Who keeps competing?

Abstract

We follow 724,333 lifters through 2,544,449 meet entries since 2010 and ask after each meet whether the lifter competes again within two years. After the debut, only 52% of men and 58% of women do. Every further meet raises the odds, to 87% after the tenth for men. Five years after the debut, 14% of men and 11% of women are still competing. A logistic model of the return shows what goes with staying: strength (61% back in the weakest fifth against 74% in the strongest, adjusted for the other factors), a valid total (59% after bombing out), a drug-tested federation (68% against 62%) and age: lifters past 45 come back most often. These are associations, not causes.

In plain words

48%
of men do not compete again within two years of their first meet. For women it is 42%. The second meet is the hardest one to get to.
87%
of men come back after their tenth or any later meet. Every meet a lifter does makes the next one likelier.
24%
of lifters who debut at 45 or later are still competing five years on, more than in any younger group. Late starters stay.

Data

The data is the public OpenPowerlifting archive. A lifter’s career is every meet they appear in, in any discipline and any equipment: a lifter who switches to bench-only meets has not quit. A lifter entered in two divisions of the same meet counts once for that meet.

What counts as coming back. After each meet we ask whether the lifter competes again within two years. Two years cover 93% of all gaps between meets, and a fixed window treats every meet alike. Only meets up to October 2024 count, so that two full years of data follow each one.

The math: the return indicator

Let $t$ be a meet of lifter $i$ on date $d_{it}$ , and $d_{i,t+1}$ the date of their next meet, if any. The lifter returns if

R_{it} = \mathbf{1}\big[d_{i,t+1} - d_{it} \le 730 \text{ days}\big]. \tag{1}

The model in Section 5 uses the raw full-power meets in the sample, where a total and a Dots score can be compared.

Coming back

The share of lifters who compete again rises with every meet they have done.

Share who compete again within two years, by meet number

Figure 1. Share of lifters who compete again within two years (1), by the number of the meet in their career. 10 includes all later meets.

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

The big step is the first. After the debut, 48% of men and 42% of women stop. After the second meet the return rate is 66% for men, after the tenth 87%. Part of this is selection: lifters still around at meet ten are the ones who like competing. Part may be habit: a lifter with a few meets behind them has a coach, a team, a plan.

How long careers last

Coming back once is not a career. For the length of a career we count the years from the debut to the last meet.

The math: the Kaplan–Meier estimator

Whether a career has ended is only clear two years after its last meet. A career with a meet in the last two years of the data may still be running; all we know is that it was still running in October 2024, two years before the end of the data, and it is censored there. Careers that began after that are left out. With $n_j$ careers still running just before time $t_j$ , and $e_j$ of them ending at $t_j$ , the share still competing after $t$ years is

\hat S(t) = \prod_{t_j \le t} \left(1 - \frac{e_j}{n_j}\right). \tag{2}

A censored career counts in $n_j$ for as long as it is observed and never as an ending.

Share still competing, by years since the debut

Figure 2. Share of lifters still competing, by years since the debut and age at the debut, Kaplan–Meier estimate (2). Debuts from 2010 to October 2024. At 0 years: the share who ever enter another meet.

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

A year after the debut, 39% of men are still competing; after five years 14%, after ten 5%. Women start out the same but stop a little sooner: after five years 11% of them are still competing. The clearest difference is age at the debut: of lifters who start at 45 or later, 24% are still competing five years on, against 16% of those who start between 24 and 34 and 12% of those who start under 18.

Retention has also fallen while the sport has grown. Of lifters who debuted in 2010, 21% were still competing five years later; of those who debuted in 2018, only 11%. The decline began before the pandemic.

**Table 2.** Share still competing after the debut, Kaplan–Meier estimate (2).
Group	1 year	2 years	5 years	10 years	Lifters

Age groups count lifters whose first meet was a raw full-power meet with a recorded age. Teenage lifters stop most often, plausibly when school ends or the school team does; late starters have usually chosen the sport deliberately.

What goes with coming back

Strong lifters come back more often, but strong lifters are also more experienced, older, more often in tested federations. To pull these apart, a model predicts the return from all of them at once.

The math: the model and adjusted rates

For raw full-power meets, with $x_{it}$ a set of indicators for the meet number, sex, age band, Dots fifth (by sex) or a meet without a valid total, a new best total, a class win, a tested federation and the calendar year, the probability of not coming back is modelled as

\Pr(R_{it} = 0 \mid x_{it}) = \sigma\big(\beta_0 + x_{it}^{\top}\beta\big), \qquad \sigma(z) = \frac{1}{1 + e^{-z}}. \tag{3}

This is a discrete-time hazard model: one row per meet, and the event is stopping after it. To report a factor in percent, we give every meet in the data that factor’s level, keep everything else as it is, and average the predicted return (g-computation):

\bar R(v) = \frac{1}{n} \sum_{i,t} \Big(1 - \sigma\big(\beta_0 + x_{it}(v)^{\top}\beta\big)\Big), \tag{4}

where $x_{it}(v)$ is $x_{it}$ with the factor set to level $v$ . For a new best total, which can only happen from the second meet on, the average runs over those meets.

What goes with coming back

Figure 3. Share of lifters back within two years, raw and adjusted for the other factors (4). Raw full-power meets.

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

Strength. Raw, the strongest fifth comes back 80% of the time and the weakest 59%. Adjusted, the gap shrinks to 74% against 61%: part of what looks like strength is experience.
Bombing out. A meet without a valid total, mostly a bomb-out, is followed by a return 59% of the time, about as low as for the weakest fifth.
A new best total. From the second meet on, a new best goes with 75% back against 73% without. A small difference.
Winning the class. 69% against 66%. Small, too.
Testing. Lifters in drug-tested federations come back 68% of the time, in untested ones 62%.
Age. Lifters aged 45 and over come back 74% of the time, those aged 24 to 34 only 65%.

None of this says what makes a lifter stay. A new best may be motivating, or simply a sign of a lifter who trains a lot and would have come back anyway.

Limits and next steps

Gaps in the archive. A lifter who moves to a federation the archive does not cover looks like someone who stopped. OpenPowerlifting covers most large federations, but not all.
Changing cohorts. Figure 2 pools all debuts since 2010. Only the early cohorts, who stayed longer, reach ten years, so the long end of the curve mixes early and late cohorts.
Names. The archive tells lifters apart by name, with extra marks where two people share one. A few careers are split or merged by mistake.
Associations. Every factor in Section 5 travels with things the data does not see, such as a coach, a team or the reason someone entered their first meet.

How it was computed. The dataset is built from the OpenPowerlifting dump with DuckDB; the model is fitted in Python with scikit-learn. Every number on this page is recomputed with each data update.

Will you be back?

The calculator at the end of the page puts you into the model. Enter how many meets you have done and how the last one went. The calculator applies model (3) to a lifter like you at a meet in the last fully observed year, and gives the share of such lifters who compete again within two years.

References

OpenPowerlifting. OpenPowerlifting data dump, public domain. openpowerlifting.org.
Kaplan, E. L. and Meier, P. (1958). Nonparametric estimation from incomplete observations. Journal of the American Statistical Association 53, 457–481.
Singer, J. D. and Willett, J. B. (1993). It’s about time: using discrete-time survival analysis to study duration and the timing of events. Journal of Educational Statistics 18, 155–195.
Robins, J. (1986). A new approach to causal inference in mortality studies with a sustained exposure period. Mathematical Modelling 7, 1393–1512 (g-computation).
Pedregosa, F. et al. (2011). Scikit-learn: machine learning in Python. JMLR 12, 2825–2830.