Scientific stagnation

Shrinking marginal returns to science and technology
2012-02-242013-12-30 abandoned certainty: possible importance: 9

One com­mon ob­jec­tion to the idea of a is that any ex­po­nen­tial in­crease in some­thing like com­put­ing power may be off­set by a cor­re­spond­ing in­crease in the diffi­culty of the re­main­ing sci­en­tific & prac­ti­cal prob­lems, lead­ing to es­sen­tially a stale­mate and an in­defi­nite con­tin­u­a­tion of the sta­tus quo.

This is defi­nitely a con­cern, and one widely dis­cussed un­der var­i­ous terms like “the great stag­na­tion”, “pick­ing the low-hang­ing fruit”, and “di­min­ish­ing mar­ginal re­turns”. I think there’s a lot of force to the S-curve view of things and it’s one of the most plau­si­ble rea­sons that there might be no Sin­gu­lar­ity in a sense we’d rec­og­nize now.

There’s a few diffi­cul­ties in grap­pling with this pro­pos­al:

  1. it’s not clear that this de­feats all forms of Sin­gu­lar­i­ty; for ex­am­ple, right now hu­mans are es­sen­tially com­pletely un-self­mod­i­fy­ing. A healthy ge­nius to­day is smart in ba­si­cally the same ways as a healthy ge­nius 1500 years ago. Sure, we have things like Google and com­put­ers to help them out a lot, but the differ­ence does­n’t seem that big. When we fi­nally get full AI, it’s not clear that it will be a sim­ple ex­trap­o­la­tion of cur­rent hu­mans and sub­ject to the same di­min­ish­ing re­turns and so es­sen­tially busi­ness as usu­al. It may well be wildly differ­ent and an en­tirely new S-curve of its own which will pe­ter out, if it ever does, in a wildly differ­ent post-hu­man regime. This is a way in which mar­ginal re­turns could be di­min­ish­ing, the S-curve model of tech­no­log­i­cal im­prove­ment true, but a Sin­gu­lar­ity hap­pen just as pre­dicted by the Vingean and Yud­kowskian schools.
  2. on its own terms, it makes lit­tle un­con­di­tional pre­dic­tion: maybe prob­lem diffi­culty is in­creas­ing ex­po­nen­tial­ly, but the re­sources be­ing de­voted to the prob­lem are in­creas­ing ex­po­nen­tially too - differ­ent ex­po­nents, or differ­ent con­stant fac­tors. There’s a bil­lion Chi­nese ‘com­ing on­line’ for R&D, one might say, and that com­pen­sates for a lot.
  3. the lit­er­a­ture is big, the prox­ies for re­turn on in­vest­ments poor, and it’s hard to get a grasp on it all.

I was read­ing up on the sub­ject and wrote a lit­tle bit here, but I even­tu­ally re­al­ized I could­n’t keep up with all the rel­e­vant pa­pers or con­tribute any­thing new, so I aban­doned the top­ic. My notes are left be­low for those in­ter­ested in the top­ic.

Diminishing marginal returns

The Long Stag­na­tion the­sis can be sum­ma­rized as: “West­ern civ­i­liza­tion is ex­pe­ri­enc­ing a gen­eral de­cline in mar­ginal re­turns to in­vest­ment”. That is, every $1 or other re­source (such as ‘trained sci­en­tist’) buys less in hu­man well-be­ing or tech­nol­ogy than be­fore, ag­gre­gated over the en­tire econ­o­my.

This does not im­ply any of the fol­low­ing:

  1. No ex­po­nen­tial curves ex­ist (rather, they are ex­po­nen­tial curves which are part of sig­moids which have yet to level off; Moore’s law and stag­na­tion can co-ex­ist)

    Sud­den dra­matic curves can ex­ist even amid an econ­omy of di­min­ish­ing mar­ginal re­turns; to over­turn the over­all curve, such a spike would have to be a mas­sive so­ci­ety-wide rev­o­lu­tion that can make up for huge short­falls in out­put.

  2. Any met­rics in ab­solute num­bers have ceased to in­crease or have be­gun to fall (patents can con­tinue grow­ing each year if the amount in­vested in R&D or num­ber of re­searchers in­creas­es)

  3. We can­not achieve mean­ing­ful in­creases in stan­dards of liv­ing or ca­pa­bil­i­ties (the In­ter­net is a ma­jor ac­com­plish­ment)

  4. Spe­cific sci­en­tific or tech­no­log­i­cal will not be achieved (eg. AI or nan­otech) or be achieved by cer­tain dates

  5. The stag­na­tion will be vis­i­ble in a dra­matic way (eg. bar­bar­ians loot­ing New York City)

The stag­na­tion the­sis in­stead sug­gests that naive fore­casts have un­der­-ap­pre­ci­ated er­ror-bars, which can­not be re­duced with­out care­ful con­sid­er­a­tion. It sug­gests that we may un­der­-es­ti­mate the chance of pe­ri­ods of es­sen­tially no pro­gress, like Eu­clid­ean geom­e­try or as­tron­omy be­tween the first cen­tury BC and the Re­nais­sance, pe­ri­ods where the forces that were able to over­come di­min­ish­ing re­turns sud­denly hit hard brick walls1. In par­tic­u­lar, if the stag­na­tion the­sis is true and the over­all land­scape of sci­ence/tech­nol­ogy looks like broad slow grad­ual im­prove­ments with oc­ca­sional sig­moids blast­ing up ex­po­nen­tials and put­ter­ing out, as op­posed to a more tech­no-op­ti­mistic Kurzweil­ian sce­nario of ‘ac­cel­er­at­ing re­turns’ over much of the sci­ence/tech­nol­ogy land­scape, then we should as­sign more weight to sce­nar­ios in which tech­nolo­gies are ‘im­bal­anced’; for ex­am­ple, a sce­nario where the Moore’s law sig­moid does not flat­ten out un­til 2040 but the AI soft­ware curve con­tin­ues to fol­low a grad­ual in­crease poses a se­ri­ous ‘hard­ware over­hang’ risk as enor­mous com­put­ing power sits around wait­ing for the first AI pro­gram just barely well-de­signed enough to make use of it and fix its prim­i­tive al­go­rithms to make proper use of said com­put­ing pow­er. An im­bal­ance fa­vor­ing ei­ther hard­ware or soft­ware may be dis­as­trous: if the luck of the sig­moid draw fa­vors hard­ware, then the first prim­i­tive pro­gram to come along will win all the mar­bles; if chance in­stead sends soft­ware up on a sig­moid rock­et, then this in­cen­tivizes an arms race to as­sem­ble enough com­put­ing power to run one’s faith­ful mil­i­ta­rized AI and win all the mar­bles be­fore an­other ac­tor can run their slav­ishly loyal AI. (Whereas in a Kurzweil­ian sce­nario of in­ter­lock­ing feed­back loops, the AI pro­gram would be de­vel­oped roughly around the same time a com­puter pow­er­ful enough to run it at all is de­vel­oped, and any over­hang po­ten­tial will be lim­ited com­pared to the other sce­nar­ios.) Damien Brod­er­ick:

What if, as Ver­nor Vinge pro­posed, ex­po­nen­tially ac­cel­er­at­ing sci­ence and tech­nol­ogy are rush­ing us into a Sin­gu­lar­ity (Vinge, 1986; 1993), what I have called the Spike? Tech­no­log­i­cal time will be nei­ther an ar­row nor a cy­cle (in Stephen Jay Gould’s phrase), but a se­ries of up­wardly ac­cel­er­at­ing lo­gis­ti­cal S-curves, each sup­plant­ing the one be­fore it as it flat­tens out. Then there’s no pat­tern of rea­soned ex­pec­ta­tion to be mapped, no know­able Cher­nobyl or Fukushima Dai­ichi to de­plore in ad­vance. Merely - opac­i­ty.

The stag­na­tion the­sis is as big as his­to­ry, and has a long lit­er­a­ture of ‘de­clin­ism’, eg. Spen­gler’s 1918 , and so it is tempt­ing to take the Out­side View and mock it as ob­vi­ously fal­si­fied - but this is where the caveats about mar­ginal re­turns come into play. To give a sim­ple ex­am­ple: world pop­u­la­tion in 1900 was around ~1.6 bil­lion, and in 2000 ~6 bil­lion, an in­crease by a fac­tor of 3.75. Plau­si­bly, the pop­u­la­tions of ed­u­cated sci­en­tists and other such peo­ple in­creased even more2 (the frac­tion of the Amer­i­can pop­u­lace go­ing to col­lege in 1900 was, shall we say, smaller than in 2000). So even if the 2000 sci­en­tists were shock­ingly 50% less ‘pro­duc­tive’ than their 1900 coun­ter­parts, be­cause there are >3.75 times as many, we will still wit­ness >1.8x as much pro­duc­tiv­i­ty. It would be very easy to sim­ply com­pare 2000 and 1900 and say talk of stag­na­tion is lu­di­crous. So we see the bur­den ‘mar­ginal re­turns’ puts on us: we need to be con­stantly ad­just­ing any ab­solute fig­ures - which are hard enough to come by - for the size of the rel­e­vant pop­u­la­tion.

When we seek to mea­sure stag­na­tion, we have sev­eral main ar­eas of in­ter­est:

  1. pure sci­ence

    • cost:

      • per­sons per break­through or pa­per or other met­ric
      • dol­lars per met­ric
      • age at dis­cov­ery
    • ben­e­fit:

      • judge­ment of con­tem­po­raries: ci­ta­tions & awards
      • judge­ment of pos­ter­i­ty:
  2. com­mer­cial­iza­tion

    • cost:

      • per­sons per re­sult
      • R&D bud­gets per re­sult
    • ben­e­fit:

      • sales
      • in­creases in met­rics of qual­ity of life (eg. lifes­pan)
  3. gen­eral eco­nom­ics

    • growth rates
    • pro­duc­tiv­ity changes

With this schemat­ic, we can slot re­sults in nice­ly. For ex­am­ple, ’s Hu­man Ac­com­plish­ment en­gages in his­to­ri­om­e­try, find­ing that on a pop­u­la­tion-ad­justed (per cap­i­ta) ba­sis, the most pro­duc­tive pe­riod for the sci­ences was the late 1800s; his re­sults would be listed un­der pure sci­ence, ben­e­fit, judge­ment of pos­ter­i­ty, since he uses only ref­er­ences writ­ten after 1950 about peo­ple & re­sults be­fore 1950 (to con­trol for the ob­vi­ous bi­as­es). On the other hand, Tyler Cowen’s eco­nomic ci­ta­tions in The Great Stag­na­tion would go into gen­eral eco­nom­ics, while Joseph Tain­ter’s Col­lapse of Com­plex So­ci­eties will strad­dle all 3 cat­e­gories but mostly eco­nom­ics.



All else equal, the less time it takes a sci­en­tist to make a break­through, the cheaper the break­through is - less op­por­tu­nity cost to him, the sooner oth­ers can build on it, less main­te­nance & de­pre­ci­a­tion, less effort re­quired to cover pre­req­ui­sites and get up to speed, etc. Mur­ray un­for­tu­nately does not give us a deep his­tory of ‘age of great­est ac­com­plish­ment’, but we can still look at the 20th cen­tu­ry’s No­bel Prizes (eg. Chem­istry ages); Jones & Wein­berg 2011 matched win­ners with their age when they per­formed the award-win­ning work, sum­mary:

A study of No­bel Lau­re­ates from 1901 to 2008 in these three fields ex­am­ined the age at which sci­en­tists did their prize-win­ning work. Re­sults showed that be­fore 1905, about two-thirds of win­ners in all three fields did their prize-win­ning work be­fore age 40, and about 20% did it be­fore age 30. But by 2000, great achieve­ments be­fore age 30 nearly never oc­curred in any of the three fields. In physics, great achieve­ments by age 40 only oc­curred in 19% of cases by the year 2000, and in chem­istry, it nearly never oc­curred….Ear­lier work on cre­ativ­ity in the sci­ences has em­pha­sized differ­ences in the ages when cre­ativ­ity peaks across var­i­ous sci­en­tific dis­ci­plines, as­sum­ing that those differ­ences were sta­ble over time, Wein­berg said. But this new work sug­gests that the differ­ences in the age of cre­ativ­ity peaks be­tween fields like chem­istry and physics are ac­tu­ally quite small com­pared to the differ­ences in cre­ativ­ity peaks be­tween time pe­ri­ods within each dis­ci­pline…­For the study, the re­searchers an­a­lyzed the com­plete set of 525 No­bel Prizes given be­tween 1901 and 2008 in the three fields - 182 in physics, 153 in chem­istry and 190 in med­i­cine. Through ex­ten­sive his­tor­i­cal and bi­o­graph­i­cal analy­sis, they de­ter­mined the ages at which each No­bel Prize win­ner pro­duced their prize-win­ning work. In gen­er­al, there was an ag­ing pat­tern over the 20th cen­tury as to when sci­en­tists made their break­through dis­cov­er­ies, al­though there were differ­ences be­tween the three fields. The most in­ter­est­ing case is physics, Wein­berg said. In physics, there was an es­pe­cially no­table in­crease in the early 20th cen­tury in the fre­quency of young sci­en­tists pro­duc­ing prize-win­ning work. The pro­por­tion of physi­cists who did their prize-win­ning work by age 30 peaked in 1923 at 31%. Those that did their best work by age 40 peaked in 1934 at 78%. The pro­por­tion of physi­cists un­der age 30 or 40 pro­duc­ing No­bel Prize-win­ning work then de­clined through­out the rest of the cen­tu­ry…An­other rea­son that younger sci­en­tists may have made more sig­nifi­cant con­tri­bu­tions early in the 20th cen­tury is that they fin­ished their train­ing ear­lier in life. The ma­jor­ity of No­bel Lau­re­ates re­ceived their doc­toral de­grees by age 25 in the early 20th cen­tu­ry, the re­searchers found. How­ev­er, all three fields showed sub­stan­tial de­clines in this ten­den­cy, with nearly no physics or chem­istry lau­re­ates re­ceiv­ing their de­grees that early in life by the end of the cen­tu­ry. In an­other analy­sis, the re­searchers ex­am­ined the age of stud­ies ref­er­enced in im­por­tant sci­en­tific pa­pers in the three fields through the 20th cen­tu­ry. They found that in the early part of the 1900s – the time when quan­tum me­chan­ics made its mark – there was a strong ten­dency for physics to ref­er­ence mostly re­cent work. “The ques­tion is, how much old knowl­edge of the field do you need to know to make im­por­tant sci­en­tific con­tri­bu­tions in your field?” Wein­berg said. “The fact that physi­cists in the early 20th cen­tury were cit­ing mostly re­cent work sug­gests that older sci­en­tists did­n’t have any ad­van­tage – their more com­plete knowl­edge of older work was­n’t nec­es­sary to make im­por­tant con­tri­bu­tions to the field. That could be one rea­son why younger sci­en­tists made such a mark.” But now, physi­cists are more likely to cite older stud­ies in their pa­pers, he said. That means older sci­en­tists may have an ad­van­tage be­cause of their depth of knowl­edge.

Rather, a re­searcher’s out­put tends to rise steeply in the 20’s and 30’s peak in the late 30’s or early 40’s and then trail off slowly through later years (Lehman, 1953; Si­mon­ton, 1991).

  • Lehman HC (1953) Age and Achieve­ment (Prince­ton Uni­ver­sity Press, Prince­ton, NJ)
  • Si­mon­ton DK (1991) “Ca­reer land­marks in sci­ence: In­di­vid­ual differ­ences and in­ter­dis­ci­pli­nary con­trasts”. Dev Psy­chol 27:119-130

Jones 2006 cov­ered a sim­i­lar up­ward shift in age for noted in­ven­tors

[Why peak in the early 40s? The age-re­lated de­cline in in­tel­li­gence is steady from 20s. There must be some bal­ance be­tween ac­quir­ing and ex­ploit­ing in­for­ma­tion with one’s dis­ap­pear­ing in­tel­li­gence which pro­duces a peak in the 40s be­fore the de­cline kills pro­duc­tiv­i­ty; the graph in ag­ing sec­tion of the DNB cu­ri­ously shows late 40s is where you hit 0 std-de­vs, in­tel­li­gence/mem­o­ry-wise, vis-a-vis the gen­eral pop­u­la­tion]

This rea­son­ing is ex­plored in Jones (2005), which stud­ies “or­di­nary” in­ven­tors, look­ing at all U.S. patents in the 1975-2000 pe­ri­od. is ris­ing at a rate of 6 years/­cen­tu­ry.

Jones 2005:

The es­ti­mates sug­gest that, on av­er­age, the great minds of the 20th Cen­tury typ­i­cally be­came re­search ac­tive at age 23 at the start of the 20th Cen­tu­ry, but only at age 31 at the end - an up­ward trend of 8 years. Mean­while, there has been no com­pen­sat­ing shift in the pro­duc­tiv­ity of in­no­va­tors be­yond mid­dle age.

The tech­no­log­i­cal al­manacs com­pile key ad­vances in tech­nol­o­gy, by year, in sev­eral differ­ent cat­e­gories such as elec­tron­ics, en­er­gy, food & agri­cul­ture, ma­te­ri­als, and tools & de­vices. The year (and there­fore age) of great achieve­ment is the year in which the key re­search was per­formed. For the tech­no­log­i­cal al­manacs, this is sim­ply the year in which the achieve­ment is list­ed.

The largest mass of great in­no­va­tions in knowl­edge came in the 30’s (42%), but a sub­stan­tial amount also came in the 40’s (30%), and some 14% came be­yond the age of 50. Sec­ond, there are no ob­ser­va­tions of great achiev­ers be­fore the age of 19. Dirac and Ein­stein prove quite un­usu­al, as only 7% of the sam­ple pro­duced a great achieve­ment at or be­fore the age of 26. Third, the age dis­tri­b­u­tion for the No­bel Prize win­ners and the great in­ven­tors, which come from in­de­pen­dent sources, are ex­tremely sim­i­lar over the en­tire dis­tri­b­u­tions. Only 7% of in­di­vid­u­als in the data ap­pear in both the No­bel Prize and great in­ven­tors data sets.

While lab­o­ra­tory ex­per­i­ments do sug­gest that cre­ative think­ing be­comes more diffi­cult with age (e.g. Reese et al, 2001), the de­cline in in­no­v­a­tive out­put at later ages may largely be due to de­clin­ing effort, which a range of so­ci­o­log­i­cal, psy­cho­log­i­cal, in­sti­tu­tion­al, and eco­nomic the­o­ries have been var­i­ously pro­posed to ex­plain (see Si­mon­ton 1996 for a re­view).

  • Reese, H,W. Lee, L.J., Co­hen, S. H., and Puck­ett, J.M. “Effects of In­tel­lec­tual Vari­ables, Age, and Gen­der on Di­ver­gent Think­ing in Adult­hood”, In­ter­na­tional Jour­nal of Be­hav­ioral De­vel­op­ment, No­vem­ber 2001, 25 (6), 491-500
  • Si­mon­ton “Cre­ativ­i­ty,” in The En­cy­clo­pe­dia of Geron­tol­ogy, San Diego, CA: Aca­d­e­mic Press, 1996

In fact, ag­gre­gate data pat­terns, much de­bated in the growth lit­er­a­ture, have noted long-s­tand­ing de­clines in the per-capita out­put of R&D work­ers, both in terms of patent counts and pro­duc­tiv­ity growth (Machlup 1962; Even­son, 1991; Jones 1995a; Ko­r­tum, 1997). Sim­ple cal­cu­la­tions from ag­gre­gate data sug­gest that the typ­i­cal R&D worker con­tributes ap­prox­i­mately 30% as much to ag­gre­gate pro­duc­tiv­ity gains to­day as she did at the open­ing of the 20th Cen­tu­ry.29

29: Com­bin­ing Machlup’s data on growth in knowl­edge pro­duc­ing oc­cu­pa­tions for 1900-1959 (Machlup 1962, Ta­ble X-4) with sim­i­lar NSF data for 1959-1999 (Na­tional Sci­ence Foun­da­tion, 2005), we see that the to­tal num­ber of knowl­edge-pro­duc­ing work­ers in the United States has in­creased by a fac­tor of ap­prox­i­mately 19. Mean­while, the U.S. per-capita in­come growth rate, which prox­ies for pro­duc­tiv­ity growth over the long-run, sug­gests a 6-fold in­crease in pro­duc­tiv­ity lev­els (based on a steady growth rate of 1.8%; see Jones 1995b). The av­er­age rate at which in­di­vid­ual R&D work­ers con­tribute to pro­duc­tiv­ity growth is A=LR , or gA=LR , where A is ag­gre­gate pro­duc­tiv­i­ty, g is the pro­duc­tiv­ity growth rate, and LR is the ag­gre­gate num­ber of R&D work­ers. The av­er­age con­tri­bu­tion of the in­di­vid­ual R&D worker in the year 2000 is then a frac­tion A2000 =A1900 =(L2000 =L1900 ) = 6=19 (32%) of what it was in 1900.

