K Culture WireKorean pop culture, in numbers

IU and Im Yoon-ah started before 18. Ma Dong-seok started at 32 — and it is not just career length

Name: IU and Im Yoon-ah started before 18. Ma Dong-seok started at 32 — and it is not just career length
Creator: K Culture Wire

Korean stars who started before they were 18 are read more today than those who started at 26. Is that about starting young, or about having been around longer?

Of 584 Korean actors and singers whose career start year is on record, the 105 who started before 18 — IU, Im Yoon-ah, Bae Suzy among them — are read 11.89 times per million. The 67 who started at 26 or later, including Ma Dong-seok, Lee Sun-kyun, Lee Joo-bin, are read 2.43.

The obvious objection, answered first

The obvious objection is that someone who started at ten has simply had longer to collect readers. It does not hold: the median career length is within two years of itself in all four bands.

Started at	People	Median career, years
under 18	105	22
18 to 21	253	22
22 to 25	159	20
26 or older	67	22

We also held career length inside a window of 10 to 20 years and measured again. The ladder survives, but the two youngest bands come together — so it becomes three steps rather than four.

The ladder, and the names in each step

Started at	People	Reads per million	Career held at 10–20 years	Most-read in this band
under 18	105	11.89	13.42	IU 208.57 · Im Yoon-ah 205.69 · Bae Suzy 165.9
18 to 21	253	6.68	12.65	Cha Eun-woo 200.33 · Park Ji-hoon 128.42 · Ji Chang-wook 127.62
22 to 25	159	4.8	8.58	Go Youn-jung 273.06 · Kim Seon-ho 220.56 · Song Joong-ki 106.36
26 or older	67	2.43	2.9	Ma Dong-seok 59.84 · Lee Sun-kyun 43.55 · Lee Joo-bin 28.55

Every step falls, both before and after holding career length — 4.89× end to end raw, 4.63× with career held. But look at the fourth column. Once career is held, the two youngest bands land at 13.42 and 12.65, close enough that we would not separate them. The four-step ladder becomes three.

Every median here survives removing any single person from its band, which rules out one way it could be fragile and does not show that it is stable. The check is a jackknife, which is known to understate how much a median moves. The method and its limits are here.

What it says about one person

Take one star who started before 18 and one who started at 26 or later, at random. How often is the early starter the more-read one?

72.7%

The chance that a randomly chosen star who started before 18 is read more than a randomly chosen star who started at 26 or later, counting ties as half. This is the common-language effect size (McGraw and Wong 1992); it equals the Mann-Whitney U statistic divided by the number of pairs (Mann and Whitney 1947). Fifty per cent would mean the bands say nothing at all about an individual.

What this page cannot tell you

Not cause. Starting young may bring readers, being the kind of child who gets cast may be the thing that shows up later, or something we have not measured drives both.

Not present age. Debut age and current age move together here and we cannot separate them.

Not everyone. Only stars whose career start year is on Wikidata are in this panel.

Not popularity. Reads count people opening an encyclopaedia article in four languages.

The four bands are ours. We cut at 18, 22 and 26 because those splits leave every band with enough people to measure; a different cut would give different medians.

How this was counted

Source — Wikidata (CC0) for the year a career started and the date of birth; Wikimedia Pageviews API for reads
Window — 2025-08 through 2026-07, 12 months, human traffic only · careers counted to 2026
Method — We report the median rather than the mean. A handful of very large values would pull a mean away from where most of the sample actually sits, and in reads-per-title data those large values are the norm, not an error. Reads are expressed per million reads of that Wikipedia in that month, so that a large edition and a small one can be compared without the larger one winning by size alone. Stability of a median is checked with a jackknife — remove one observation, recompute, repeat for each (Quenouille 1949, Tukey 1958). The Korean figure is read against a control drawn the same way, because a movement that appears in the control as well is not a fact about Korea. Each band is summarised by its median, and then measured again with career length held inside a ten-year window, because the first thing a reader will ask is whether early starters have simply had more time. Beside the medians we report the chance that one person from the youngest band outreads one person from the oldest.
Limits — A median tells you where the middle sits and nothing about the shape around it, so it should be read next to the range or the full distribution, not alone. Dividing by an edition's own total means a month when that edition was unusually busy lowers every article in it. The ratio measures share of attention inside an edition, not how many people read something. The jackknife understates how much a median varies, because removing one observation barely moves a median, and it is not consistent for the median (Miller 1974). A swing it finds is therefore real; a swing it misses is not evidence of stability. Establishing stability would need a delete-d jackknife or a bootstrap (Efron 1979), which we have not built. A control rules an explanation out; it does not identify the one that replaces it. We say what the control removes and stop there. Debut age and present age are tangled and this data cannot separate them: someone who started at five in 1992 is in their late thirties now, and we cannot tell whether the reading follows when they started or how old they are today. The panel is people whose career start year is recorded on Wikidata, which is not everyone, and a missing start year is more likely for someone less written about. Nothing here says starting young causes anything.

The same people by zodiac sign · How many charting titles they have · How we count