. We fit

D_{it} = \alpha_i + \sum_{a \neq 25} \beta_a \,\mathbf{1}[A_{it} = a] + \sum_{k=2}^{6} \gamma_k \,\mathbf{1}[K_{it} = k] + \varepsilon_{it}. \tag{1}

The lifter effect $\alpha_i$ absorbs everything constant about a person. $\beta_a$ is the age profile, measured against age 25; $\gamma_k$ is what the $k$ -th meet adds over the first at the same age.

The $\alpha_i$ are removed by subtracting each lifter’s own mean from every variable,

\ddot D_{it} = D_{it} - \bar D_i, \qquad \ddot x_{it} = x_{it} - \bar x_i, \tag{2}

and fitting $\ddot D$ on the demeaned indicators $\ddot x$ by least squares. Only lifters with two or more meets contribute. Uncertainty comes from 200 bootstrap samples that draw whole careers with replacement, which keeps the dependence between one lifter’s meets intact.

Experience counts separately. Lifters improve fast over their first meets, whatever their age: at the same age, the sixth meet and later score +56 Dots above the first for women and +40 Dots for men. Since lifters are also older at their sixth meet than at their first, leaving these terms out would credit the learning to age.

Dots gained over the first meet, at the same age

Figure 2. Dots gained over the first meet at the same age,

\gamma_k

in (1). The 95% bootstrap intervals are at most about two Dots wide.

Third Attempt by Daniel Vahl · Data: OpenPowerlifting, current dump

With experience held apart, the age profile of the same lifters looks very different from the snapshot. Women keep improving well into their thirties and peak at 35 (95% interval 34–39), men at 31 (30–32). At 50, women are only 17 Dots below their peak and men 37 Dots. Studies of world-class careers put the peak of powerlifters at about 35, for men and women alike: close to the women here and later than for the men.

The trend problem

There is a catch, and it is not a statistical detail. Inside one career, a year older is always also a year later. If everyone in the sport gets stronger over the years, through better coaching, better programs or simply more lifters taking it seriously, that rise shows up in (1) as an effect of age.

And the sport does get stronger. Among lifters aged 24–30, the median Dots score has risen by about 3.6 Dots a year for women and 1.6 Dots for men since 2012, about 1.1% and 0.4% a year.

The sport gets stronger every year

Figure 3. Median Dots score of lifters aged 24 to 30 by year. The slope of a straight line through these points is the trend

\tau

in (4).

Third Attempt by Daniel Vahl · Data: OpenPowerlifting, current dump

Can the data tell age and trend apart? No. This is the classic age-period-cohort problem.

The math: why age and trend cannot be separated

Write $P_{it}$ for the calendar year of the meet and $C_i$ for the lifter’s birth year. Then

A_{it} = P_{it} - C_i. \tag{3}

Suppose the truth is an age profile $f(a)$ plus a period effect $\pi(p)$ . For any number $\delta$ ,

f(A) + \pi(P) = \big[f(A) + \delta A\big] + \big[\pi(P) - \delta P\big] + \delta C,

and $\delta C$ is constant for each lifter, so $\alpha_i$ absorbs it. Every choice of $\delta$ fits the data equally well: a steeper age profile with a flatter trend, or the reverse. The data pins down everything except one straight line.

So the answer needs an assumption about that line. We take the trend $\tau$ from Figure 3 and remove a share $s$ of it from the age profile:

\beta^{(s)}_a = \beta_a - s\,\tau\,(a - 25), \qquad 0 \le s \le 1. \tag{4}

With $s = 0$ the sport stands still and every gain in a career is age. With $s = 1$ the whole rise of the field reaches every lifter, at every age.

Three ways to measure strength by age

Figure 4. Dots relative to age 25: the snapshot from Figure 1 (dashed), the same lifters over time (1) and the same lifters with the full trend removed (4),

s = 1

. Shaded: 95% bootstrap intervals.

Third Attempt by Daniel Vahl · Data: OpenPowerlifting, current dump

With the full trend removed, the peak moves back to 25 for women (24–26) and 27 for men (25–27), close to the snapshot. The decline after it is steeper, though: at 50, women are 93 Dots below their peak and men 70 Dots, more than the snapshot shows. That steeper decline comes from the assumption itself: with $s = 1$ , every year of age is charged the full trend $\tau$ , so age 50 drops by $25\tau$ against age 25.

**Table 2.** Peak age by method, with 95% bootstrap intervals for the models.
Method	Men	Women

The intervals cover sampling noise only. The choice of $s$ moves the peak much further than any interval.

Which $s$ is right? Probably neither end. Part of the trend in Figure 3 is real progress that reaches everyone: better programs, more knowledge, better coaching. Part of it may be who competes: as the sport grows and the federations in the archive change, the mix of lifters changes, and a different mix does not make any single lifter stronger. So $s = 0$ overstates the peak age as soon as there is any real progress, and $s = 1$ understates it if the mix has shifted toward stronger lifters. Halfway, the peak is at 28 for women and 29 for men.

Whichever $s$ between 0 and 1 you choose, the decline after 40 is slow: between 40 and 60, the total of the same lifters falls by 0.5 to 1.5% a year for women and 1.0 to 1.4% for men, the lower figure with $s = 0$ and the higher with $s = 1$ .

Limits and next steps

Who keeps competing. The model follows lifters only while they compete. If lifters whose strength fades are more likely to stop, the decline is understated. A later study looks at who keeps competing.
One straight line. The trend is taken as linear and equal for all ages. Older lifters may profit less from new training ideas, or more.
Bodyweight. Dots compares lifters of different size, but bodyweight changes with age, too. The calculator below works with the total, which includes it.

How it was computed. The dataset is built from the OpenPowerlifting dump with DuckDB, and the models are fitted in Python with NumPy. Every number on this page is recomputed with each data update.

Your strength at another age

The calculator at the end of the page carries your own total to other ages. The calculator uses the same model as (1), with the logarithm of the total in place of Dots, so changes are in percent and include bodyweight. The slider sets the share $s$ in (4): how much of the sport’s yearly rise you think is your own.

The math: carrying a total to another age

Let $\ell^{(s)}(a)$ be the age profile of $\log T$ , the logarithm of the total, from (1), with the trend in (4) measured as the slope of the median log Dots. For a total $T$ at age $a$ , the estimate at age $a'$ is

\widehat T(a') = T \exp\big(\ell^{(s)}(a') - \ell^{(s)}(a)\big). \tag{5}

References

OpenPowerlifting. OpenPowerlifting data dump, public domain. openpowerlifting.org.
Wooldridge, J. M. (2010). Econometric Analysis of Cross Section and Panel Data, 2nd ed. MIT Press (fixed effects).
Efron, B. and Tibshirani, R. J. (1993). An Introduction to the Bootstrap. Chapman & Hall.
Yang, Y. and Land, K. C. (2013). Age-Period-Cohort Analysis: New Models, Methods, and Empirical Applications. CRC Press.
Solberg, P. A., Hopkins, W. G., Paulsen, G. and Haugen, T. A. (2019). Peak age and performance progression in world-class weightlifting and powerlifting athletes. International Journal of Sports Physiology and Performance 14, 1357–1363.