Fig­ure 5 com­pares the es­ti­mated life-cy­cle curves for the year 1900 and the year 2000, us­ing spec­i­fi­ca­tion (3). We see that the peak abil­ity to pro­duce great achieve­ments in knowl­edge came around age 30 in 1900 but shifted to nearly age 40 by the end of the cen­tu­ry. An in­ter­est­ing as­pect of this graph is the sug­ges­tion that, other things equal, life­time in­no­va­tion po­ten­tial has de­clined.

The first analy­sis looks di­rectly at ev­i­dence from Ph.D. age and shows that Ph.D. age in­creases sub­stan­tially over the 20th Cen­tu­ry. The sec­ond analy­sis har­nesses world wars, as ex­oge­nous in­ter­rup­tions to the young ca­reer, to test the ba­sic idea that train­ing is an im­por­tant pre­lim­i­nary in­put to in­no­va­tion. I show that, while the world wars do not ex­plain the 20th cen­tu­ry’s age trend, they do in­di­cate the un­avoid­able na­ture of train­ing: lost years of train­ing ap­pear to be “made up” after the war. The fi­nal analy­sis ex­plores cross-field, cross-time vari­a­tion. I show that vari­a­tions in train­ing du­ra­tion pre­dict vari­a­tions in age at great in­ven­tion

In­deed, sev­eral stud­ies have doc­u­mented up­ward trends in ed­u­ca­tional at­tain­ment among the gen­eral pop­u­la­tion of sci­en­tists. For ex­am­ple, the age at which in­di­vid­u­als com­plete their doc­tor­ates rose gen­er­ally across all ma­jor fields in a study of the 1967-1986 pe­ri­od, with the in­crease ex­plained by longer pe­ri­ods in the doc­toral pro­gram (Na­tional Re­search Coun­cil, 1990). The du­ra­tion of doc­tor­ates as well as the fre­quency and du­ra­tion of post-doc­tor­ates has been ris­ing across the life-sciences since the 1960s (Til­gh­man et al, 1998). A study of elec­tri­cal en­gi­neer­ing over the course of the 20th cen­tury de­tails a long-s­tand­ing up­ward trend in ed­u­ca­tional at­tain­ment, from an ini­tial propen­sity for bach­e­lor de­grees as the ed­u­ca­tional cap­stone to a world where Ph.D.’s are com­mon (Ter­man, 1998).

  • Ter­man, F.E. “A Brief His­tory of Elec­tri­cal En­gi­neer­ing Ed­u­ca­tion”, Pro­ceed­ings of the IEEE, Au­gust 1998, 86 (8), 1792-1800

Most strik­ing­ly, both achieve­ment and Ph.D. age in Physics ex­pe­ri­enced a unique de­cline in the early 20th cen­tu­ry. This un­usual fea­ture, be­yond re­in­forc­ing the re­la­tion­ship be­tween train­ing and achieve­ment age, may also serve to in­form more ba­sic the­o­ries for the un­der­ly­ing dy­nam­ics and differ­ences across fields.

[that there was a fall in age for this rad­i­cal and rev­o­lu­tion­ary pe­riod in physics val­i­dates the gen­eral ap­proach]

First, mean life ex­pectancy at age 10 was al­ready greater than 60 in 1900, while it is clear from Sec­tions 2 and 3 that in­no­va­tion po­ten­tial is mod­est be­yond 60, so that adding years of life be­yond this age would have at most mild effects on the op­ti­miza­tion.25 Re­lat­ed, even mod­est dis­count­ing would sub­stan­tially limit the effect of gains felt 35+ years be­yond the end of train­ing on the mar­ginal train­ing de­ci­sion. Next, com­mon life ex­pectancy changes can­not ex­plain the unique cross-field and cross-time vari­a­tion ex­plored in Sec­tion 4, such as the unique be­hav­ior of physics. More­over, Fig­ure 7 sug­gests, if any­thing, ac­cel­er­at­ing age trends after the sec­ond world war, which is hard to ex­plain with in­creased longevi­ty, where post-war gains have slowed.

One proxy mea­sure is re­search col­lab­o­ra­tion in patent­ing - mea­sured as team size - which is in­creas­ing at over 10% per decade.28 A more di­rect mea­sure of spe­cial­iza­tion con­sid­ers the prob­a­bil­ity that an in­di­vid­ual switches tech­no­log­i­cal ar­eas be­tween con­sec­u­tive patents. Jones (2005) shows that the prob­a­bil­ity of switch­ing tech­no­log­i­cal ar­eas is sub­stan­tially de­clin­ing with time. These analy­ses in­di­cate that train­ing time, E, is ris­ing, while mea­sures of breadth, b, are si­mul­ta­ne­ously de­clin­ing. It is then diffi­cult to es­cape the con­clu­sion that the dis­tance to the knowl­edge fron­tier is ris­ing.28 Large and gen­eral up­ward trends in re­search col­lab­o­ra­tion are also found in jour­nal pub­li­ca­tions (e.g. Adams et al, 2004).

  • Adams, James D., Black, Grant C., Clem­mons, J.R., and Stephan, Paula E. “Sci­en­tific Teams and In­sti­tu­tional Col­lab­o­ra­tions: Ev­i­dence from U.S. Uni­ver­si­ties, 1981-1999”, NBER Work­ing Pa­per #10640, July 2004

Anal­o­gous­ly, prob­lems that re­quire more ex­pe­ri­en­tial train­ing have older peak ages. For in­stance, Jones (2006) finds that the peak age for nat­ural sci­en­tists has drifted higher over the twen­ti­eth cen­tu­ry. Rel­a­tive to 100 years ago, more ex­pe­ri­ence now needs to be ac­cu­mu­lated to reach the cut­ting edge of sci­en­tific fields.

lifes­pan in­creases can­not make up for this: http://­less­wrong.­com/l­w/7jh/liv­ing_­forever_is_hard_­part_2_adult_­longevi­ty/

Jones, Ben­jamin F. “The Bur­den of Knowl­edge and the Death of the Re­nais­sance Man: Is In­no­va­tion Get­ting Hard­er?” NBER Work­ing Pa­per #11360, 2005

Up­ward trends in aca­d­e­mic col­lab­o­ra­tion and length­en­ing doc­tor­ates, which have been noted in other re­search, can also be ex­plained by the mod­el, as can much-de­bated trends re­lat­ing pro­duc­tiv­ity growth and patent out­put to ag­gre­gate in­ven­tive effort. The knowl­edge bur­den mech­a­nism sug­gests that the na­ture of in­no­va­tion is chang­ing, with neg­a­tive im­pli­ca­tions for long-run eco­nomic growth.

Given this in­creas­ing ed­u­ca­tional at­tain­ment, in­no­va­tors will only be­come more spe­cial­ized if the bur­den of knowl­edge mech­a­nism is suffi­ciently strong. More sub­tly, in­come ar­bi­trage pro­duces the sur­pris­ing re­sult that ed­u­ca­tional at­tain­ment will not vary across tech­no­log­i­cal fields, re­gard­less of vari­a­tion in the bur­den of knowl­edge or in­no­v­a­tive op­por­tu­ni­ties.

Any re­la­tion to ?

“As Sci­ence Evolves, How Can Sci­ence Pol­i­cy?”, Jones 2010; sum­mary of all the Jones pa­pers

First, R&D em­ploy­ment in lead­ing economies has been ris­ing dra­mat­i­cal­ly, yet TFP growth has been flat (Jones, 1995b). Sec­ond, the av­er­age num­ber of patents pro­duced per R&D worker or R&D dol­lar has been falling over time across coun­tries (Even­son 1984) and U.S. man­u­fac­tur­ing in­dus­tries (Ko­r­tum 1993). These ag­gre­gate data trends can be seen in the model as an effect of in­creas­ingly nar­row ex­per­tise, where in­no­va­tors are be­com­ing less pro­duc­tive as in­di­vid­u­als and are re­quired to work in ever larger teams.

  • Jones “Time Se­ries Tests of En­doge­nous Growth Mod­els,” Quar­terly Jour­nal of Eco­nom­ics, 1995b, 110, 495-525

Es­sen­tial­ly, the greater the growth in the bur­den of knowl­edge, the greater must be the growth in the value of knowl­edge to com­pen­sate. Ar­tic­u­lated views of why in­no­va­tion may be get­ting harder in the growth lit­er­a­ture (Ko­r­tum 1997, Segerstrom 1998) have fo­cused on a “fish­ing out” idea; that is, on the pa­ra­me­ter χ. The in­no­va­tion lit­er­a­ture also tends to fo­cus on “fish­ing out” themes (e.g. Even­son 1991, Cock­burn & Hen­der­son, 1996). This pa­per offers the bur­den of knowl­edge as an al­ter­na­tive mech­a­nism, one that makes in­no­va­tion hard­er, acts sim­i­larly on the growth rate, and can ex­plain ag­gre­gate data trends (see Sec­tion 4). Most im­por­tant­ly, the model makes spe­cific pre­dic­tions about the be­hav­ior of in­di­vid­ual in­no­va­tors, al­low­ing one to get un­der­neath the ag­gre­gate facts and test for a pos­si­ble ris­ing bur­den of knowl­edge us­ing mi­cro-data

  • Ko­r­tum, Samuel S. “Equi­lib­rium R&D and the De­cline in the Paten­t-R&D Ra­tio: U.S. Ev­i­dence,” Amer­i­can Eco­nomic Re­view Pa­pers and Pro­ceed­ings, May 1993, 83, 450-457
  • Even­son 1991 “Patent Data by In­dus­try: Ev­i­dence for In­ven­tion Po­ten­tial Ex­haus­tion?” Tech­nol­ogy and Pro­duc­tiv­i­ty: The Chal­lenge for Eco­nomic Pol­icy, 1991, Paris: OECD, 233-248

This re­sult is con­sis­tent with Hen­der­son & Cock­burn’s (1996) find­ing that re­searchers in the phar­ma­ceu­ti­cal in­dus­try are hav­ing a greater diffi­culty in pro­duc­ing in­no­va­tions over time.

  • Hen­der­son, Re­becca and Cock­burn, Iain. “Scale, Scope, and Spillovers: The De­ter­mi­nants of Re­search Pro­duc­tiv­ity in Drug Dis­cov­ery,” Rand Jour­nal of Eco­nom­ics, Spring 1996, 27, 32-59.

The age at which in­di­vid­u­als com­plete their doc­tor­ates rose gen­er­ally across all ma­jor fields from 1967-1986, with the in­crease ex­plained by longer pe­ri­ods in the doc­toral pro­gram (Na­tional Re­search Coun­cil, 1990). The du­ra­tion of doc­tor­ates as well as the fre­quency of post-doc­tor­ates has been ris­ing across the life-sciences since the 1960s (Til­gh­man et al, 1998). An up­ward age trend has also been noted among the great in­ven­tors of the 20th Cen­tury at the age of their noted achieve­ment (Jones, 2005), as shown in Ta­ble 1. Mean­while, like the gen­eral trends in in­no­va­tor team­work doc­u­mented here, up­ward trends in aca­d­e­mic coau­thor­ship have been doc­u­mented in many aca­d­e­mic lit­er­a­tures, in­clud­ing physics and bi­ol­ogy (Zuck­er­man & Mer­ton, 1973), chem­istry (Cronin et al, 2004), math­e­mat­ics (Gross­man, 2002), psy­chol­ogy (Cronin et al, 2003), and eco­nom­ics (M­c­Dow­ell & Melv­in, 1983; Hud­son, 1996; La­band & Tol­lison, 2000). These coau­thor­ship stud­ies show con­sis­tent and, col­lec­tive­ly, gen­eral up­ward trends, with some of the data sets go­ing back as far as 1900.

  • Na­tional Re­search Coun­cil, On Time to the Doc­tor­ate: A Study of the Length­en­ing Time to Com­ple­tion for Doc­tor­ates in Sci­ence and En­gi­neer­ing, Wash­ing­ton, DC: Na­tional Acad­emy Press, 1990
  • Tilgh­man, Shirley (chair) et al. Trends in the Early Ca­reers of Life Sci­ences, Wash­ing­ton, DC: Na­tional Acad­emy Press, 1998
  • Zuck­er­man, Har­riet and Mer­ton, Robert. “Age, Ag­ing, and Age Struc­ture in Sci­ence,” in Mer­ton, Robert, The So­ci­ol­ogy of Sci­ence, Chicago, IL: Uni­ver­sity of Chicago Press, 1973, 497-559
  • Cronin et al, 2004 “Vis­i­ble, Less Vis­i­ble, and In­vis­i­ble Work: Pat­terns of Col­lab­o­ra­tion in 20th Cen­tury Chem­istry,” Jour­nal of the Amer­i­can So­ci­ety for In­for­ma­tion Sci­ence and Tech­nol­ogy, 2004, 55(2), 160-168
  • Gross­man, Jer­ry. “The Evo­lu­tion of the Math­e­mat­i­cal Re­search Col­lab­o­ra­tion Graph,” Con­gres­sus Nu­mer­an­tium, 2002, 158, 202-212
  • Cron­in, Blaise, Shaw, Deb­o­ra, and La Bar­re, Kathryn. “A Cast of Thou­sands: Coau­thor­ship and Sub­au­thor­ship Col­lab­o­ra­tion in the 20th Cen­tury as Man­i­fested in the Schol­arly Jour­nal Lit­er­a­ture of Psy­chol­ogy and Phi­los­o­phy,” Jour­nal of the Amer­i­can So­ci­ety for In­for­ma­tion Sci­ence and Tech­nol­ogy, 2003, 54(9), 855-871
  • Mc­Dow­ell, John, and Melv­in, Michael. “The De­ter­mi­nants of Coau­thor­ship: An Analy­sis of the Eco­nom­ics Lit­er­a­ture,” Re­view of Eco­nom­ics and Sta­tis­tics, Feb­ru­ary 1983, 65, 155-160
  • Hud­son, John. “Trends in Mul­ti­-Au­thored Pa­pers in Eco­nom­ics,” Jour­nal of Eco­nomic Per­spec­tives, Sum­mer 1996, 10, 153-158
  • La­band, David and Tol­lison, Robert. “In­tel­lec­tual Col­lab­o­ra­tion,” Jour­nal of Po­lit­i­cal Econ­omy, June 2000, 108, 632-662

Of fur­ther in­ter­est is the drop in to­tal patent pro­duc­tion per to­tal re­searchers, which has been doc­u­mented across a range of coun­tries and in­dus­tries and may go back as far as 1900 and even be­fore (Machlup 1962). Cer­tain­ly, not all re­searchers are en­gag­ing in patentable ac­tiv­i­ties, and it is pos­si­ble that much of this trend is ex­plained by rel­a­tively rapid growth of re­search in ba­sic sci­ence.22 How­ev­er, the re­sults here in­di­cate that among those spe­cific in­di­vid­u­als who pro­duce patentable in­no­va­tions, the ra­tio of patents to in­di­vid­u­als is in fact de­clin­ing. In par­tic­u­lar, the re­cent drop in patents per U.S. R&D work­er, a drop of about 50% since 1975 (see Segerstrom 1998), is roughly con­sis­tent in mag­ni­tude with the rise in team size over that pe­ri­od.

TODO Per­for­mance Curve Data­base - many sig­moids or lin­ear graphs?


One warn­ing sign is ci­ta­tion by con­tem­po­raries. If the per­cent­age of pa­pers which never get cited in­creas­es, this sug­gests that ei­ther the re­search it­self is no good (even a null re­sult is worth cit­ing as in­for­ma­tion about what we know is not the case), or fel­low re­searchers have a rea­son not to cite them (rea­sons which range from the ma­lign like re­searchers are so over­loaded that they can­not keep up with the lit­er­a­ture, which im­plies di­min­ish­ing mar­ginal re­turns, or pro­fes­sional jeal­ousy, which im­plies the process of sci­ence is be­ing cor­rupt­ed, to the merely pos­si­bly harm­ful, like length lim­it­s). Uncit­ed­ness is most fa­mous in the hu­man­i­ties, which are not nec­es­sar­ily of ma­jor con­cern, but I have col­lected es­ti­mates that there are mul­ti­ple hard fields like chem­istry where uncit­ed­ness may range up to 70+%, and there’s a trou­bling in­di­ca­tion that uncit­ed­ness may be in­creas­ing in even the top sci­en­tific jour­nals.

de­mand for math in­elas­tic and col­lab­o­ra­tion not help­ful?

“The Col­lapse of the So­viet Union and the Pro­duc­tiv­ity of Amer­i­can Math­e­mati­cians”, by George J. Bor­jas and Kirk B. Do­ran, NBER Work­ing Pa­per No. 17800, Feb­ru­ary 2012

We use unique in­ter­na­tional data on the pub­li­ca­tions, ci­ta­tions, and affil­i­a­tions of math­e­mati­cians to ex­am­ine the im­pact of a large post-1992 in­flux of So­viet math­e­mati­cians on the pro­duc­tiv­ity of their Amer­i­can coun­ter­parts. We find a neg­a­tive pro­duc­tiv­ity effect on those math­e­mati­cians whose re­search over­lapped with that the So­vi­ets. We also doc­u­ment an in­creased mo­bil­ity rate (to low­er-qual­ity in­sti­tu­tions and out of ac­tive pub­lish­ing) and a re­duced like­li­hood of pro­duc­ing “home run” pa­pers. Al­though the to­tal prod­uct of the pre-ex­ist­ing Amer­i­can math­e­mati­cians shrank, the So­viet con­tri­bu­tion to Amer­i­can math­e­mat­ics filled in the gap. How­ev­er, there is no ev­i­dence that the So­vi­ets greatly in­creased the size of the “math­e­mat­ics pie.”

Ben Jones, a pro­fes­sor at the Kel­logg School of Man­age­ment, at North­west­ern Uni­ver­si­ty, has quan­ti­fied this trend. By an­a­lyz­ing 19.9 mil­lion peer-re­viewed aca­d­e­mic pa­pers and 2.1 mil­lion patents from the past fifty years, he has shown that lev­els of team­work have in­creased in more than nine­ty-five per cent of sci­en­tific sub­fields; the size of the av­er­age team has in­creased by about twenty per cent each decade. The most fre­quently cited stud­ies in a field used to be the prod­uct of a lone ge­nius, like Ein­stein or Dar­win. To­day, re­gard­less of whether re­searchers are study­ing par­ti­cle physics or hu­man ge­net­ics, sci­ence pa­pers by mul­ti­ple au­thors re­ceive more than twice as many ci­ta­tions as those by in­di­vid­u­als. This trend was even more ap­par­ent when it came to so-called “home­-run pa­pers”-pub­li­ca­tions with at least a hun­dred ci­ta­tions. These were more than six times as likely to come from a team of sci­en­tists.

…A few years ago, Isaac Ko­hane, a re­searcher at Har­vard Med­ical School, pub­lished a study that looked at sci­en­tific re­search con­ducted by groups in an at­tempt to de­ter­mine the effect that phys­i­cal prox­im­ity had on the qual­ity of the re­search. He an­a­lyzed more than thir­ty-five thou­sand peer-re­viewed pa­pers, map­ping the pre­cise lo­ca­tion of co-au­thors. Then he as­sessed the qual­ity of the re­search by count­ing the num­ber of sub­se­quent ci­ta­tions. The task, Ko­hane says, took a “small army of un­der­grad­u­ates” eigh­teen months to com­plete. Once the data was amassed, the cor­re­la­tion be­came clear: when coau­thors were closer to­geth­er, their pa­pers tended to be of sig­nifi­cantly higher qual­i­ty. The best re­search was con­sis­tently pro­duced when sci­en­tists were work­ing within ten me­tres of each oth­er; the least cited pa­pers tended to emerge from col­lab­o­ra­tors who were a kilo­me­tre or more apart. “If you want peo­ple to work to­gether effec­tive­ly, these find­ings re­in­force the need to cre­ate ar­chi­tec­tures that sup­port fre­quent, phys­i­cal, spon­ta­neous in­ter­ac­tions,” Ko­hane says. “Even in the era of big sci­ence, when re­searchers spend so much time on the In­ter­net, it’s still so im­por­tant to cre­ate in­ti­mate spaces.”




Drugs are per­haps the most spec­tac­u­lar ex­am­ple of a sig­moid sud­denly hit­ting and de­stroy­ing a lot of ex­pec­ta­tions. If you read the tran­shu­man­ist lit­er­a­ture from the ’80s or ’90s, even the more sober and well-in­formed pro­jec­tions, it’s strik­ing how much faith is put into ever-new mir­a­cle drugs com­ing out. And why not? still did­n’t seem like it was go­ing too badly - per­haps it would take more than $1 bil­lion, but real progress was be­ing made - and a crop of nootrop­ics came out in the ’70s and ’80s like , fol­lowed up with fa­mous block­busters like Vi­a­gra, and then the Hu­man Genome Project would tri­umphantly fin­ish in a decade or so, where­upon things would re­ally get cook­ing.

What went un­no­ticed in all this was the di­min­ish­ing mar­ginal re­turns. First, it turned out that phar­ma­ceu­ti­cal com­pa­nies were do­ing a pretty good job at search­ing all the ‘small’ (lighter weight) chem­i­cals:

“What Do Med­i­c­i­nal Chemists Ac­tu­ally Make? A 50-Year Ret­ro­spec­tive”

The idea is to sur­vey the field from a longer per­spec­tive than some of the other pa­pers in this vein, and from a wider per­spec­tive than the pa­pers that have looked at mar­keted drugs or struc­tures re­ported as be­ing in the clin­ic. I’m re­pro­duc­ing the plot for the mol­e­c­u­lar weights of the com­pounds, since it’s an im­por­tant mea­sure and rep­re­sen­ta­tive of one of the trends that shows up. The promi­nent line is the plot of mean val­ues, and a blue square shows that the mean for that pe­riod was sta­tis­ti­cally differ­ent than the 5-year pe­riod be­fore it (it’s red if it was­n’t). The lower dashed line is the me­di­an. The dot­ted line, how­ev­er, is the mean for ac­tual launched drugs in each pe­riod with a grey band for the 95% con­fi­dence in­ter­val around it.

In­crease in av­er­age size of drugs 1960-2004

As a whole, the mean mol­e­c­u­lar weight of a J. Med. Chem. has gone up by 25% over the 50-year pe­ri­od, with the steeped in­crease com­ing in 1990-1994. “Why, that was the golden age of com­bichem”, some of you might be say­ing, and so it was. Since that pe­ri­od, though, mol­e­c­u­lar weights have just in­creased a small amount, and may now be lev­el­ing off. Sev­eral other mea­sures show sim­i­lar trends. “Fifty Years of Med-Chem Mol­e­cules: What Are They Telling Us?”

The more atoms in a par­tic­u­lar drug, the more pos­si­ble per­mu­ta­tions and arrange­ments; a log­a­rith­mic slow­down in av­er­age weight would be ex­pected of a search through an ex­po­nen­tially in­creas­ing space of pos­si­bil­i­ties. Smaller drugs are much more de­sir­able than larger ones: they often are eas­ier & cheaper to syn­the­size, more often sur­vive pas­sage through the gut or can pass the blood­-brain bar­ri­er, etc. So there are many more large drugs than small drugs, but the small ones are much more de­sir­able; hence, one would ex­pect a bal­ance be­tween them, with no clear shift - if we were far from ex­ploit­ing all the low-hang­ing fruit. In­stead, we see a very steady trend up­wards, as if there were ever fewer worth­while small drugs to be found.

(One could make an anal­ogy to oil field ex­plo­ration: the big oil fields are eas­i­est to find and also the best, while small ones are both hard to find and the worst; if the big ones ex­ist, the oil com­pa­nies will ex­ploit them as much as pos­si­ble and ne­glect the small ones; hence, a chart show­ing ever de­creas­ing av­er­age size of pro­duc­ing oil fields smells like a strong warn­ing sign that there are few big oil fields left.)

Si­mul­ta­ne­ously with this in­di­ca­tion that good drugs are get­ting harder to find, we find that re­turns are di­min­ish­ing to each dol­lar spent on drug R&D (it takes more dol­lars to pro­duce one drug):

Al­though mod­ern phar­ma­ceu­ti­cals are sup­posed to rep­re­sent the prac­ti­cal pay­off of ba­sic re­search, the R&D to dis­cover a promis­ing new com­pound now costs about 100 times more (in in­fla­tion-ad­justed dol­lars) than it did in 1950. (It also takes nearly three times as long.) This trend shows no sign of let­ting up: In­dus­try fore­casts sug­gest that once fail­ures are taken into ac­count, the av­er­age cost per ap­proved mol­e­cule will top $3.8 bil­lion by 2015. What’s worse, even these “suc­cess­ful” com­pounds don’t seem to be worth the in­vest­ment. Ac­cord­ing to one in­ter­nal es­ti­mate, ap­prox­i­mately 85 per­cent of new pre­scrip­tion drugs ap­proved by Eu­ro­pean reg­u­la­tors pro­vide lit­tle to no new ben­e­fit. We are wit­ness­ing Moore’s law in re­verse.3

The av­er­age drug de­vel­oped by a ma­jor phar­ma­ceu­ti­cal com­pany costs at least $4 bil­lion, and it can be as much as $11 bil­lion…The drug in­dus­try has been toss­ing around the $1 bil­lion num­ber for years. It is based largely on a study (sup­ported by drug com­pa­nies) by Joseph Di­Masi of Tufts Uni­ver­si­ty…But as Bernard Munos of the In­no­Think Cen­ter for Re­search In Bio­med­ical In­no­va­tion has not­ed, just ad­just­ing that es­ti­mate for cur­rent fail­ure rates re­sults in an es­ti­mate of $4 bil­lion in re­search dol­lars spent for every drug that is ap­proved…­Forbes (that would be Scott De­Carlo and me) took Munos’ count of drug ap­provals for the ma­jor phar­mas and com­bined it with their re­search and de­vel­op­ment spend­ing as re­ported in an­nual earn­ings fil­ings go­ing back fifteen years…The range of money spent is stun­ning. As­traZeneca has spent $12 bil­lion in re­search money for every new drug ap­proved, as much as the top-selling med­i­cine ever gen­er­ated in an­nual sales; Am­gen spent just $3.7 bil­lion.4

Kindler’s at­tempts to fig­ure out what to do about re­search were even more an­guished. He was right that the old Pfizer model was­n’t work­ing. Big­ger was­n’t bet­ter when it came to pro­duc­ing new drugs. Stud­ies by Bernard Munos, a re­tired strate­gist at Eli Lilly (LLY), show that both mas­sive in­creases in re­search spend­ing and cor­po­rate merg­ers have failed to in­crease R&D pro­duc­tiv­i­ty. Be­tween 2000 and 2008, ac­cord­ing to Munos, Pfizer spent $60 bil­lion on re­search and gen­er­ated nine drugs that won FDA ap­proval – an av­er­age cost of $6.7 bil­lion per prod­uct. At that rate, Munos con­clud­ed, the com­pa­ny’s in­ter­nal pipeline sim­ply could­n’t sus­tain its profits.5

But the very op­po­site of Moore’s Law is hap­pen­ing at the down­stream end of the R&D pipeline. The num­ber of new mol­e­cules ap­proved per bil­lion dol­lars of in­fla­tion-ad­justed R&D has de­clined in­ex­orably at 9% a year and is now 1/100th of what it was in 1950. The nine biggest drug com­pa­nies spend more than $60 bil­lion a year on R&D but are find­ing new ther­a­pies at such a slow rate that, as a group, they’ve lit­tle chance of re­coup­ing that mon­ey. Mean­while, block­buster drugs are los­ing patent pro­tec­tion at an ac­cel­er­at­ing rate. The next few years will take the in­dus­try over a “patent cliff” of $170 bil­lion in global an­nual rev­enue. On top of this, nat­ural se­lec­tion is pro­duc­ing re­sis­tant dis­ease strains that un­der­mine the effi­cacy not only of ex­ist­ing an­tibi­otics and an­tivi­rals but (even faster) of an­ti-cancer drugs. Many peo­ple be­lieve that some­thing is ter­ri­bly wrong with the way the in­dus­try works. The prob­lem, some think, is that sci­ence-to mix clichés-is scrap­ing the bot­tom of the bi­o­log­i­cal bar­rel after pluck­ing the low-hang­ing fruit.6

“Eroom’s law”: “Di­ag­nos­ing the de­cline in phar­ma­ceu­ti­cal R&D effi­ciency”; Scan­nell et al 2012 (Lowe):

The past 60 years have seen huge ad­vances in many of the sci­en­tific, tech­no­log­i­cal and man­age­r­ial fac­tors that should tend to raise the effi­ciency of com­mer­cial drug re­search and de­vel­op­ment (R&D). Yet the num­ber of new drugs ap­proved per bil­lion US dol­lars spent on R&D has halved roughly every 9 years since 1950, falling around 80-fold in in­fla­tion-ad­justed terms. There have been many pro­posed so­lu­tions to the prob­lem of de­clin­ing R&D effi­cien­cy. How­ev­er, their ap­par­ent lack of im­pact so far and the con­trast be­tween im­prov­ing in­puts and de­clin­ing out­put in terms of the num­ber of new drugs make it sen­si­ble to ask whether the un­der­ly­ing prob­lems have been cor­rectly di­ag­nosed. Here, we dis­cuss four fac­tors that we con­sider to be pri­mary caus­es, which we call the ‘bet­ter than the Bea­t­les’ prob­lem; the ‘cau­tious reg­u­la­tor’ prob­lem; the ‘throw money at it’ ten­den­cy; and the ‘ba­sic re­search-brute force’ bias. Our aim is to pro­voke a more sys­tem­atic analy­sis of the causes of the de­cline in R&D effi­cien­cy.

sim­i­lar graph: http://dl.­drop­box.­com/u/85192141/2008-wobbe.pdf “Fig­ure 3. New drugs dis­cov­ered per bil­lion dol­lars R&D spend­ing and an­nual R&D spend­ing”

“Drug de­vel­op­ment: Raise stan­dards for pre­clin­i­cal can­cer re­search”, C. Glenn Be­g­ley & Lee M. El­lis 2012 - aca­d­e­mic stud­ies ir­re­pro­ducible


Since 2007, real me­dian house­hold in­come has de­clined 6.4% and is 7.1 % be­low the me­dian house­hold in­come peak prior to the 2001 re­ces­sion. Justin Wolfers,

2010 Cen­sus data: > The U.S. Cen­sus Bu­reau an­nounced to­day that in 2010, me­dian house­hold in­come de­clined, the poverty rate in­creased and the per­cent­age with­out health in­sur­ance cov­er­age was not sta­tis­ti­cally differ­ent from the pre­vi­ous year. > Real me­dian house­hold in­come in the United States in 2010 was $49,445, a 2.3 per­cent de­cline from the 2009 me­di­an. > The na­tion’s offi­cial poverty rate in 2010 was 15.1 per­cent, up from 14.3 per­cent in 2009 ─ the third con­sec­u­tive an­nual in­crease in the poverty rate. There were 46.2 mil­lion peo­ple in poverty in 2010, up from 43.6 mil­lion in 2009 ─ the fourth con­sec­u­tive an­nual in­crease and the largest num­ber in the 52 years for which poverty es­ti­mates have been pub­lished. > The num­ber of peo­ple with­out health in­sur­ance cov­er­age rose from 49.0 mil­lion in 2009 to 49.9 mil­lion in 2010, while the per­cent­age with­out cov­er­age −16.3 per­cent - was not sta­tis­ti­cally differ­ent from the rate in 2009.

Un­til this morn­ing, the offi­cial data showed that the U.S. pro­duc­tiv­ity growth ac­cel­er­ated dur­ing the fi­nan­cial cri­sis. Non­farm busi­ness pro­duc­tiv­ity growth sup­pos­edly went from a 1.2% an­nual rate in 2005-2007, to a 2.3% an­nual rate in 2007-2009. Many com­men­ta­tors sug­gested that this pro­duc­tiv­ity gain, in the face of great dis­rup­tions, showed the flex­i­bil­ity of the U.S. econ­o­my.

Uh, oh. The lat­est re­vi­sion of the na­tional in­come ac­counts, re­leased this morn­ing, makes the whole pro­duc­tiv­ity ac­cel­er­a­tion van­ish. Non­farm busi­ness pro­duc­tiv­ity growth in the 2007-09 pe­riod has now been cut al­most in half, down to only 1.4% per year. http­s://in­no­va­tio­nand­growth.­word­press.­com/2011/07/29/pro­duc­tiv­i­ty-surge-of-2007-09-melt­s-away-in-new-data/

I would pay up to $500 per year [for a search en­gine like Google]. It’s that valu­able to me. What about you?

Last year three re­searchers at Uni­ver­sity of Michi­gan per­formed a small ex­per­i­ment to see if they could as­cer­tain how much or­di­nary peo­ple might pay for search. Their method was to ask stu­dents in­side a well-s­tocked uni­ver­sity li­brary to an­swer ques­tions asked on Google, but to find the an­swers only us­ing the ma­te­ri­als in the li­brary. They mea­sured how long it took the stu­dents to an­swer a ques­tion in the stacks. On av­er­age it took 22 min­utes. That’s 15 min­utes longer that the 7 min­utes it took to an­swer the same ques­tion, on av­er­age, us­ing Google. Fig­ur­ing a na­tional av­er­age wage of $22/hour, this works out to a sav­ings of $1.37 per search.­thetech­ni­um/archives/2011/04/­would_y­ou_­pay_f.php

‘A sur­vey in­di­cated that 46 per­cent of Amer­i­cans would be un­will­ing to give up tele­vi­sion for the rest of their lives in re­turn for a mil­lion dol­lars.’ Cowen on: “Would You Give Up TV for a Mil­lion Bucks?” 1992. TV Guide, Oc­to­ber 10, pp. 10-15

How much would some­one have to pay you to give up the In­ter­net for the rest of your life? Would a mil­lion dol­lars be enough? Twenty mil­lion? How about a bil­lion dol­lars? “When I ask my stu­dents this ques­tion, they say you could­n’t pay me enough,” says Pro­fes­sor Michael Cox, di­rec­tor of the O’Neil Cen­ter for Global Mar­kets and Free­dom at South­ern Methodist Uni­ver­si­ty’s Cox School of Busi­ness. http://rea­­s/show/­would-y­ou-give-up-the-in­ter­net

An 86-page 2010 FCC study con­cludes that “a rep­re­sen­ta­tive house­hold would be will­ing to pay about $59 per month for a less re­li­able In­ter­net ser­vice with fast speed (”Ba­sic“), about $85 for a re­li­able In­ter­net ser­vice with fast speed and the pri­or­ity fea­ture (”Pre­mium“), and about $98 for a re­li­able In­ter­net ser­vice with fast speed plus all other ac­tiv­i­ties (”Pre­mium Plus“). An im­prove­ment to very fast speed adds about $3 per month to these es­ti­mates.” http://siepr.s­tan­ford.e­du/sys­tem/­files/shared/­Fi­nal_Rosston_Sav­age_Wald­man_02_04_10__1_.pdf

A study from Japan found that: “The es­ti­mated WTP for avail­abil­ity of e-mail and web brows­ing de­liv­ered over per­sonal com­put­ers are 2,709 Yen ($35) and 2,914 Yen ($38), on a monthly ba­sis, re­spec­tive­ly, while av­er­age broad­band ac­cess ser­vice costs ap­prox­i­mately 4,000 Yen ($52) in Japan”

The Aus­tan Gools­bee pa­per, based on 2005 data, does a time study to find that the con­sumer sur­plus of the in­ter­net is about two per­cent of in­come. http://­fac­ul­ty.chicago­b­ooth.e­du/aus­tan.­gools­bee/re­search/­timeuse.pdf

The pro­duc­tiv­ity of U.S. work­ers dropped from April through June for the sec­ond con­sec­u­tive quar­ter, lead­ing to an in­crease in la­bor costs that may re­strain gains in profits. The mea­sure of em­ployee out­put per hour fell at a 0.3 per­cent an­nual rate in the sec­ond quar­ter after a re­vised 0.6 per­cent drop in the prior three months, fig­ures from the La­bor De­part­ment showed to­day in Wash­ing­ton. The me­dian es­ti­mate of 60 econ­o­mists sur­veyed by Bloomberg News pro­jected a 0.9 per­cent de­crease. Ex­penses per em­ployee climbed at a 2.2 per­cent rate.

…From the sec­ond quar­ter of 2010, pro­duc­tiv­ity climbed 0.8 per­cent com­pared with a 1.2 per­cent year-over-year in­crease in the first quar­ter. La­bor costs rose 1.3 per­cent from the year- ear­lier pe­riod fol­low­ing a 1.1 per­cent in­crease in the 12 months ended in the first quar­ter. To­day’s pro­duc­tiv­ity re­port in­cor­po­rated re­vi­sions to prior years. Worker effi­ciency was re­vised to 4.1 per­cent in 2010 from a pre­vi­ously re­ported 3.9 per­cent. For 2009, it was re­vised down to 2.3 per­cent from 3.7 per­cent. La­bor costs fell 2 per­cent in 2010, the biggest de­cline since records be­gan in 1948. Gross do­mes­tic prod­uct ex­panded at a 1.3 per­cent an­nual pace from April through June, after a 0.4 per­cent rate in the pre­vi­ous three months, the Com­merce De­part­ment said on July 29. House­hold spend­ing rose at 0.1 per­cent pace, the weak­est since the same pe­riod in 2009.­com/news/2011-08-09/pro­duc­tiv­i­ty-in-u-s-fall­s-for-sec­ond-s­traight-quar­ter-as-labor-cost­s-rise.html

Cana­dian man­u­fac­tur­ing & goods pro­duc­tiv­ity stag­nat­ing 2000-2010; re­source ex­trac­tion falling 60% since 1960! http://­mar­gin­al­rev­o­lu­tion.­com/w­p-con­tent/u­pload­s/2011/08/­canada1.png no­tice this is de­spite in­creas­ing to­tal ab­solute ex­trac­tion rate: http://u­pload­.wiki­me­di­­a/­com­mon­s/b/b8/­Cana­di­an_Oil_Pro­duc­tion_1960_­to_2020.png > You can see that Min­ing and Ex­trac­tion TFP takes a long plunge, even though Canada to­day pros­pers through sell­ing nat­ural re­sources. So what’s up? One of Gor­don’s ar­gu­ments against TFP is his claim that this graph im­plies ear­lier min­ing tech­nolo­gies were bet­ter than cur­rent min­ing tech­nolo­gies (un­like­ly), but that is a mis­un­der­stand­ing of what TFP mea­sures. Think of TFP as try­ing to pick “the stuff we get for free through in­no­va­tion.” Falling TFP in min­ing re­flects Canada’s move from “suck it up with a straw” oil to com­plex, high cost ex­trac­tion tar sands projects and the like. They have moved down this curve a long, long way. > > Yet Canada still pros­pers: some­one is will­ing to pay for all the time and trou­ble they put into ex­trac­tion, be­cause the other nat­ural re­source op­tions are cost­lier at the rel­e­vant mar­gin. An­other way to make the point is that this graph, and the em­bed­ded story of pro­duc­tiv­i­ty, is very bad news for some­one, just not Canada, at least not so far. http://­mar­gin­al­rev­o­lu­tion.­com/­mar­gin­al­rev­o­lu­tion/2011/08/is-there-a-pro­duc­tiv­i­ty-cri­sis-in-cana­da.html

The sta­tis­ti­cal trend for growth in to­tal econ­omy LP ranged from 2.75 per­cent in early 1962 down to 1.25 per­cent in late 1979 and re­cov­ered to 2.45 per­cent in 2002. Our re­sults on pro­duc­tiv­ity trends iden­tify a prob­lem in the in­ter­pre­ta­tion of the 2008-09 re­ces­sion and con­clude that at present sta­tis­ti­cal trends can­not be ex­tended past 2007.

For the longer stretch of his­tory back to 1891, the pa­per pro­vides nu­mer­ous cor­rec­tions to the growth of la­bor qual­ity and to cap­i­tal quan­tity and qual­i­ty, lead­ing to sig­nifi­cant re­arrange­ments of the growth pat­tern of MFP, gen­er­ally low­er­ing the un­ad­justed MFP growth rates dur­ing 1928-50 and rais­ing them after 1950. Nev­er­the­less, by far the most rapid MFP growth in U. S. his­tory oc­curred in 1928-50, a phe­nom­e­non that I have pre­vi­ously dubbed the “one big wave.”

Its con­clu­sion is that over the next 20 years (2007-2027) growth in real po­ten­tial GDP will be 2.4 per­cent (the same as in 2000-07), growth in to­tal econ­omy la­bor pro­duc­tiv­ity will be 1.7 per­cent, and growth in the more fa­mil­iar con­cept of NFPB sec­tor la­bor pro­duc­tiv­ity will be 2.05 per­cent. The im­plied fore­cast 1.50 per­cent growth rate of per-capita real GDP falls far short of the his­tor­i­cal achieve­ment of 2.17 per­cent be­tween 1929 and 2007 and rep­re­sents the slow­est growth of the mea­sured Amer­i­can stan­dard of liv­ing over any two-decade in­ter­val recorded since the in­au­gu­ra­tion of George Wash­ing­ton.


Aus­tralia min­ing pro­duc­tiv­ity falling:

“Every­one here also knows that it is now just about im­pos­si­ble to avoid the con­clu­sion that pro­duc­tiv­ity growth per­for­mance has been quite poor since at least the mid-2000s,” he said. Based on the out­put per hours worked, the best pro­duc­tiv­ity per­form­ers over the past five years were in­for­ma­tion, me­dia and telecom­mu­ni­ca­tions (up 6.1 per cent a year on av­er­age), fol­lowed by agri­cul­ture, forestry and fish­ing (up 3.8 per cent) and fi­nan­cial and in­sur­ance ser­vices (up 3.7 per cen­t). In con­trast, min­ing pro­duc­tiv­ity went back­wards by 4.9 per cent per year on av­er­age, and elec­tric­i­ty, gas and waste ser­vices by 5.1 per cent. http://www.theaus­tralian.­­ness/e­co­nom­ic­s/min­ing-drags-down-pro­duc­tiv­i­ty/s­to­ry-e6frg926-1226107755451

Chad Jones (Fig. 1, p. 763, and in his short, read­able text In­tro­duc­tion to Eco­nomic Growth) has re­minded econ­o­mists that the num­ber of sci­en­tists and re­searchers has more than dou­bled in the G-5 coun­tries since 1950, while the growth rate of liv­ing stan­dards has­n’t budged: Twice the re­searchers, zero effect on growth.

The rea­sons for re­tract­ing 742 Eng­lish lan­guage re­search pa­pers re­tracted from the PubMed data­base be­tween 2000 and 2010 were eval­u­at­ed. Rea­sons for re­trac­tion were ini­tially di­chotomised as fraud or er­ror and then analysed to de­ter­mine spe­cific rea­sons for re­trac­tion.

Re­sults: Er­ror was more com­mon than fraud (73.5% of pa­pers were re­tracted for er­ror (or an undis­closed rea­son) vs 26.6% re­tracted for fraud). Eight rea­sons for re­trac­tion were iden­ti­fied; the most com­mon rea­son was sci­en­tific mis­take in 234 pa­pers (31.5%), but 134 pa­pers (18.1%) were re­tracted for am­bigu­ous rea­sons. Fab­ri­ca­tion (in­clud­ing data pla­gia­rism) was more com­mon than text pla­gia­rism. To­tal pa­pers re­tracted per year have in­creased sharply over the decade (r=0.96; p<0.001), as have re­trac­tions specifi­cally for fraud (r=0.89; p<0.001). Jour­nals now reach far­ther back in time to re­tract, both for fraud (r=0.87; p<0.001) and for sci­en­tific mis­takes (r=0.95; p<0.001). Jour­nals often fail to alert the naïve read­er; 31.8% of re­tracted pa­pers were not noted as re­tracted in any way.


http://pm­re­trac­t.heroku.­com/byyear seems to fol­low an ex­po­nen­tial in­crease in re­trac­tion per­cent­age from 2000-2010

Be­low is a fig­ure con­structed us­ing the quar­terly TFP [to­tal fac­tor pro­duc­tiv­i­ty] se­ries of John Fer­nald at the San Fran­cisco Fed. (ex­treme di­ver­gence from the ex­po­nen­tial growth, around 1970-1973, to some­thing that looks lin­ear - with no ac­cel­er­a­tion in the ’90s or 2000s)

Us­ing coun­try-level analy­sis as a base, we es­ti­mated that the to­tal gross value of In­ter­net search across the global econ­omy was $780 bil­lion in 2009, equiv­a­lent to the GDP of the Nether­lands or Turkey. By this es­ti­mate, each search is worth about $0.50. Of that val­ue, $540 bil­lion-69 per­cent of the to­tal and 25 times the an­nual value added (profits) of search com­pa­nies-flowed di­rectly to global GDP, chiefly in the form of e-com­merce, ad­ver­tis­ing rev­enues, and higher cor­po­rate pro­duc­tiv­i­ty. Search ac­counted for 1.2 per­cent of US and for 0.5 per­cent of In­di­a’s GDP. The re­main­ing $240 bil­lion (31 per­cent) does not show up in GDP sta­tis­tics. It is cap­tured by in­di­vid­u­als rather than com­pa­nies, in the form of con­sumer sur­plus, and arises from un­mea­sured ben­e­fits, such as lower prices, con­ve­nience, and the time saved by swift ac­cess to in­for­ma­tion. We es­ti­mate those ben­e­fits at $20 a month for con­sumers in France, Ger­many, and the United States and at $2 to $5 a month for their coun­ter­parts in Brazil and In­dia. http­s://www.m­ck­in­seyquar­ter­ly.­com/­Mar­ket­ing/Dig­i­tal_­Mar­ket­ing/Mea­sur­ing_the_­val­ue_of_search_2848

  • A typ­i­cal In­ter­net search for aca­d­e­mic in­for­ma­tion takes seven min­utes. Re­ly­ing on phys­i­cal ref­er­ences takes 22 min­utes.44
  • A con­sumer gen­er­ally finds time to per­form ten searches on­line but only two searches offline for each pur­chase.45
  • It takes the same amount of time to do three searches in an on­line busi­ness di­rec­tory as it does to do one in a phys­i­cal di­rec­to­ry.46

Analy­sis for this re­port sug­gests that knowl­edge work­ers in busi­ness each save 30 to 45 hours per year as a re­sult of search. When it comes to price trans­paren­cy, aca­d­e­mic re­search shows that the more vis­its made to price com­par­i­son Web sites, the lower prices fall and the greater the differ­ence be­tween the av­er­age and min­i­mum price for a par­tic­u­lar good.77 Thus, price trans­parency has a dis­ci­plin­ing effect on the mar­gins re­tail­ers can ex­pect, which ben­e­fits con­sumers. Pre­lim­i­nary re­search shows prices on­line are, on av­er­age, 10 per­cent lower than those offline as a re­sult of the price trans­parency afforded by search tool­s.78 Bet­ter match­ing is par­tic­u­larly valu­able to con­sumers when they want long-tail items. Re­search shows that con­sumers value a hard-to-find, long-tail prod­uct any­where be­tween 1.3 to 1.8 times the ac­tual price of the prod­uc­t.79 Con­sumers there­fore cap­ture sig­nifi­cant amounts of sur­plus when they buy prod­ucts in the long tail. With re­gard to time saved, var­i­ous stud­ies taken to­gether sug­gest that con­sumers who search on­line for their pur­chase can save 10 to 20 hours a year.80 Us­ing data from aca­d­e­mic stud­ies, we val­ued that time at be­tween $0.5 and $7 per hour, based on av­er­age, after-tax in­come per house­hold in each coun­try and the as­sump­tion that a con­sumer’s leisure time was worth 65 per­cent of this fig­ure.81,82

It seems like mar­ket fore­casts of low real yields 30 years into the fu­ture sup­port TGS. How long does it take for long-run money neu­tral­ity to win out? If the yield curve showed low yields 100 years out, would that dis­suade those look­ing for a mon­e­tary so­lu­tion? http://­mar­gin­al­rev­o­lu­tion.­com/­mar­gin­al­rev­o­lu­tion/2011/08/­cap­i­tal-de­pre­ci­a­tion-as-s­tim­u­lus.htm­l#­com­ments yields: http://www.trea­sury.­gov­/re­source-cen­ter/­data-chart-cen­ter/in­ter­est-rates/­Pages/­­px?­data=re­alyield

To­day, very few teenagers work full time jobs, and the num­ber of teens em­ployed in sum­mer jobs has de­creased from ~60% in 1994, to ~40% in 2008.[29]

[29]: Ca­marota S, Jense­nius, K. : “A Drought of Sum­mer Jobs: Im­mi­gra­tion and the Long-Term De­cline in Em­ploy­ment Among U.S.-Born Teenagers”. In: Back­grounder. Cen­ter for Im­mi­gra­tion Stud­ies; 2010.


Here, we see that the per­cent­age rep­re­sen­ta­tion of teens in the U.S. work­force in 2010 is 5.1% less than the level recorded in 2002. That fig­ure con­firms that teens are in­deed be­ing dis­placed from the U.S. work­force at the min­i­mum wage lev­el….In prac­ti­cal terms, for the 5.1% per­cent­age de­cline from 2002 through 2010 in the teen share of Amer­i­can fed­eral min­i­mum wage earn­ers, ap­prox­i­mate half were dis­placed by young adults Age 20-34 (2.7%), while the re­main­der were dis­placed by geezers Age 45-59 (2.4%). http://po­lit­i­cal­cal­cu­la­tion­s.blogspot.­com/2011/07/how-much-are-geezer­s-dis­plac­ing-teen­s.html based on “the Bu­reau of La­bor Sta­tis­tics’ an­nual re­ports on the Char­ac­ter­is­tics of Min­i­mum Wage Work­ers.” see also http://po­lit­i­cal­cal­cu­la­tion­s.blogspot.­com/2011/07/dis­ap­pear­ing-teen-job­s-and-min­i­mum-wage_14.html

This pa­per out­lines a sim­ple re­gres­sion-based method to de­com­pose the vari­ance of an ag­gre­gate time se­ries into the vari­ance of its com­po­nents, which is then ap­plied to mea­sure the rel­a­tive con­tri­bu­tions of pro­duc­tiv­i­ty, hours per work­er, and em­ploy­ment to cycli­cal out­put growth across a panel of coun­tries. Mea­sured pro­duc­tiv­ity con­tributes more to the cy­cle in Eu­rope and Japan than in the United States. Em­ploy­ment con­tributes the largest pro­por­tion of the cy­cle in Eu­rope and the United States (but not Japan), which is in­con­sis­tent with the idea that higher lev­els of em­ploy­ment pro­tec­tion in Eu­rope dampen cycli­cal em­ploy­ment fluc­tu­a­tion­s…In the United States, pro­duc­tiv­ity only con­tributes about 27% of the cy­cle and la­bor in­put four-fifths. Mean­while, in France and Ger­many, pro­duc­tiv­ity con­tributes 43% and 38% of the cy­cle, re­spec­tive­ly. Japan is more Eu­ro­pean than Eu­rope in this re­gard; pro­duc­tiv­ity con­tributes 59% of the cy­cle there, while Ko­rea looks more like the United States. I’m the au­thor of that pa­per. My own in­ter­pre­ta­tion of my pa­per is that Japan sees a lot more la­bor hoard­ing than Eu­rope or the United States, so un­em­ploy­ment is a par­tic­u­larly bad mea­sure of the cy­cle there. We’d want to look at some mea­sure of the out­put gap as a cycli­cal in­di­ca­tor since la­bor mar­ket in­di­ca­tors from Japan don’t carry much in­for­ma­tion about the macro sit­u­a­tion. BUT, Karl has a ma­jor point, which is what the sec­ond quote was about. If we look at out­put, our analy­sis is com­pli­cated by the fact that the trends which we saw through 1990 or so-con­ver­gence in pro­duc­tiv­ity and un­usu­ally high hours worked per work­er-have stopped. With­out putting words in Tyler’s mouth, Japan picked the low-hang­ing fruit and now its pro­duc­tiv­ity has been at 70% of that of the United States for some time now. We can’t just naively ex­trap­o­late that trend and ex­pect a large amount of growth. Com­bine that with low pop­u­la­tion growth and a sharp down­ward trend in hours worked, and the Japan­ese growth slow­down since then is not sur­pris­ing. http://­mar­gin­al­rev­o­lu­tion.­com/­mar­gin­al­rev­o­lu­tion/2011/08/where-does-the-japan­ese-s­low­down-come-from.htm­l#­com­men­t-body-157487734

The need for fur­ther pro­duc­tiv­ity gains does­n’t re­ally make sense to me as an ex­pla­na­tion. Japan has low hang­ing pro­duc­tiv­ity fruit out the wa­zoo. The stereo­typ­i­cal salary­man stays out late “work­ing” every night. Send the same dude home at 5:00pm and he’d get just as much done and in­crease pro­duc­tiv­ity by 4 hours a day, eas­i­ly. Or is the sug­ges­tion that Japan­ese cul­ture is too re­sis­tant to this kind of change, hence pro­duc­tiv­ity could­n’t grow, hence it got hit by TGS? http://­mar­gin­al­rev­o­lu­tion.­com/­mar­gin­al­rev­o­lu­tion/2011/08/where-does-the-japan­ese-s­low­down-come-from.htm­l#­com­men­t-157487642

The pa­per mea­sures pro­duc­tiv­ity growth in sev­en­teen coun­tries in the nine­teenth and twen­ti­eth cen­turies. GDP per worker and cap­i­tal per worker in 1985 US dol­lars were es­ti­mated for 1820, 1850, 1880, 1913, and 1939 by us­ing his­tor­i­cal na­tional ac­counts to back cast Penn World Ta­ble data for 1965 and 1990. Fron­tier and econo­met­ric pro­duc­tion func­tions are used to mea­sure neu­tral tech­ni­cal change and lo­cal tech­ni­cal change. The lat­ter in­cludes con­cur­rent in­creases in cap­i­tal per worker and out­put per worker be­yond the high­est val­ues achieved. These in­creases were pi­o­neered by the rich coun­tries of the day. An in­crease in the cap­i­tal-la­bor ra­tio was usu­ally fol­lowed by a half cen­tury in which rich coun­tries raised out­put per worker at that higher ra­tio. Then the rich coun­tries moved on to a higher cap­i­tal-ra­tio, and tech­ni­cal progress ceased at the lower ra­tio they aban­doned. Most of the ben­e­fits of tech­ni­cal progress ac­crued to the rich coun­tries that pi­o­neered it. It is re­mark­able that coun­tries in 1990 with low cap­i­tal la­bor ra­tios achieved an out­put per worker that was no higher than coun­tries with the same cap­i­tal la­bor ra­tio in 1820. In the course of the last two hun­dred years, the rich coun­tries cre­ated the pro­duc­tion func­tion of the world that de­fines the growth pos­si­bil­i­ties of poor coun­tries to­day.

“Tech­nol­ogy and the great di­ver­gence: Global eco­nomic de­vel­op­ment since 1820”, Allen 2011



Wik­ileaks diplo­matic ca­ble http://www.ze­ro­hedge.­com/news/wik­ileak­s-ca­ble-re­veal­s-chi­ne­se-warn­ing-do­mes­tic-as­set-bub­bles-over­ca­pac­i­ty-ear­ly-2010-bash­ing-

China will need to re­struc­ture its econ­omy so that it has a sig­nifi­cantly higher share of knowl­edge-based ser­vices, es­pe­cially re­search and de­vel­op­ment. How­ever Chi­na’s “ter­ri­ble” ed­u­ca­tional sys­tem, which pro­motes copy­ing and past­ing over cre­ative and in­de­pen­dent thought, is the largest im­ped­i­ment the coun­try faces on this front, our IFC con­tact said….

\1\1. (SBU) How­ev­er, Lai [Con­sul Gen­er­al, the head of IFC’s Chengdu office, Lai Jin­chang] iden­ti­fied Chi­na’s “ter­ri­ble” ed­u­ca­tional sys­tem as pre­sent­ing a se­ri­ous im­ped­i­ment to­ward achiev­ing a shift to a more knowl­edge-based econ­o­my. The cur­rent sys­tem pro­motes copy­ing and past­ing over cre­ative and in­de­pen­dent thought. Lai said that the sys­tem re­wards stu­dents for think­ing “within a frame­work” in or­der to get the grade. He de­scribed the nor­mal process un­der­taken by stu­dents when writ­ing as es­sen­tially col­lect­ing sen­tences from var­i­ous sources with­out any orig­i­nal think­ing. He com­pared the writ­ing abil­ity of a typ­i­cal Chi­nese Phd as pal­ing in com­par­i­son to his “un­skilled” staff dur­ing his decade of work with the IFC in Africa.


stronger in Chi­na? , 2005:

We tar­geted 13 gene-dis­ease as­so­ci­a­tions, each al­ready as­sessed by meta-analy­ses, in­clud­ing at least 15 non-Chi­nese stud­ies. We searched the Chi­nese Jour­nal Ful­l-Text Data­base for ad­di­tional Chi­nese stud­ies on the same top­ics. We iden­ti­fied 161 Chi­nese stud­ies on 12 of these gene-dis­ease as­so­ci­a­tions; only 20 were Pub­Med-in­dexed (seven Eng­lish ful­l-tex­t). Many stud­ies (14-35 per top­ic) were avail­able for six top­ics, cov­er­ing dis­eases com­mon in Chi­na. With one ex­cep­tion, the first Chi­nese study ap­peared with a time lag (2-21 y) after the first non-Chi­nese study on the top­ic. Chi­nese stud­ies showed sig­nifi­cantly more promi­nent ge­netic effects than non-Chi­nese stud­ies, and 48% were sta­tis­ti­cally sig­nifi­cant per se, de­spite their smaller sam­ple size (me­dian sam­ple size 146 ver­sus 268, p< 0.001). The largest ge­netic effects were often seen in Pub­Med-in­dexed Chi­nese stud­ies (65% sta­tis­ti­cally sig­nifi­cant per se). Non-Chi­nese stud­ies of Asian-de­s­cent pop­u­la­tions (27% sig­nifi­cant per se) also tended to show some­what more promi­nent ge­netic effects than stud­ies of non-Asian de­scent (17% sig­nifi­cant per se).

“Chi­nese In­no­va­tion Is a Pa­per Tiger: A closer look at Chi­na’s patent fil­ings and R&D spend­ing re­veals a coun­try that has a long way to go”, WSJ, by Anil K. Gupta and Haiyan Wang:

Chi­na’s R&D ex­pen­di­ture in­creased to 1.5% of GDP in 2010 from 1.1% in 2002, and should reach 2.5% by 2020. Its share of the world’s to­tal R&D ex­pen­di­ture, 12.3% in 2010, was sec­ond only to the U.S., whose share re­mained steady at 34%-35%. Ac­cord­ing to the World In­tel­lec­tual Prop­erty Or­ga­ni­za­tion, Chi­nese in­ven­tors filed 203,481 patent ap­pli­ca­tions in 2008. That would make China the third most in­no­v­a­tive coun­try after Japan (502,054 fil­ings) and the U.S. (400,769)…Ac­cord­ing to the Or­ga­ni­za­tion for Eco­nomic Co­op­er­a­tion and De­vel­op­ment, in 2008, the most re­cent year for which data are avail­able, there were only 473 tri­adic patent fil­ings from China ver­sus 14,399 from the U.S., 14,525 from Eu­rope, and 13,446 from Japan. Starkly put, in 2010 China ac­counted for 20% of the world’s pop­u­la­tion, 9% of the world’s GDP, 12% of the world’s R&D ex­pen­di­ture, but only 1% of the patent fil­ings with or patents granted by any of the lead­ing patent offices out­side Chi­na. Fur­ther, half of the Chi­na-o­ri­gin patents were granted to sub­sidiaries of for­eign multi­na­tion­al­s….A 2009 sur­vey by the China As­so­ci­a­tion for Sci­ence and Tech­nol­ogy re­ported that half of the 30,078 re­spon­dents knew at least one col­league who had com­mit­ted aca­d­e­mic fraud. Such a cul­ture in­hibits se­ri­ous in­quiry and wastes re­sources.

In­sid­ers agree; a dean at Ts­inghua Uni­ver­sity (first or sec­ond best uni­ver­sity in Chi­na):

In re­al­i­ty, how­ev­er, ram­pant prob­lems in re­search fund­ing-some at­trib­ut­able to the sys­tem and oth­ers cul­tur­al-are slow­ing down Chi­na’s po­ten­tial pace of in­no­va­tion.

Al­though sci­en­tific merit may still be the key to the suc­cess of smaller re­search grants, such as those from Chi­na’s Na­tional Nat­ural Sci­ence Foun­da­tion, it is much less rel­e­vant for the megapro­ject grants from var­i­ous gov­ern­ment fund­ing agen­cies…the guide­lines are often so nar­rowly de­scribed that they leave lit­tle doubt that the “needs” are any­thing but na­tion­al; in­stead, the in­tended re­cip­i­ents are ob­vi­ous. Com­mit­tees ap­pointed by bu­reau­crats in the fund­ing agen­cies de­ter­mine these an­nual guide­lines. For ob­vi­ous rea­sons, the chairs of the com­mit­tees often lis­ten to and usu­ally co­op­er­ate with the bu­reau­crats.

“Ex­pert opin­ions” sim­ply re­flect a mu­tual un­der­stand­ing be­tween a very small group of bu­reau­crats and their fa­vorite sci­en­tists. This top-down ap­proach sti­fles in­no­va­tion and makes clear to every­one that the con­nec­tions with bu­reau­crats and a few pow­er­ful sci­en­tists are para­mount, dic­tat­ing the en­tire process of guide­line prepa­ra­tion. To ob­tain ma­jor grants in Chi­na, it is an open se­cret that do­ing good re­search is not as im­por­tant as schmooz­ing with pow­er­ful bu­reau­crats and their fa­vorite ex­perts.

This prob­lem­atic fund­ing sys­tem is fre­quently ridiculed by the ma­jor­ity of Chi­nese re­searchers. And yet it is al­so, para­dox­i­cal­ly, ac­cepted by most of them. Some be­lieve that there is no choice but to ac­cept these con­ven­tions. This cul­ture even per­me­ates the minds of those who are new re­turnees from abroad; they quickly adapt to the lo­cal en­vi­ron­ment and per­pet­u­ate the un­healthy cul­ture. A sig­nifi­cant pro­por­tion of re­searchers in China spend too much time on build­ing con­nec­tions and not enough time at­tend­ing sem­i­nars, dis­cussing sci­ence, do­ing re­search, or train­ing stu­dents (in­stead, us­ing them as la­bor­ers in their lab­o­ra­to­ries). Most are too busy to be found in their own in­sti­tu­tions. Some be­come part of the prob­lem: They use con­nec­tions to judge grant ap­pli­cants and un­der­value sci­en­tific mer­it. ed­i­to­ri­al, “Chi­na’s Re­search Cul­ture”, Sci­ence

An in­ves­ti­ga­tion by the Chi­nese As­so­ci­a­tion of Sci­en­tists has re­vealed that only about 40 per­cent of the funds al­lo­cated for sci­en­tific re­search is used on the projects they are meant for. The rest is usu­ally spent on things that have noth­ing to do with re­search. Some re­search project lead­ers use the money to buy fur­ni­ture, home ap­pli­ances and, hold your breath, even apart­ments. In the most ap­palling scan­dal, an ac­coun­tant in the Na­tional Sci­ence Foun­da­tion of China mis­ap­pro­pri­ated more than 200 mil­lion yuan ($3.12 mil­lion) in eight years un­til he was ar­rested in 2004. Be­sides, the de­gree of earnest­ness most sci­en­tists show in their re­search projects nowa­days is ques­tion­able. En­gag­ing in sci­en­tific re­search projects funded by the State has turned out to be an op­por­tu­nity for some sci­en­tists to make mon­ey. There are ex­am­ples of some sci­en­tists get­ting re­search funds be­cause of their con­nec­tions with offi­cials rather than their in­no­va­tion ca­pac­i­ty. Qian Xue­sen, known as the fa­ther of Chi­na’s atomic bomb and satel­lites, used to say dur­ing the last few years be­fore his death in 2009 that the biggest prob­lem is that Chi­nese uni­ver­si­ties can­not cul­ti­vate top-class sci­en­tists. “Hon­est Re­search Needed”, China Daily (gov­ern­ment pa­per)

Zinch Chi­na, a con­sult­ing com­pany that ad­vises Amer­i­can col­leges and uni­ver­si­ties about Chi­na, pub­lished a re­port last year that found cheat­ing on col­lege ap­pli­ca­tions to be “per­va­sive in Chi­na, dri­ven by hy­per­-com­pet­i­tive par­ents and ag­gres­sive agents.”

From the sur­vey’s in­tro­duc­tion: “Our re­search in­di­cates that 90 per­cent of rec­om­men­da­tion let­ters are fake, 70 per­cent of es­says are not writ­ten by the ap­pli­cant, and 50 per­cent of high school tran­scripts are fal­si­fied.”


But there’s grow­ing ev­i­dence that the in­no­va­tion short­fall of the past decade is not only real but may also have con­tributed to to­day’s fi­nan­cial cri­sis. Think back to 1998, the early days of the dot-com bub­ble. At the time, the news was filled with re­ports of star­tling break­throughs in sci­ence and med­i­cine, from new can­cer treat­ments and gene ther­a­pies that promised to cure in­tractable dis­eases to high­-speed satel­lite In­ter­net, cars pow­ered by fuel cells, mi­cro­ma­chines on chips, and even cloning. These tech­nolo­gies seemed to be com­mer­cial­iz­ing at “In­ter­net speed,” cre­at­ing com­pa­nies and draw­ing in enor­mous in­vest­ments from profit-seek­ing ven­ture cap­i­tal­ist­s-and or­di­nar­ily cau­tious cor­po­rate gi­ants. Fed­eral Re­serve Chair­man Alan Greenspan summed it up in a 2000 speech: “We ap­pear to be in the midst of a pe­riod of rapid in­no­va­tion that is bring­ing with it sub­stan­tial and last­ing ben­e­fits to our econ­o­my.” With the hind­sight of a decade, one thing is abun­dantly clear: The com­mer­cial im­pact of most of those break­throughs fell far short of ex­pec­ta­tion­s-not just in the U.S. but around the world. No gene ther­apy has yet been ap­proved for sale in the U.S. Rural dwellers can get satel­lite In­ter­net, but it’s far slow­er, with longer lag times, than the am­bi­tious satel­lite ser­vices that were be­ing de­vel­oped a decade ago. The eco­nom­ics of al­ter­na­tive en­ergy haven’t changed much. And while the biotech in­dus­try has con­tin­ued to grow and pro­duce im­por­tant drugs-such as Avastin and Gleevec, which are used to fight can­cer-the gains in health as a whole have been dis­ap­point­ing, given the enor­mous sums in­vested in re­search. As Gary P. Pisano, a Har­vard Busi­ness School ex­pert on the biotech busi­ness, ob­serves: “It was a much harder road com­mer­cially than any­one be­lieved.”…With far fewer break­through prod­ucts than ex­pect­ed, Amer­i­cans had lit­tle new to sell to the rest of the world. Ex­ports stag­nat­ed, stuck at around 11% of gross do­mes­tic prod­uct un­til 2006, while im­ports soared. That forced the U.S. to bor­row tril­lions of dol­lars from over­seas. The same surges of im­ports and bor­row­ing also dis­torted eco­nomic sta­tis­tics so that growth from 1998 to 2007, rather than av­er­ag­ing 2.7% per year, may have been closer to 2.3% per year.

…Even the se­quenc­ing of the hu­man genome-an ac­claimed sci­en­tific achieve­men­t-has not re­duced the cost of de­vel­op­ing profitable drugs. One in­di­ca­tor of the prob­lem’s scope: 2008 was the first year that the U.S. biotech in­dus­try col­lec­tively made a profit, ac­cord­ing to a re­cent re­port by Ernst & Young-and that per­for­mance is not ex­pected to be re­peated in 2009.

…If an in­no­va­tion boom were truly hap­pen­ing, it would likely push up stock prices for com­pa­nies in such lead­ing-edge sec­tors as phar­ma­ceu­ti­cals and in­for­ma­tion tech­nol­o­gy. In­stead, the stock in­dex that tracks the phar­ma­ceu­ti­cal, biotech, and life sci­ences com­pa­nies in the Stan­dard & Poor’s (MHP) 500-s­tock in­dex dropped 32% from the end of 1998 to the end of 2007, after ad­just­ing for in­fla­tion. The in­for­ma­tion tech­nol­ogy in­dex fell 29%. To pick out two ma­jor com­pa­nies: The stock price of Merck de­clined 35% be­tween the end of 1998 and the end of 2007, after ad­just­ing for in­fla­tion, while the stock price of Cisco Sys­tems (CSCO) was down 9%. Con­sider an­other in­di­ca­tor of com­mer­cially im­por­tant in­no­va­tion: the trade bal­ance in ad­vanced tech­nol­ogy prod­ucts. The Cen­sus Bu­reau tracks im­ports and ex­ports of goods in 10 high­-tech ar­eas, in­clud­ing life sci­ences, biotech, ad­vanced ma­te­ri­als, and aero­space. In 1998 the U.S. had a $30 bil­lion trade sur­plus in these ad­vanced tech­nol­ogy prod­ucts; by 2007 that had flipped to a $53 bil­lion deficit. Sur­pris­ing­ly, the U.S. was run­ning a trade deficit in life sci­ences, an area where it is sup­posed to be a leader.

…The fi­nal clue: the ag­o­niz­ingly slow im­prove­ment in death rates by age, de­spite all the money thrown into health-care re­search. Yes, ad­vances in health care can affect the qual­ity of life, but one would ex­pect any big in­no­va­tion in med­ical care to re­sult in a faster de­cline in the death rate as well. The offi­cial death-rate stats offer a mixed but mostly dis­ap­point­ing pic­ture of how med­ical in­no­va­tion has pro­gressed since 1998. On the plus side, Amer­i­cans 65 and over saw a faster de­cline in their death rate com­pared with pre­vi­ous decades. The bad news: Most age groups un­der 65 saw a slower fall in the death rate. For ex­am­ple, for chil­dren ages 1 to 4, the death rate fell at a 2.3% an­nual pace be­tween 1998 and 2006, com­pared with a 4% de­cline in the pre­vi­ous decade. And sur­pris­ing­ly, the death rate for peo­ple in the 45-to-54 age group was slightly higher in 2006 than in 1998.

http://www.busi­ness­week.­com/­magazine/­con­tent/09_24/b4135000953288.htm Man­del; the point about stock mar­ket is in­ter­est­ing, be­cause a de­fender of no-de­clines could say that any failed pre­dic­tions of in­no­va­tions - like the ones sur­round­ing the Hu­man Genome Project - have been cher­ry-picked by de­clin­ists, but the stock mar­ket should be im­mune to such cher­ry-pick­ing. Yet, it was­n’t.

Howard 2001 Search­ing the Real World for Signs of Ris­ing Pop­u­la­tion In­tel­li­gence:

Howard (1999) looked at chess per­for­mance since the in­au­gural FIDE (in­ter­na­tional chess fed­er­a­tion) rat­ing list in 1970. The list is based on an ob­jec­tive mea­sure of each play­er’s chess per­for­mance, on a scale from about 2000 to 2800. The rat­ing changes with each game played, de­pend­ing on re­sult and op­po­nen­t’s strength, and thus re­flects cur­rent prowess….S­ince 1970, play­ers were reach­ing high per­for­mance lev­els at pro­gres­sively ear­lier ages. For ex­am­ple, the me­dian age of the top ten dropped from the late 30s in the 1970s to the mid-20s in the 1990s. Ev­i­dence dis­cussed in de­tail sug­gested that the trend was due to ris­ing in­tel­li­gence.

  • Howard, R. W. (1999). “Pre­lim­i­nary re­al-world ev­i­dence that av­er­age hu­man in­tel­li­gence re­ally is ris­ing”. In­tel­li­gence, 27, 235±250

…How­ev­er, in the So­viet Union where chess was the na­tional sport, this had been oc­cur­ring since the 1920s (Char­ness & Ger­chak, 1996). If g was not ris­ing, the age trend should have started much ear­li­er.

  • Char­ness, N., & Ger­chak, Y. (1996). “Par­tic­i­pa­tion rates and max­i­mal per­for­mance”. Psy­cho­log­i­cal Sci­ence, 7, 46±51

…An in­for­mal study by Nunn (1999) sup­ports the view. Us­ing the com­puter pro­gram Fritz’s blundercheck mode, which scans games for se­ri­ous er­rors, he com- pared the stan­dard of play in two ma­jor tour­na­ments across the cen­tu­ry; Carls­bad, 1911 and the Biel In­ter­zon­al, 1993. Both had many of their er­a’s best play­ers. Per­for­mance was much bet­ter in 1993, play­ers mak­ing many fewer se­ri­ous er­rors. Nunn con­cluded that the 1911 tour­na­ment would be con­sid­ered very weak to­day.

Howard (1999) noted that, since 1970, chess has had an in­creas­ing num­ber of prodi­gies (chess gifted chil­dren), de­spite fewer young­sters in the ag­ing West­ern pop­u­la­tion. Some pre-1970 data re­lat­ing to the Chess Olympiad and the pres­ti­gious in­ter­na­tional grand­mas­ter ti­tle were ob­tained. The ti­tle it­self dates back to 1914 but only in 1950 did FIDE offi­cially award it. Ta­ble 1 shows the age records for gain­ing the grand­mas­ter ti­tle from 1950, ei­ther a play­er’s ex­act age when re­ceiv­ing the ti­tle (if known) or age on July 1 of the year re­ceiv­ing it. The ta­ble shows the record be­ing bro­ken sev­eral times in the 1950s, but the 1957 record stood un­til 1991, and there­after was re­peat­edly bro­ken. The record set­ters in the 1950s were ex­cep­tion­ally tal­ent­ed, all ex­cept Bron­stein be­com­ing world cham­pi­on.

The same age record de­crease has oc­curred with an­other pres­ti­gious per­for­mance-based ti­tle, the US Chess Fed­er­a­tion (USCF) mas­ter ti­tle. The age record has been bro­ken sev­eral times re­cent­ly, ex­tremely young play­ers gain­ing the ti­tle. In the last few years, some record-hold­ers have been; Jordy Rey­naud aged 10 years, 7 months; Vinay Bhat 10 years, 6 months, and in 1998 Hikaru Naka­mura at 10 years, 2 months, only about 29 months after learn­ing to play. How­ev­er, efforts to gain lon­gi­tu­di­nal data on this ti­tle from the USCF were un­suc­cess­ful. Par­en­thet­i­cal­ly, in 1998, Irina Krush set an­other age record by win­ning the US Wom­en’s Cham­pi­onship aged only 14 years.

  • Nunn, J. (1999). John Nun­n’s chess puz­zle book. Lon­don: Gam­bit Pub­li­ca­tions.

…Fran­cis, Fran­cis and Tr­uscott (1994) pro­vide data on play­ers and tour­na­ment re­sults. Some ad­di­tional data were ob­tained from bridge fed­er­a­tions. Ta­ble 1 presents age records for the US Con­tract Bridge League life mas­ter ti­tle. There seem to be many more bridge prodi­gies with time, the age record steadily drop­ping in bridge as in chess. The present record holder re­port­edly only be­gan play­ing bridge the year be­fore. It is in­ter­est­ing to note that the USCF chess mas­ter and US bridge mas­ter age records have de­creased to about the same age.

  • Fran­cis, H. G., Fran­cis, D. A., & Tr­uscott, A. E. (1994). The offi­cial en­cy­clo­pe­dia of bridge (5th ed.). Mem­phis, TN: Amer­i­can Con­tract Bridge League

…Fig. 2 presents me­dian age of the play­ers in the World Open Cham­pi­onship ti­tles (con­sist­ing of two player team­s). All ages are as at the age on Jan­u­ary 1 of the year con­sid­ered, as most birth dates avail­able listed only year. The event was held every 2 years. Be­cause of the small sam­ples, data are me­dian age of all play­ers on the win­ning teams for each decade. The trend partly par­al­lels the trend for chess grand­mas­ters, go­ing down­wards from the 1960s, but then it rises in the 1990s.

Fig. 3 gives me­dian age of the six play­ers in each win­ning team in the Bridge Olympiad, held every 4 years since 1960. The me­dian age in­creased from 1960 to 1972, then de­clined and then rose from 1982. Clear­ly, top Bridge Olympiad play­ers are not get­ting pro­gres­sively younger, with play­ers be­ing dis­placed by younger, stronger play­ers. The trend up­wards from 1964 oc­curred be­cause the ex­act same French team won three times in a row.

…How­ev­er, go has a ma­jor prob­lem­at­i­cal as­pect for the present study. Un­like chess and bridge, there are great bar­ri­ers to en­try at up­per lev­els. Play­ers gen­er­ally must start train­ing in el­e­men­tary school and must serve a lengthy ap­pren­tice­ship with a top play­er. They only are ad­mit­ted to the ranks of pro­fes­sion­als and to dan lev­els by vote of other pro­fes­sion­als (Bozulich, 1992). This sys­tem favours the pre-em­i­nence of older es­tab­lished play­ers who could keep out young, tal­ented play­ers. The time re­quired and diffi­culty of ris­ing may dis­cour­age great tal­ent.

…These are the pres­ti­gious Ki­sei, Ten­gen, Mei­jin, Hon­in­bo, Ju­dan, Go­sei, and Oza ti­tles. Most were first awarded in the 1950s. The com­pe­ti­tions for each ti­tle usu­ally are held every year and the win­ner is de­ter­mined by a se­ries of match­es. Is the age of ti­tle win­ners drop­ping? Fig. 2 gives the me­dian age of all ti­tle win­ners com­bined (“go: all”) and of first-time win­ners (“go: unique”) of each ti­tle in each decade. The age trend partly par­al­lels that for chess and bridge, de­creas­ing from the 1960s to the 1970s but then ris­ing some­what. How­ev­er, the unique go ti­tle win­ners in the later decades are much younger than those in the 1950s and 1960s. There is no real go olympiad. Per­haps the clos­est equiv­a­lent is the an­nual (usu­al­ly) team match be­tween the two strongest na­tions, Japan and Chi­na, which ran un­til 1996. The span of years is fairly short. Fig. 3 presents me­dian age of the win­ning team over this pe­ri­od. The data are quite vari­able, usu­ally be­cause the Chi­nese team started much younger and got older and the Japan­ese team got younger. The data show no clear down­wards age trend.

  • Bozulich, R. (1992). The go play­er’s al­manac. Tokyo: The Ishi Press

…Var­i­ous fac­tors vary­ing over the decades may affect sci­en­tific pro­duc­tiv­i­ty, mask­ing any effects of ris­ing g. First, fund­ing for ba­sic re­search may vary great­ly, and par­tic­u­lar fields may fall in or out of favour. Sec­ond, fields change over time, mak­ing com­par­isons be­tween decades prob­lem­at­i­cal. It may take much longer to reach the fron­tiers of knowl­edge in later decades, for ex­am­ple. In the early stages, there may be rel­a­tively few re­searchers and differ­ent prob­lems to solve (Gupta & Karisid­dap­pa, 1996). A field’s easy prob­lems may be solved and the re­main­ing ones be in­tractable. Some fields even be­come rel­a­tively worked out, with their ma­jor prob­lems solved, and so pro­duc­tiv­ity de­clines. Hor­gan (1996) even ar­gues that sci­ence it­self soon will be worked out. The in­creas­ing cost of equip­ment has meant more team work in some fields. A par­ti­cle physics pa­per may have hun­dreds of au­thors.

  • Gup­ta, B. M., & Karisid­dap­pa, C. R. (1996). “Au­thor pro­duc­tiv­ity pat­terns in the­o­ret­i­cal pop­u­la­tion ge­net­ics (1900-1980)”. Sci­en­to­met­rics, 36, 19±41
  • Hor­gan, J. (1996). The end of sci­ence. Read­ing, MA: Ad­dis­on-Wes­ley.

…Stephan and Levin (1992) ar­gue that the sci­en­tific ca­pac­ity of the United States has de­clined over the last few decades, partly be­cause the sci­en­tific com­mu­nity is ag­ing and be­cause they say that the av­er­age qual­ity of new sci­en­tists is de­clin­ing. Sci­ence has be­come a less at­trac­tive ca­reer. The United States pro­duces about a third of the world’s sci­ence but ev­i­dence sug­gests that in­tel­lec­tual tal­ent has been shift­ing to more at­trac­tive fields. For ex­am­ple, Bowen and Schus­ter (1986) say that, be­tween 1945 and 1969, 1.2 times as many Phi Beta Kappa (an elite stu­dent so­ci­ety) mem­bers chose ca­reers in busi­ness, law and med­i­cine as in acad­eme. But in the 1970s, five times as many did. US sci­ence grad­u­ate stu­dents now often are for­eign­ers as lo­cals shift to bet­ter paid fields (North, 1995). In Aus­tralia, the en­trance exam mark cut­offs to en­ter uni­ver­sity sci­ence courses have steadily dropped over the years as fewer stu­dents ap­ply, while top marks are needed for courses in fi­nance, law and med­i­cine.

I ex­am­ined some In­sti­tute of Sci­en­tific In­for­ma­tion (ISI) data from 1955 to 1997, from the ISI’s Sci­ence Ci­ta­tion In­dex Guide in 1997, which in­cludes lists of source pub­li­ca­tions. Fig. 4 presents num­bers of ar­ti­cles pub­lished in each year and num­ber of unique source au­thors. The lat­ter cat­e­gory nat­u­rally would not in­clude all sci­en­tists, as many PhD grad­u­ates never pub­lish an ar­ti­cle (Cole & Phe­lan, 1999). Data on au­thor num­bers from 1966 to 1979 could not be ob­tained, de­spite re­peated re­quests to ISI. Al­so, ISI’s pub­lished fig­ure for au­thors in 1965, nearly dou­ble that of 1964, may be a mis­print. Fig. 4 shows a huge rise in num­ber of ar­ti­cles pub­lished. So, by this mea­sure sci­en­tific pro­duc­tiv­ity has risen great­ly. How­ev­er, the num­ber of unique au­thors has also risen, while the ac­tual pro­duc­tiv­ity per unique au­thor has de­clined slight­ly, from 0.967 in 1955 to 0.771 in 1997. This may have many caus­es, such as the trend to mul­ti­-au­thor pa­pers, ris­ing cost of equip­ment, shorter ca­reer spans, and so on. The data sug­gest that sci­en­tific pro­duc­tiv­ity has risen. In­deed, in many fields of sci­ence and in math­e­mat­ics, the an­nual num­ber of ar­ti­cles pub­lished is dou­bling every 10-15 years (Od­lyzko, 1995). The num­bers in Fig. 4 even may un­der­es­ti­mate the growth in pro­duc­tiv­i­ty. Com­pe­ti­tion for pub­li­ca­tion space often is se­vere. Many jour­nals have high re­jec­tion rates, tak­ing only the best of those sub­mit­ted. The num­ber of ar­ti­cles never pub­lished may have risen great­ly, too.

  • Odlyzko, A. M. (1995). Tragic loss or good rid­dance? The im­pend­ing demise of tra­di­tional schol­arly jour­nals. In­ter­na­tional Jour­nal of Hu­man-Com­puter Stud­ies, 42, 71±122

“Phil­an­thropy’s suc­cess sto­ries”, co-founder Holden Karnof­sky:

“One ex­cep­tion is the Case­book for The Foun­da­tion: A Great Amer­i­can Se­cret, which lists and dis­cusses”100 of the high­est-achiev­ing foun­da­tion ini­tia­tives" since 1900…I thor­oughly ex­am­ined this vol­ume, and col­lected some ba­sic notes into a spread­sheet…The most im­pres­sive cases (in my view) are mostly the ear­lier ones. Though the Case­book fo­cuses on more re­cent phil­an­thropy (78 of its 100 cases are post-1950), 9 of the 14 cases I found most im­pres­sive are pre-1950 (and a 10th is from 1952).

A pos­si­ble ex­pla­na­tion is that the space of do­ing good has be­come more crowded over time. For ex­am­ple, note that

  • To­tal U.S. gov­ern­ment health spend­ing was 0.26% of GDP in 1902 and 0.92% of GDP in 1950; by con­trast, in 2009, it was 7.06% of GDP (these fig­ures are in the spread­sheet linked above), and even most de­vel­op­ing coun­tries spend 2%+ of GDP on in this area (source). In 1927, the Com­mon­wealth Fund pi­loted a rural hos­pi­tal pro­gram; there aren’t a lot of “phil­an­thropic op­por­tu­ni­ties” that look like that to­day.
  • To­tal U.S. gov­ern­ment ed­u­ca­tion spend­ing was 1.07% of GDP in 1902 and 3.28% of GDP in 1950; by con­trast, in 2009, it was 6.16% of GDP (these fig­ures are in the spread­sheet linked above), and even most de­vel­op­ing coun­tries spend 3%+ of GDP on in this area (source). In 1902, the Rock­e­feller Foun­da­tion funded ad­vo­cacy for pro­vid­ing pub­lic schools in the U.S. South; there aren’t a lot of “phil­an­thropic op­por­tu­ni­ties” that look like that to­day.
  • More con­text: The De­part­ment of Ed­u­ca­tion was cre­ated in 1979, the Na­tional Sci­ence Foun­da­tion was cre­ated in 1950, and the Na­tional In­sti­tutes of Health be­gan in 1930 (but have grown sig­nifi­cantly since; in fact one of the “suc­cess sto­ries” in the Case­book dis­cusses the growth of the NIH bud­get from $2.4 mil­lion in 1945 to $5.5 bil­lion in 1985)."

http­s://www.g­w­­doc­s/al­ger­non/2012-wood­ley.pdf :

In­no­va­tion rates were ob­tained from Hueb­ner (2005a), who de­fines this vari­able in terms of the num­ber of im­por­tant sci­en­tific and tech­no­log­i­cal de­vel­op­ments per year di­vided by the world pop­u­la­tion. This met­ric there­fore cap­tures the in­no­v­a­tive ca­pac­ity of pop­u­la­tions on a yearly ba­sis. In de­vel­op­ing his in­no­va­tion rate mea­sures Hueb­ner ob­tained a list of 7198 im­por­tant events in the his­tory of sci­ence and tech­nol­ogy com­piled by Bunch and Helle­mans (2004), which spans from 1455 to 2004. By curve-fit­ting these data to a Gauss­ian dis­tri­b­u­tion, Hueb­ner at­tempts to pre­dict fu­ture in­no­va­tion rates out to the 22nd cen­tu­ry. Hueb­n­er’s his­tor­i­cal and fu­ture world pop­u­la­tion es­ti­mates were de­rived from the U.S. Cen­sus Bu­reau (2004a, 2004b). The es­ti­mates were avail­able on a decadal ba­sis and were ob­tained from Hueb­n­er’s Fig. 1 (p. 982).

Mur­ray’s in­dex [Hu­man ac­com­plish­ment: The pur­suit of ex­cel­lence in the arts and sci­ences, 800 BC to 1950] is com­puted on the ba­sis of the weighted per­cent­age of sources (i.e. mul­ti­ple lists of key events in the his­tory of sci­ence and tech­nol­o­gy), which in­clude a par­tic­u­lar key event. Al­though Mur­ray’s data are not as ex­ten­sive in time as are Hueb­n­er’s, it is ap­par­ent that rate of ac­com­plish­ment in­creases com­men­su­rately with Hueb­n­er’s in­dex in the pe­riod from 1455 to the mid­dle of the 19th cen­tu­ry, and then de­clines to­wards the end of that cen­tury and into the 20th. Mur­ray’s in­dex was found to cor­re­late highly with Hueb­n­er’s (r=.865, p < .01, N = 50 decades). In an ear­lier un­pub­lished study, Gary (1993) com­puted in­no­va­tion rates us­ing Asi­mov’s (1994) Chronol­ogy of Sci­ence and Dis­cov­ery. He found the same shaped curve as that de­scribed by both Hueb­ner and Mur­ray, with an in­no­va­tion peak oc­cur­ring at the end of the 19th cen­tu­ry. Hueb­n­er’s in­dex cor­re­lates strongly with Gary’s (r=.853, Pb .01, N =21 time points). It should be noted that the ob­ser­va­tion of peak in­no­va­tion at the end of the 19th cen­tury dates back to the work of Sorokin (1942), thus it is con­cluded that Hueb­n­er’s in­dex ex­hibits high con­ver­gent va­lid­i­ty.

  • Gary, B. L. (1993). “A new timescale for plac­ing hu­man events, de­riva­tion of per capita rate of in­no­va­tion, and a spec­u­la­tion on the tim­ing of the demise of hu­man­ity”. Un­pub­lished Man­u­script
  • Sorokin, P. A. (1942). The cri­sis of our age: The so­cial and cul­tural out­look. Boston: E. P. Dut­ton

To con­trol for this Hueb­n­er’s crit­ics sug­gest re-es­ti­mat­ing in­no­va­tion rates us­ing just the in­no­va­tion-gen­er­at­ing coun­tries. This analy­sis was con­ducted us­ing raw decadal in­no­va­tion data from Bunch and Helle­mans (2004), along with data on Eu­ro­pean pop­u­la­tion growth from 1455 to 1995 (from McEvedy & Jones [1978] and the US Cen­sus Bu­reau) com­bined with data on US pop­u­la­tion growth from 1795 to 1995 (from var­i­ous sta­tis­ti­cal ab­stracts of the United States avail­able from the US Cen­sus Bu­reau). The re­sul­tant in­no­va­tion rates were found to cor­re­late at r = .927 (P b .01, N = 55 decades) with Hueb­n­er’s orig­i­nal es­ti­mates, which in­di­cates that the in­no­va­tion rate data are in­sen­si­tive to de­ci­sion rules con­cern­ing which set of pop­u­la­tion es­ti­mates are used. Where choice of pop­u­la­tion mat­ters is in ex­trap­o­lat­ing fu­ture de­clines in in­no­va­tion rate.

  • McEvedy & Jones [1978] ???

Whilst a geno­typic IQ de­cline of be­tween 1 and 2 points a gen­er­a­tion does not seem large, it is im­por­tant to stress the im­pact that such a change can have on the fre­quen­cies of those with the high­est lev­els of IQ. A 105-109 point de­cline in the West­ern geno­typic IQ mean would have de­creased the pro­por­tion of the pop­u­la­tion with the sort of IQ needed for sig­nifi­cant in­no­va­tion (i.e. ≥ 135) by ~55-75% per­cent. The world­wide in­crease in the rate of in­no­va­tion from 1455 to 1873 fol­lowed by a sharp de­cline is con­sis­tent not only with con­tin­ued dys­ge­n­e­sis in the West since the lat­ter half of the 19th cen­tu­ry, but also with the ex­is­tence of a “eu­genic phase” in the pop­u­la­tion cy­cle (Weiss, 2008). Dur­ing this phase geno­typic in­tel­li­gence was ris­ing and in­no­va­tors were be­com­ing more com­mon on a per capita ba­sis, con­gru­ent with pos­i­tive di­rec­tional se­lec­tion for ‘bour­geois’ traits.

It must be noted that to­tal num­bers of in­no­va­tions are not as strongly re­lated to geno­typic IQ as are in­no­va­tion rates (r=.512, pb.01, N=55). To­tal num­bers of in­no­va­tions (which based on Bunch and Helle­mans [2004] ap­pear to have peaked in the 1960’s) re­late more strongly to the size of the most in­no­v­a­tive pop­u­la­tions. This re­la­tion­ship sug­gests that big­ger pop­u­la­tions con­tain more in­no­va­tors, how­ever dys­ge­n­e­sis is es­sen­tially ‘di­lut­ing’ the im­pact of in­no­va­tors, such that per capita in­no­v­a­tive ca­pac­ity de­clines with the pas­sage of time. This process should be ap­par­ent in the ways in which sci­ence is or­ga­nized in the mod­ern world. For ex­am­ple, if rel­a­tive to the pop­u­la­tion as a whole high in­tel­li­gence in­di­vid­u­als are be­com­ing scarcer, es­tab­lished sci­en­tists might have to re­sort to re­cruit­ing in­di­vid­u­als of more mediocre abil­i­ty. This might ex­plain the ten­dency for con­tem­po­rary sci­en­tists, more so than sci­en­tists of ear­lier gen­er­a­tions, to se­lect for con­sci­en­tious and so­cia­ble work­ers as high con­sci­en­tious­ness does not re­quire high IQ (Charl­ton, 2008). Con­sis­tent with Charl­ton’s (2008) ar­gu­ment, it has been found that whilst the size of sci­en­tific teams has been in­creas­ing, the rel­a­tive im­pact of in­di­vid­ual sci­en­tists has been de­creas­ing (Jones, 2009; Wuchty, Jones, & Uzzi, 2007).

  • Charl­ton, B. G. (2008). Why are mod­ern sci­en­tists so dull? How sci­ence se­lects for per­se­ver­ance and so­cia­bil­ity at the ex­pense of in­tel­li­gence and cre­ativ­i­ty. Med­ical Hy­pothe­ses, 72, 237-243
  • Jones, B. (2009). The bur­den of knowl­edge and the “death of the re­nais­sance man”: Is in­no­va­tion get­ting hard­er? The Re­view of Eco­nomic Stud­ies, 76, 283-317
  • Wuchty, S., Jones, B. F., & Uzzi, B. (2007). The in­creas­ing dom­i­nance of teams in the pro­duc­tion of knowl­edge. Sci­ence, 316, 1036-1039

An­other haz­ard is that in the ab­sence of a “crit­i­cal mass” of suffi­ciently in­tel­li­gent in­di­vid­u­als en­gen­der­ing an ap­pro­pri­ate level of sci­en­tific rig­or, “junk sci­ence” has the po­ten­tial to pro­lif­er­ate to an ex­tent never be­fore seen in free na­tions (cf. Cof­nas, 2012).

This trend may also cou­ple with the an­ti-Flynn effect, which has been ob­served in a num­ber of West­ern na­tions over the last cou­ple of decades, and is char­ac­ter­ized by sig­nifi­cant losses in phe­no­typic IQ (Fly­nn, 2009b; Shayer & Gins­burg, 2009; Sun­det, Bar­laug, & Tor­jussen, 2004; Teas­dale & Owen, 2008).

  • Fly­nn, J. R. (2009b). Re­quiem for nu­tri­tion as the cause of IQ gains: Raven’s gains in Britain 1938-2008. Eco­nom­ics and Hu­man Bi­ol­o­gy, 7, 18-27
  • Shay­er, M. D., & Gins­burg, D. (2009). Thirty years on - A large an­ti-Flynn effect? (I­I): 13- and 14-year olds. Pi­aget­ian tests of for­mal op­er­a­tions norms 1976-2006/7. British Jour­nal of Ed­u­ca­tional Psy­chol­o­gy, 79, 409-418
  • Sun­det, J. M., Bar­laug, D. G., & Tor­jussen, T. M. (2004). The end of the Flynn effect? A study of sec­u­lar trends in mean in­tel­li­gence test scores of Nor­we­gian con­scripts dur­ing half a cen­tu­ry. In­tel­li­gence, 32, 349-362
  • Teas­dale, T. W., & Owen, D. R. (2008). Sec­u­lar de­clines in cog­ni­tive test scores: A re­ver­sal of the Flynn effect. In­tel­li­gence, 36, 121-126

Tech­nolo­gies like ga­mete cloning, when ma­ture enough, may per­mit in­di­vid­u­als to se­lect for IQ en­hanc­ing al­le­les, but would only re­al­is­ti­cally be able to raise the IQ of off­spring by a point or two at best (Lee, 2010).

  • Lee, J. J. (2010). Re­view of in­tel­li­gence and how to get it: Why schools and cul­tures count, R.E. Nis­bett, Nor­ton, New York, NY (2009). ISBN: 9780393065053. Per­son­al­ity and In­di­vid­ual Differ­ences, 48, 247-255 [count me skep­ti­cal, given the ex­ist­ing ge­netic vari­a­tion… we should be able to get more than a point or two]

…it is in­di­cated that the Flynn effect has­n’t re­ally started to take off in these na­tions, but that it has the po­ten­tial to do so (Wicherts, Dolan, Carl­son, & van der Maas, 2010). This is ev­i­denced by ob­ser­va­tions of a nascent Flynn effect in South Africa (te Ni­jen­huis, Mur­phy, & van Ee­den, 2011), Kenya (Da­ley, Wha­ley, Sig­man, Es­pinosa, & Neu­mann, 2003), Do­minica (Meisen­berg et al., 2005), Saudi Ara­bia (Bat­ter­jee, 2011) and else­where. It is en­tirely pos­si­ble there­fore that many of the less de­vel­oped na­tions are en­ter­ing into the early stages of an “en­hanced growth” phase in the Flynn effect, a con­se­quence of which might be sig­nifi­cant de­creases in pover­ty, such as is cur­rently oc­cur­ring in Africa (Sala-i-Martin & Pinkovskiy, 2010).

  • Wicherts, J. M., Dolan, C. V., Carl­son, J. S., & van der Maas, H. L. J. (2010). “Raven’s test per­for­mance of sub­-Sa­ha­ran Africans: Mean lev­el, psy­cho­me­t­ric prop­er­ties, and the Flynn effect”. Learn­ing and In­di­vid­ual Differ­ences, 20, 135-151
  • te Ni­jen­huis, J., Mur­phy, R., & van Ee­den, R. (2011). “The Flynn effect in South Africa”. In­tel­li­gence, 36, 456-467
  • Da­ley, T. C., Wha­ley, S. E., Sig­man, M. D., Es­pinosa, M. P., & Neu­mann, C. (2003). “IQ on the rise: the Flynn effect in rural Kenyan chil­dren”. Psy­cho­log­i­cal Sci­ence, 14, 215-219
  • Meisen­berg, G., Law­less, E., Lam­bert, E., & New­ton, A. (2005). “The Flynn effect in the Caribbean: Gen­er­a­tional change of cog­ni­tive test per­for­mance in Do­minica”. Mankind Quar­ter­ly, 46, 29-69.
  • Bat­ter­jee, A. (2011). “In­tel­li­gence and ed­u­ca­tion: the Saudi case”. Mankind Quar­ter­ly, 52, 133-190
  • Sala-i-Mart­in, X., & Pinkovskiy, M. (2010). “African poverty is falling… much faster than you think!” Na­tional Bu­reau of Eco­nomic Re­search Work­ing Pa­per No. 15775

http://dl.­drop­box.­com/u/85192141/2012-meisen­berg.pdf :

The re­la­tion­ship be­tween in­tel­li­gence and fer­til­ity has been in­ves­ti­gated since the early years of the 20th cen­tu­ry. Dur­ing the first half of the cen­tu­ry, stud­ies in Britain and the United States usu­ally found a neg­a­tive re­la­tion­ship be­tween IQ and com­pleted fam­ily size (Anas­tasi, 1956; Cat­tell, 1936, 1937; Daw­son, 1932/33), al­though atyp­i­cal re­sults were ob­tained oc­ca­sion­ally (Willoughby & Coogan, 1940). These early re­sults were chal­lenged by a se­ries of stud­ies with mainly White mid­dle-class groups in the United States at the time of the baby boom, which re­ported a neg­li­gi­ble or slightly pos­i­tive re­la­tion­ship be­tween IQ and num­ber of chil­dren (Ba­je­ma, 1963, 1968; Falek, 1971; Hig­gins et al., 1962; Waller, 1971). These re­sults were com­ple­mented by the ob­ser­va­tion that sub­fer­til­ity of men in Who’s Who in Amer­ica dis­ap­peared for co­horts born after 1910 (Kirk, 1957). The con­clu­sion at that time was that dys­genic fer­til­ity for in­tel­li­gence was a tem­po­rary phe­nom­e­non dur­ing the de­mo­graphic tran­si­tion when the more in­tel­li­gent pi­o­neered the use of con­tra­cep­tion, but dis­ap­peared at a later stage when con­tra­cep­tive habits had diffused through the en­tire pop­u­la­tion (Os­born & Ba­je­ma, 1972).

Al­though these re­sults seemed to make the re­la­tion­ship be­tween in­tel­li­gence and re­pro­duc­tion a dead is­sue, stud­ies of co­horts who re­pro­duced after the 1960s again showed the fa­mil­iar neg­a­tive re­la­tion­ship. As early as 1978, a neg­a­tive re­la­tion­ship was ob­served in mar­ried White Amer­i­can women. Re­mark­ably, this neg­a­tive re­la­tion­ship re­mained sig­nifi­cant even with ed­u­ca­tion and so­cioe­co­nomic back­ground con­trolled (Udry, 1978). How­ev­er, the value of this re­sult is am­bigu­ous be­cause most of the women in this sam­ple (aged 15-44) had not yet com­pleted their child­bear­ing. More sub­stan­tial were the find­ings of Vin­ing (1982, 1986, 1995), who pro­vided ev­i­dence for the reemer­gence of a dys­genic trend among those born after 1935. Vin­ing’s con­clu­sions were fur­ther sup­ported by Rether­ford and Sewell (1988, 1989), who found a neg­a­tive re­la­tion­ship be­tween in­tel­li­gence at age 17 and num­ber of chil­dren at age 35 for a pre­dom­i­nantly White sam­ple of high school se­niors in Wis­con­sin. Ad­di­tional ev­i­dence was found in the Gen­eral So­cial Sur­vey (van Court & Bean, 1985; Lynn & van Court, 2004), which showed a neg­a­tive re­la­tion­ship be­tween word knowl­edge and the num­ber of chil­dren. Lynn and van Court (2004) con­cluded that the re­la­tion­ship had been neg­a­tive for all co­horts born after 1900, al­though it was weaker for those born 1920-1929. In these stud­ies, the dys­genic fer­til­ity was far stronger in fe­males than males.

…Ta­bles 3 and 4 also show that in these mod­els the like­li­hood of be­ing mar­ried is in­creased by IQ and g, both di­rectly and, to a lesser ex­tent, in­di­rectly through re­li­gious at­ten­dance. How­ev­er, in the White groups this effect is op­posed by the an­ti-mar­riage effect of ed­u­ca­tion. These re­sults con­trast with a British study, which found that nev­er-mar­ried women had higher child­hood IQs than mar­ried women al­though mar­ried men tended to have higher IQs than nev­er-mar­ried men (Tay­lor et al., 2005). In an Afro-Caribbean pop­u­la­tion, how­ev­er, high IQ raised the mar­riage rate for both males and fe­males (Meisen­berg, Law­less, Lam­bert, & New­ton, 2006).

…S­e­lec­tion against high in­tel­li­gence has been ob­served through­out most of the 20th cen­tury in Eu­rope and the United States (Cat­tell, 1936, 1937; Lynn & van Court 2004; Rether­ford & Sewell, 1988; van Court & Bean, 1985), where it was prob­a­bly present since the be­gin­ning of the fer­til­ity tran­si­tion in the 19th cen­tury (Notestein, 1936; Steven­son, 1920). We can es­ti­mate that with­out this se­lec­tion effect the av­er­age in­tel­li­gence in these coun­tries to­day would be up to 5 points higher than it is-about as high as the av­er­age IQ in China to­day (Lynn & Van­hanen, 2006), where re­pro­duc­tive differ­en­tials still fa­vored wealth, lit­er­acy and pre­sum­ably in­tel­li­gence in the early part of the 20th cen­tury (Lam­son, 1935; Notestein, 1938). In pre-in­dus­trial so­ci­eties, fer­til­ity usu­ally was high­est among the wealthy classes (Clark & Hamil­ton, 2006; Hadeishi, 2003; Har­rell, 1985; Lam­son, 1935) and also among the ed­u­cat­ed, at least in the few stud­ies that in­cluded a mea­sure of ed­u­ca­tion (Clark & Hamil­ton, 2006; Hadeishi, 2003; Lam­son, 1935). Al­though se­lec­tion against high ed­u­ca­tional at­tain­ment, and pre­sum­ably high in­tel­li­gence, is found world­wide to­day (Meisen­berg, 2008; Wein­berg­er, 1987), his­tor­i­cally it presents a novel phe­nom­e­non.

Dur­ing the 20th cen­tu­ry, the small ge­netic de­cline was masked by mas­sive en­vi­ron­men­tal im­prove­ments, es­pe­cially in the ed­u­ca­tional sys­tem, which caused IQ gains on the or­der of 10 points per gen­er­a­tion (Fly­nn, 1987). This en­vi­ron­men­tal effect was at least ten times greater than the de­cline pre­dicted from ge­netic se­lec­tion, and thus made ge­netic se­lec­tion seem ir­rel­e­vant. How­ev­er, re­cent re­sults show that this ris­ing trend, known as the Flynn effect, is ei­ther di­min­ish­ing or re­vers­ing in the most ad­vanced so­ci­eties. A mar­ginal Flynn effect was still ob­served among chil­dren born be­tween 1973 and 1995 in the United States (Rodgers & Wän­ström, 2007), most re­cent trends in Britain are am­bigu­ous (Fly­nn, 2009; Shay­er, Gins­burg & Coe, 2007), and mil­i­tary con­scripts born after about 1980 in Den­mark (Teas­dale & Owen, 2008) and Nor­way (Sun­det, Bar­laug & Tor­jussen, 2004) show stag­nat­ing in­tel­li­gence or a slow de­cline.

…The im­pli­ca­tions of the present find­ings for the United States need to be stated clear­ly: As­sum­ing an in­defi­nite con­tin­u­a­tion of cur­rent fer­til­ity pat­terns, an un­chang­ing en­vi­ron­ment and a gen­er­a­tion time of 28 years, the IQ will de­cline by about 2.9 points/­cen­tury as a re­sult of ge­netic se­lec­tion. The pro­por­tion of highly gifted peo­ple with an IQ higher than 130 will de­cline by 11.5% in one gen­er­a­tion and by 37.7% in one cen­tu­ry. Since many im­por­tant out­comes, in­clud­ing eco­nomic wealth (Rin­der­mann, 2008a) and democ­racy (Rin­der­mann, 2008b), are fa­vored by high in­tel­li­gence, ad­verse long-term con­se­quences of such a trend would be ex­pected al­though short­-term con­se­quences on a time scale of less than one cen­tury are neg­li­gi­ble.

Flynn effect on hol­low IQ? Wicherts et al 2004: “Are in­tel­li­gence tests mea­sure­ment in­vari­ant over time? In­ves­ti­gat­ing the na­ture of the Flynn effect”

Mea­sure­ment in­vari­ance im­plies that gains over the years can be at­trib­uted to in­creases in the la­tent vari­ables that the tests pur­port to mea­sure. The stud­ies re­ported con­tain orig­i­nal data of Dutch Wech­sler Adult In­tel­li­gence Scale (WAIS) gains from 1967 to 1999, Dutch Differ­en­tial Ap­ti­tude Test (DAT) gains from 1984 to 1995, gains on a Dutch chil­dren in­tel­li­gence test (RAKIT) from 1982 to 1993, and re­analy­ses of re­sults from Must, Must, and Raudik [In­tel­li­gence 167 (2003) 1-11] and Teas­dale and Owen [In­tel­li­gence 28 (2000) 115-120]. The re­sults of multi­group con­fir­ma­tory fac­tor analy­ses clearly in­di­cate that mea­sure­ment in­vari­ance with re­spect to co­horts is un­ten­able. Uni­form mea­sure­ment bias is ob­served in some, but not all sub­tests. The im­pli­ca­tions of these find­ings are dis­cussed.


As you de­velop more drugs, your stan­dard for safety should go up be­cause it be­comes ever less likely that a new drug is su­pe­rior to any of the old ones but the chance it fooled your tests re­mains pretty much the same.

Imag­ine you have a lit­tle ran­dom num­ber gen­er­a­tor 1-100, and you want to max­i­mize the num­ber you draw, but you also have, say, a 10% chance of mis­read­ing the num­ber each time. Ini­tially you’d keep dis­card­ing your num­ber - pfft, a 50? pfft, a 65? I can do bet­ter than that! - but once you’ve suc­cess­fully drawn a 95, then you want to start ex­am­in­ing the num­bers care­ful­ly. ‘I just drew a 96, but the odds of get­ting a num­ber >95 is just 4%! It’s more likely that I just mis­read this 96… oh wait, it was ac­tu­ally 69. My bad.’

(I’m not sure how ac­cu­rate this lit­tle model is, but it cap­tures how I feel about it.)

In gen­er­al, I don’t buy the ar­gu­ment that the FDA is stran­gling all sorts of fan­tas­tic in­no­va­tion through its fo­cus on safe­ty. If safety is cost­ing us bil­lions and bil­lions every year in op­por­tu­nity cost for ei­ther drug ap­provals or re­search, we ought to see coun­tries with laxer reg­u­la­tions and sci­en­tific ca­pa­bil­ity - like in East Asia - start­ing up mas­sive phar­ma­ceu­ti­cal gi­ants and steam­rolling West­ern corps with the low-hang­ing fruit we have fas­tid­i­ously turned up our noses at. We don’t ob­serve this. We ob­serve oc­ca­sional in­no­va­tions and con­tri­bu­tions, and ap­par­ently they’re pretty ac­tive in stem cell re­search (which Amer­ica did re­press), but at the rates one would gen­er­ally ex­pect. The fall in re­turns is pretty huge, and if it was due solely to safe­ty, aban­don­ing safety ought to lead to pro­duc­tiv­ity gains of 2, 3, maybe 10 times the Amer­i­can sci­en­tist equiv­a­lents. I think we would have heard if that were the case.

What causes di­min­ish­ing re­turns? Dun­no. It’s a pretty com­mon phe­nom­e­non.

ge­net­ics un­der­achiev­ing: http://blog.sethrobert­­nomic­s-con­fi­den­tial-the-faux-won­der­land-of-ice­land/

Ques­tion: Dick, would you care to com­ment on the rel­a­tive effec­tive­ness be­tween giv­ing talks, writ­ing pa­pers, and writ­ing books?

Ham­ming: In the short­-haul, pa­pers are very im­por­tant if you want to stim­u­late some­one to­mor­row. If you want to get recog­ni­tion long-haul, it seems to me writ­ing books is more con­tri­bu­tion be­cause most of us need ori­en­ta­tion. In this day of prac­ti­cally in­fi­nite knowl­edge, we need ori­en­ta­tion to find our way. Let me tell you what in­fi­nite knowl­edge is. Since from the time of New­ton to now, we have come close to dou­bling knowl­edge every 17 years, more or less. And we cope with that, es­sen­tial­ly, by spe­cial­iza­tion. In the next 340 years at that rate, there will be 20 dou­blings, i.e. a mil­lion, and there will be a mil­lion fields of spe­cialty for every one field now. It is­n’t go­ing to hap­pen. The present growth of knowl­edge will choke it­self off un­til we get differ­ent tools. I be­lieve that books which try to di­gest, co­or­di­nate, get rid of the du­pli­ca­tion, get rid of the less fruit­ful meth­ods and present the un­der­ly­ing ideas clearly of what we know now, will be the things the fu­ture gen­er­a­tions will val­ue. Pub­lic talks are nec­es­sary; pri­vate talks are nec­es­sary; writ­ten pa­pers are nec­es­sary. But I am in­clined to be­lieve that, in the long-haul, books which leave out what’s not es­sen­tial are more im­por­tant than books which tell you every­thing be­cause you don’t want to know every­thing. I don’t want to know that much about pen­guins is the usual re­ply. You just want to know the essence.



One of the more con­tro­ver­sial ex­pla­na­tions for di­min­ish­ing re­turns is that the di­min­ish­ing re­flects the qual­ity of the hu­man cap­i­tal: the peak qual­ity has de­clined. On this view, im­por­tant dis­cov­er­ies and in­ven­tions are dis­pro­por­tion­ately due to the smartest sci­en­tists and in­ven­tors. As the smartest cease to com­mand a re­pro­duc­tive ad­van­tage, their ranks are in­evitably im­pov­er­ished.

This might seem to con­tra­dict the well-known and also fly counter to the many IQ-en­hanc­ing in­ter­ven­tions over the past cen­turies such as vac­ci­na­tions or , ex­cept the dys­genic hy­poth­e­sis refers to the geno­typic po­ten­tial for in­tel­li­gence and only in­di­rectly to the phe­no­type. That is, in­tel­li­gence is a joint prod­uct of genes and en­vi­ron­ment: genes set a ceil­ing but the en­vi­ron­ment de­ter­mines how much of the po­ten­tial will be re­al­ized. So if the en­vi­ron­ment im­proves, more in­di­vid­u­als will be well-nur­tured - and hit their ge­netic ceil­ings. This sort of rea­son­ing pre­dicts that we could see an in­crease in pop­u­la­tion-wide av­er­ages, per the Flynn effect, and we could also see de­creases in the tail for low in­tel­li­gence (per the pub­lic health in­ter­ven­tions, eg. no more iodine-re­lated cre­tinis­m), but as­sum­ing the en­vi­ron­ment was not so ter­ri­ble that no in­di­vid­ual hit their ceil­ings, we’d see a trun­cat­ing of the bell curve with fewer in­di­vid­u­als than one would pre­dict. If the dys­genic se­lec­tion effects con­tin­ued, one might even see re­duc­tions in the ab­solute num­bers of highly in­tel­li­gent peo­ple.

So in this nar­ra­tive, genes for in­tel­li­gence cook along through his­tory in sub­par de­fi­cient en­vi­ron­ments, ek­ing out mod­est fit­ness ad­van­tages (due to pre­sum­able costs like in­creased me­tab­o­lism) and main­tain­ing their pres­ence in the gene pool, un­til the In­dus­trial Rev­o­lu­tion hap­pens, caus­ing the in which sud­denly richer coun­tries be­gin to re­pro­duce less, ap­par­ently due to wealth, and who are the wealth­i­est in those coun­tries? The most in­tel­li­gent. So even as the In­dus­trial and Sci­en­tific Rev­o­lu­tions and eco­nomic growth (pow­ered by the in­tel­li­gent) all si­mul­ta­ne­ously im­prove the en­vi­ron­ment in a myr­iad of ways, the most in­tel­li­gent are fail­ing to re­pro­duce and the geno­typic ceil­ing be­gins falling even as the phe­no­typic av­er­age con­tin­ues ris­ing, un­til the trends in­ter­sect.

This is a com­plex nar­ra­tive. There are mul­ti­ple main points to es­tab­lish, any of which could tor­pedo the over­all the­sis:

  1. that in­tel­li­gence has a sub­stan­tial ge­netic com­po­nent

    If there is no ge­netic ba­sis, then there can be no dys­gen­ics.

  2. that the in­tel­li­gent (and highly in­tel­li­gent) have not al­ways re­pro­duced less and suffered fit­ness losses

    If we ob­served that the highly in­tel­li­gent were al­ways at fit­ness dis­ad­van­tages, this im­plies var­i­ous bizarre or fal­si­fied claims (like hu­mans start­ing eons ago with IQs of 1000s), and that our ba­sic model was com­pletely wrong. The truth would have to be some­thing more ex­otic like in­tel­li­gence is de­ter­mined by spon­ta­neous mu­ta­tions or the re­pro­duc­tive penalty is bal­anced by the in­clu­sive fit­ness of close rel­a­tives with mediocre in­tel­li­gence (per­haps some ).

  3. that the in­tel­li­gent (and highly in­tel­li­gent) now re­pro­duce less and their genes suffer a loss of fit­ness

    If in­tel­li­gence is be­ing re­pro­duc­tively se­lected for, then the pres­sures would not be dys­genic in this sense but eu­genic. (Such op­po­site pres­sures would not ex­plain any di­min­ish­ing mar­ginal re­turns and ac­tu­ally ar­gue against it.)

  4. that the highly in­tel­li­gent are not in­creas­ing in mod­ern times

    An­other ba­sic san­ity check like #2: if the highly in­tel­li­gent are in­creas­ing in pro­por­tion, this is the op­po­site of what the nar­ra­tive needs.

  5. that the ab­sence of the highly in­tel­li­gent could in fact ex­plain di­min­ish­ing re­turns

    If they turn out to be only as pro­duc­tive as less ex­treme mem­bers of the bell curve, then this dis­cus­sion could be en­tirely moot al­beit in­ter­est­ing: the loss of them would be off­set by the gain of their equally pro­duc­tive but dim­mer brethren. Dys­genic pres­sures would only mat­ter if it be­gan to di­min­ish their ranks too, but this could be some sort of sta­ble equi­lib­ri­um: the dim­mer oc­ca­sion­ally give birth to brighter off­spring, who do not re­pro­duce much and also do not pro­duce any more than their par­ents, all in ac­cor­dance with the pre­vi­ous points but with no dys­genic threats to the dim­mer ranks.

  • http://www.springer­link.­com/­con­tent/p15h2830v4115281


  • “Smart Teens Don’t Have Sex (or Kiss Much Ei­ther)”, Halpern et al 2000

  • http://­coun­ter­­du/archives/­Coun­ter­point_V21_I3_2001_Nov.pdf

  • http://www.half­sig­ma.­com/2006/07/s­marter_peo­ple_.html

  • http://www.half­sig­ma.­com/2006/07/­sex_­drive_de­cre.html

  • http://www.gnx­p.­com/blog/2007/04/in­ter­course-and-in­tel­li­gence.php (testos­terone link; see also the SMAP pa­pers)

  • http://www.­sci­encedi­rec­t.­com/­science/ar­ti­cle/pi­i/S1090513805000619


  • http­s://en.wikipedi­­i/Fer­til­i­ty_and_in­tel­li­gence

http://cite­­su.e­du/view­doc/­down­load­?­doi=­type­=pdf “Ex­plor­ing Sci­en­tists’ Work­ing Timetable: Do Sci­en­tists Often Work Over­time?” Wang et al 2012; such a grind is very Con­sci­en­tious, but is it good for gen­uine cre­ativ­i­ty? does­n’t cre­ativ­ity re­quire time off and work­ing on other things?

A novel method is pro­posed to mon­i­tor and record sci­en­tists’ work­ing timetable. We record the down­loads in­for­ma­tion of sci­en­tific pa­pers re­al-timely from Springer round the clock, and try to ex­plore sci­en­tists’ work­ing habits. As our ob­ser­va­tion demon­strates, many sci­en­tists are still en­gaged in their re­search after work­ing hours every day. Many of them work far into the night, even till next morn­ing. In ad­di­tion, re­search work also in­trudes into their week­ends. Differ­ent work­ing time pat­terns are re­vealed. In the US, overnight work is more preva­lent among sci­en­tists, while Chi­nese sci­en­tists mostly have busy week­ends with their sci­en­tific re­search.

“The Gray­ing of Acad­e­mia: Will It Re­duce Sci­en­tific Pro­duc­tiv­i­ty?”, Stroebe 2010

This change re­sulted in a drop in re­tire­ments of older aca­d­e­mics and has al­ready al­tered the age struc­ture at U.S. uni­ver­si­ties (Ashen­fel­ter & Card, 2002; Clark & Ghent, 2008). On the ba­sis of data ob­tained from 16,000 older fac­ulty mem­bers at 104 col­leges and uni­ver­si­ties across the United States, Ashen­fel­ter and Card (2002) con­cluded that after the abo­li­tion of manda­tory re­tire­ment, the per­cent­age of 70-year-old pro­fes­sors con­tin­u­ing to work in­creased from 10% to 40%. In an analy­sis of data from the North Car­olina uni­ver­sity sys­tem, Clark and Ghent (2008) drew a sim­i­lar con­clu­sion: > Prior to 1994, the re­tire­ment rate was 59 per­cent for fac­ulty age 70, 67 per­cent for fac­ulty age 71 and 100 per­cent for fac­ulty age 72. After the pol­icy of manda­tory re­tire­ment was re­moved, 24 per­cent of fac­ulty age 70, 19 per­cent of fac­ulty age 71, and 17 per­cent of fac­ulty age 72 re­tired. (pp. 156 -157) As a re­sult of such changes, the per­cent­age of full­time fac­ulty mem­bers age 70 or older went up three­fold (to 2.1%) be­tween the years 1995 and 2006 (Bom­bardieri, 2006). How­ev­er, at some uni­ver­si­ties the sit­u­a­tion is more ex­treme. For ex­am­ple, in the Har­vard Uni­ver­sity Fac­ulty of Arts and Sci­ences, the per­cent­age of tenured pro­fes­sors age 70 years and older has in­creased from 0% in 1992 to 9.1% in 2006 (Bom­bardieri, 2006). The im­pact of the chang­ing age struc­ture has also been felt at the Na­tional In­sti­tutes of Health (NIH), where the av­er­age age of prin­ci­pal in­ves­ti­ga­tors for NIH grants has in­creased from 30 - 40 years in 1980 to 48 years in 2007.

al­though cre­ativ­ity is mod­er­ately pos­i­tively cor­re­lated with IQ up to in­tel­li­gence lev­els that are ap­prox­i­mately one stan­dard de­vi­a­tion above the mean, the re­la­tion­ship be­comes es­sen­tially zero for more in­tel­li­gent in­di­vid­u­als (Bar­ron & Har­ring­ton, 1981; Feist & Bar­ron, 2003). Thus, when IQ scores are cor­re­lated with some valid cri­te­rion of sci­en­tific dis­tinc­tion (e.g., num­ber of ci­ta­tion­s), the cor­re­la­tions ap­proach zero (e.g., Bayer & Fol­ger, 1966; Cole & Cole, 1973). This makes it highly un­likely that a mod­est agere­lated de­crease in in­tel­li­gence should im­pair a sci­en­tist’s abil­ity to pro­duce high­-qual­ity re­search. Sim­i­lar reser­va­tions ap­ply to mea­sures of di­ver­gent think­ing, which are con­sid­ered more closely re­lated to cre­ativ­ity than are tra­di­tional in­tel­li­gence tests (e.g., Hen­nessey & Am­a­bile, 2010). Al­though there is some ev­i­dence that age decre­ments in di­ver­gent think­ing ap­pear as early as in the 40s (e.g., Mc­Crae, Aren­berg, & Costa, 1987), age ac­counts for very lit­tle vari­ance.

The most in­flu­en­tial the­ory of the as­so­ci­a­tion of age, cog­ni­tive abil­i­ty, and sci­en­tific achieve­ment has been sug­gested by Si­mon­ton (e.g., 1985, 1988, 1997, 2002), un­doubt­edly the most im­por­tant and pro­lific re­searcher in the area of the psy­chol­ogy of sci­ence. He de­vel­oped an el­e­gant quan­ti­ta­tive model of the de­cline in cre­ative po­ten­tial, which pre­dicts that the as­so­ci­a­tion be­tween age and pro­duc­tiv­ity is curvi­lin­ear and de­clines with ca­reer age rather than chrono­log­i­cal age. The ba­sic as­sump­tion of Si­mon­ton’s the­ory is that each cre­ator starts off with a fixed amount of ini­tial cre­ative po­ten­tial. This cre­ative po­ten­tial con­sists of “con­cepts, ideas, im­ages, tech­niques, or other cog­ni­tions that can be sub­jected to free vari­a­tion” (Si­mon­ton, 1997, pp. 67- 68). Of the pos­si­ble com­bi­na­tions of the­se, only a sub­set are suffi­ciently promis­ing to jus­tify fur­ther elab­o­ra­tion. Some of them may fail the em­pir­i­cal test, but some may fi­nally be worked out into fin­ished prod­ucts that might be pub­lished. Each time in­di­vid­u­als pro­duce new re­search, they use up part of their cre­ative po­ten­tial and re­duce the ideational com­bi­na­tions that are avail­able to them. Ac­cord­ing to Si­mon­ton (1997), pro­duc­tiv­ity in­creases dur­ing the first 20 years of an in­di­vid­u­al’s ca­reer, when the in­di­vid­ual still has a rich fund of cre­ative po­ten­tial and is get­ting bet­ter and bet­ter at turn­ing these ideas into pub­lish­able out­put. How­ev­er, ap­prox­i­mately 20 years into an in­di­vid­u­al’s ca­reer, a peak is typ­i­cally reached. After that, pro­duc­tiv­ity be­gins to de­cline, be­cause the in­di­vid­ual has used up a sub­stan­tial pro­por­tion of his or her ini­tial cre­ative po­ten­tial.

It has been ar­gued that the differ­ences be­tween sci­en­tists in re­search pro­duc­tiv­ity are too ex­treme to be ex­plained merely by differ­ences in abil­ity or mo­ti­va­tion (Cole & Cole, 1973). For ex­am­ple, in a study of the sci­en­tific out­put of more than 1,000 Amer­i­can aca­d­e­mic psy­chol­o­gists, Den­nis (1954) found that the most pro­duc­tive 10% au­thored 41% of all pub­li­ca­tions, whereas the bot­tom 10% pro­duced less than 1%. In fact, the top half were re­spon­si­ble for 90% of to­tal out­put, and the bot­tom half, for only the re­main­ing 10%. Sim­i­larly bi­ased dis­tri­b­u­tions have been shown for other sci­ences as well as for the arts and hu­man­i­ties (Si­mon­ton, 2002). Find­ings such as these led Price (1963), a his­to­rian of sci­ence, to pro­pose Price’s law. Ac­cord­ing to this law, if k is the num­ber of re­searchers who have made at least one con­tri­bu­tion to a given field, the square root of k will be re­spon­si­ble for half of all con­tri­bu­tions in this field. Thus, if there are 100 con­trib­u­tors in a field, the top 10% will be re­spon­si­ble for half of the con­tri­bu­tions to this area.2

For ex­am­ple, in a study of pub­li­ca­tions by the 60 mem­bers of the ed­i­to­r­ial board of the Jour­nal of Coun­sel­ing Psy­chol­ogy in 2007, Duffy, Mar­t­in, Bryan, and Raque­-Bog­dan (2008) found num­ber of pub­li­ca­tions and num­ber of ci­ta­tions to cor­re­late .80. This cor­re­la­tion is some­what higher than the cor­re­la­tions typ­i­cally found for psy­chol­o­gy, which vary be­tween .50 and .70 (Si­mon­ton, 2002). Si­mon­ton (2002) there­fore con­cluded “that the qual­ity of out­put is a pos­i­tive func­tion of quan­tity of out­put: the more pub­li­ca­tions one pro­duces, the higher the odds that one will get cited” (p. 45). It is in­ter­est­ing to note that the same re­la­tion­ship has been ob­served in brain­storm­ing re­search, where the num­ber of ideas that are pro­duced by an in­di­vid­ual or a group is highly cor­re­lated with the num­ber of good ideas (e.g., Diehl & Stroe­be, 1987; Stroe­be, Ni­js­tad, & Ri­et­zschel, 2010).

Be­cause of the ex­po­nen­tial growth of the sci­en­tific com­mu­nity dur­ing the last few cen­turies, there has al­ways been an over­rep­re­sen­ta­tion of younger sci­en­tists (Price, 1963). Thus, even if sci­en­tific achieve­ment were un­re­lated to age, one would ex­pect more em­i­nent con­tri­bu­tions from young rather than old sci­en­tists. The same bias arises with stud­ies that use num­ber of pub­li­ca­tions in top jour­nals as their in­dex of sci­en­tific achieve­ment. For ex­am­ple, if one took the pub­li­ca­tions of 10 ma­jor sci­en­tific jour­nals as one’s sam­ple and then plot­ted the age dis­tri­b­u­tion of the au­thors of these pub­li­ca­tions, the re­sults would again be dis­torted by the fact that there are likely to have been more younger than older sci­en­tists in the pop­u­la­tion of sci­en­tists from which the suc­cess­ful pub­lish­ers were drawn.

The clas­sic study of No­bel lau­re­ates was pub­lished by Zuck­er­man (1977). It was based on 92 No­bel Prize win­ners who worked in the United States and won the No­bel Prize be­tween 1901 and 1972. She found that the av­er­age age at which these in­di­vid­u­als did their prize-win­ning re­search was 39 years, with win­ners of the prize in physics do­ing their re­search at 38.6 years and win­ners of the prize in med­i­cine and phys­i­ol­ogy do­ing it at 41.1 years. Sim­i­lar re­sults were re­ported by Stephan and Levin (1993), who in an up­date and ex­ten­sion of Zuck­er­man’s (1977) study an­a­lyzed the 414 win­ners of the No­bel Prize in the nat­ural sci­ences in the years 1901-1992. The av­er­age age for con­duct­ing the prize-win­ning re­search for all dis­ci­plines was 37.6 years, with physi­cists do­ing their re­search the ear­li­est, at 34.5 years, and med­ical re­search be­ing con­ducted by some­what older re­searchers, at 38.0 years. Al­though this is not old, it is also not pre­co­ciously young. How­ev­er, be­fore one draws any con­clu­sions, one must re­mem­ber that these find­ings in­form us only of the pro­por­tion of No­bel Prizes won by sci­en­tists of differ­ent ages. They do not tell us at which age sci­en­tists are most likely to win that prize. For this, we need to know the age dis­tri­b­u­tion of the pop­u­la­tion of sci­en­tists from which the No­bel Prize win­ners were se­lect­ed. Al­though Stephan and Levin (1993) failed to make such a cor­rec­tion, Zuck­er­man (1977) did, and she com­pared the age dis­tri­b­u­tion of her lau­re­ates to that of the gen­eral pop­u­la­tion of Amer­i­can sci­en­tists (see Fig­ure 1). This com­par­i­son shows that the only sub­stan­tial de­vi­a­tions from the gen­eral pop­u­la­tion oc­cur for the age group of 35 to 44 years, which is clearly over­rep­re­sented among the No­bel lau­re­ates, and the age group of 55 years and old­er, which is un­der­rep­re­sent­ed. Be­fore one con­cludes from this ev­i­dence that great sci­ence is re­ally the do­main of the mid­dle-aged, one should re­mem­ber that dur­ing the pe­riod con­sid­ered in these stud­ies, even Amer­i­can sci­en­tists were sub­ject to com­pul­sory re­tire­ment. Most re­search in the nat­ural sci­ences re­quires mon­e­tary re­sources, per­son­nel, and lab­o­ra­tory fa­cil­i­ties, which may have be­come un­avail­able to older sci­en­tists after their re­tire­ment. In an­tic­i­pa­tion of this fact, many sci­en­tists in their mid-50s may have al­ready stopped ini­ti­at­ing projects that they ex­pected to be un­able to fin­ish be­fore re­tire­ment.

For ex­am­ple, when Har­vey Lehman, one of the most pro­lific re­searchers on age and sci­en­tific achieve­ment, tab­u­lated the ages at which a sam­ple of 52 de­ceased philoso­phers had writ­ten their most sig­nifi­cant work, a sin­gle-peaked func­tion emerged: The mean age for pro­duc­ing a philo­soph­i­cal mas­ter­work was 41.5 years. Prac­ti­cally the same age curve also de­scribes the age at which sig­nifi­cant works were pro­duced in psy­chol­ogy (Lehman, 1966). Lehman’s (1953, 1966) re­search can be crit­i­cized for his fail­ure to take ac­count of the age dis­tri­b­u­tion of the pop­u­la­tion of philoso­phers and sci­en­tists from which he drew the sam­ple of ex­cel­lent con­tri­bu­tions. The data were not cor­rected for the fact that there were likely to be many more younger than older in­di­vid­u­als in the pop­u­la­tion of which the em­i­nent in­di­vid­u­als were a sub­sam­ple. How­ev­er, Wray (2004), who stud­ied land­mark dis­cov­er­ies in bac­te­ri­ol­ogy be­tween 1877 and 1899, also found that sci­en­tists 36 to 45 years of age were re­spon­si­ble for a dis­pro­por­tion­ate num­ber of these dis­cov­er­ies, even after he cor­rected for the likely age dis­tri­b­u­tion of sci­en­tists in the to­tal pop­u­la­tion. In con­trast, younger sci­en­tists (35 years and younger) and older sci­en­tists (46 to 65 years) were rel­a­tively un­der­rep­re­sent­ed. Fi­nal­ly, Over (1988), who used pub­li­ca­tions in Psy­cho­log­i­cal Re­view as his cri­te­rion for out­stand­ing con­tri­bu­tions (ad­mit­tedly a less de­mand­ing cri­te­rion than that of land­mark dis­cov­er­ies, even though Psy­cho­log­i­cal Re­view is one of the top jour­nals of our dis­ci­pline), found a sim­i­lar curvi­lin­ear dis­tri­b­u­tion that peaked for in­di­vid­u­als who were 12 to 17 years past their PhDs (i.e., ages 38 to 45 years) and de­clined there­after. How­ev­er, Over (1988) ar­gued that be­cause 60% of Amer­i­can psy­chol­o­gists ac­tive in re­search be­tween 1965 and 1980 were un­der 40, one could ex­pect that about 60% of the ar­ti­cles ap­pear­ing in Psy­cho­log­i­cal Re­view in this pe­riod would be au­thored by psy­chol­o­gists un­der the age of 40. In fact, 59.9% of the ar­ti­cles in his sam­ple were pub­lished by au­thors who were 0 to 11 years past their PhDs. Thus, de­spite the less de­mand­ing cri­te­ri­on, the curvi­lin­ear re­la­tion­ship be­tween age and sci­en­tific achieve­ment re­ported here is sim­i­lar to that found in stud­ies of No­bel lau­re­ates.

The pat­tern of find­ings of these early stud­ies is sim­i­lar to that found in the stud­ies of No­bel lau­re­ates and sci­en­tists with lesser achieve­ments, with age be­ing curvi­lin­early re­lated to sci­en­tific pro­duc­tiv­i­ty, which reaches a peak around ages 40 to 45 and then drops off (e.g., Bayer & Dut­ton, 1977; Cole, 1979; Den­nis, 1956; Horner, Rush­ton, & Ver­non, 1986; Kyvik, 1990; Over, 1982). This pat­tern was repli­cated in cross-sec­tional (Bayer & Dut­ton, 1977; Cole, 1979; Kyvik, 1990) and lon­gi­tu­di­nal or crossse­quen­tial stud­ies (Den­nis, 1956; Over, 1982; Horner et al., 1986) con­ducted in the United States (Bayer & Dut­ton, 1977; Cole, 1979; Horner et al., 1986) and Eu­rope (Den­nis, 1956; Kyvik, 1990; Over, 1982). How­ev­er, not all dis­ci­plines showed this pat­tern (Levin & Stephan, 1989). But the only dis­ci­pline in which a dis­crepant pat­tern has been repli­cated re­peat­edly is math­e­mat­ics. Sev­eral stud­ies of sam­ples of math­e­mati­cians re­sulted in a lin­ear re­la­tion­ship, with nei­ther an in­crease nor a de­cline in pro­duc­tiv­ity (Cole, 1979; Stern, 1978). Three ex­am­ples of stud­ies suffice to il­lus­trate the typ­i­cal pat­terns found in this re­search area. In one of the most ex­ten­sive cross-sec­tional stud­ies, Cole (1979) com­pared the pub­li­ca­tion rates in the years from 1965 to 1969 of 2,460 sci­en­tists from six differ­ent dis­ci­plines, in­clud­ing psy­chol­o­gy. Fig­ure 2 presents the over­all pro­duc­tiv­ity for the six fields com­bined, as well as the over­all ci­ta­tion rate. As the fig­ure in­di­cates, age is curvi­lin­earily re­lated to both pro­duc­tiv­ity and ci­ta­tions. Over­all, the rates for pro­duc­tiv­ity and ci­ta­tions peaked around age 40 and then dropped off. This re­la­tion­ship was valid for all dis­ci­plines, ex­cept for math­e­mat­ics, for which the re­la­tion­ship was lin­ear, “sup­port­ing the con­clu­sion that pro­duc­tiv­ity does not differ sig­nifi­cantly with age” (Cole, 1979, p. 965). Cole thus repli­cated the find­ings of Stern (1978), who con­cluded from her cross-sec­tional study that “the no­tion that younger math­e­mati­cians are, as it were, ‘phys­i­o­log­i­cally’ more able to pro­duce pa­pers would ap­pear to be in er­ror. In gen­er­al, we can state cat­e­gor­i­cally that age ex­plains very lit­tle, if any­thing, about pro­duc­tiv­ity” (p. 134). Two cross-se­quen­tial stud­ies of psy­chol­o­gists were con­ducted by Over (1982) and Horner et al. (1986). Over (1982) an­a­lyzed the re­la­tion­ship be­tween age and pro­duc­tiv­ity of a small sam­ple of British psy­chol­o­gists rang­ing in age from 26 to 65 years. These in­di­vid­u­als were as­sessed twice, once in 1968 -1970 and a sec­ond time in 1978 - 1980. British psy­chol­o­gists in gen­eral pub­lished as fre­quently in 1978 -1980 as in 1968 -1970 (i.e., there was no pe­riod effec­t). How­ev­er, both the cross-sec­tional and the lon­gi­tu­di­nal analy­ses in­di­cated that psy­chol­o­gists over 45 years of age pub­lished sig­nifi­cantly less fre­quently than their younger col­leagues. The pub­li­ca­tion rates cor­re­lated .49 across the two times of mea­sure­ment, in­di­cat­ing sub­stan­tial sta­bil­ity of in­di­vid­ual pro­duc­tiv­i­ty. Over (1982) con­cluded that “a per­son’s pre­vi­ous re­search pro­duc­tiv­ity was a far bet­ter pre­dic­tor of sub­se­quent re­search out­put than age was” (p. 519). An­other cross-se­quen­tial analy­sis on sci­en­tific pro­duc­tiv­ity was based on 1,084 Amer­i­can aca­d­e­mic psy­chol­o­gists and was con­ducted by Horner et al. (1986). Both the cross-sec­tional and the lon­gi­tu­di­nal analy­ses re­sulted in a curvi­lin­ear re­la­tion­ship be­tween age and pro­duc­tiv­i­ty. On av­er­age, the pro­duc­tiv­ity at ages 35 to 44 was sig­nifi­cantly higher than the pro­duc­tiv­ity at younger and older ages. Again, the cor­re­la­tions be­tween an in­di­vid­u­al’s num­ber of pub­li­ca­tions at differ­ent pe­ri­ods in­di­cated a great deal of sta­bil­i­ty. Fi­nal­ly, age ac­counted on av­er­age for only 6.9% of the vari­ance across time (more for low than for high pub­lish­er­s). The find­ings of these early stud­ies al­low four con­clu­sions: (a) The over­whelm­ing ma­jor­ity of stud­ies re­ported an age-re­lated de­cline in pro­duc­tiv­ity (indi­cated by num­ber of ar­ti­cles pub­lished), and most stud­ies found the as­so­ci­a­tion to be curvi­lin­ear, with a peak some­where around the ages of 40 to 45 years. (b) Even though there was a curvi­lin­ear re­la­tion­ship be­tween age and pro­duc­tiv­i­ty, age ac­counted for less than 8% of the vari­ance in pro­duc­tiv­i­ty. In math­e­mat­ics, the re­la­tion­ship be­tween age and pro­duc­tiv­ity even ap­pears to be lin­ear, with age be­ing un­re­lated to pro­duc­tiv­i­ty. (c) In con­trast, past per­for­mance was by far the best pre­dic­tor of fu­ture pro­duc­tiv­i­ty. As Si­mon­ton (2002) es­ti­mat­ed, “Be­tween one third to two thirds of the vari­ance in pro­duc­tiv­ity in any given pe­riod may be pre­dicted from the in­di­vid­ual differ­ence ob­served in the pre­vi­ous pe­riod” (p. 86). (d) Fi­nal­ly, even if older re­searchers are some­what less pro­duc­tive than their younger col­leagues, the qual­ity of their work (as re­flected by ci­ta­tions) ap­pears to be no less high. Over (1988) cor­re­lated the num­ber of ci­ta­tions each ar­ti­cle pub­lished in Psy­cho­log­i­cal Re­view had re­ceived in the first five years after pub­li­ca­tion with the age of the ar­ti­cle’s au­thor and found that the cor­re­la­tion was not sig­nifi­cantly differ­ent from ze­ro. Sim­i­lar find­ings were re­ported by Si­mon­ton (1985) in a study of the im­pact of the pub­li­ca­tions of 10 psy­chol­o­gists who had re­ceived the APA’s Award for Dis­tin­guished Sci­en­tific Con­tri­bu­tions. He found that the ra­tio of high­-im­pact pub­li­ca­tions to to­tal out­put fluc­tu­ated ran­domly through­out their ca­reers.

Al­though a re­cent lon­gi­tu­di­nal analy­sis of the as­so­ci­a­tion of age and pro­duc­tiv­ity for 112 em­i­nent mem­bers of the U.S. Na­tional Acad­emy of Sci­ences also re­sulted in a non­lin­ear re­la­tion­ship (Feist, 2006), this re­la­tion­ship was differ­ent from that re­ported in most ear­lier stud­ies. Three un­con­di­tional growth curve mod­els were con­struct­ed. The best fit to the data was achieved with a cu­bic mod­el, pro­vid­ing “pop­u­la­tion es­ti­mates on pro­duc­tiv­ity that in­crease rapidly un­til ap­prox­i­mately 20 years into one’s ca­reer, then flat­ten over the next 15 years, and then rise again over the last 5-year in­ter­val” (Feist, 2006, p. 29). Be­cause these in­di­vid­u­als started pub­lish­ing their first ar­ti­cles be­tween 22 and 25 years of age, they would have reached their first peak around age 45. After a 15-year lev­el­ing-off pe­ri­od, their pro­duc­tiv­ity would in­crease again after age 60. A some­what differ­ent pat­tern was re­ported by Joy (2006), who ex­am­ined the pub­li­ca­tion data of 1,216 fac­ulty mem­ber from 96 schools rang­ing from elite re­search uni­ver­si­ties to mi­nor un­der­grad­u­ate col­leges. Data were col­lected in 2004. Fig­ure 3 presents the mean num­ber of pub­li­ca­tions per year by ca­reer age (i.e., years since re­ceiv­ing the PhD) of ful­l-time fac­ulty mem­bers at three ho­mo­ge­neous sub­groups of in­sti­tu­tions. In the con­text of the fo­cus of this ar­ti­cle, I re­strict my­self to dis­cussing the data for the 399 fac­ulty mem­bers as­so­ci­ated with re­search uni­ver­si­ties (e.g., Prince­ton Uni­ver­si­ty, the Uni­ver­sity of Mass­a­chu­setts at Amher­st, North­east­ern Uni­ver­si­ty). These aca­d­e­mics pub­lished more dur­ing the first five years of their ca­reers than in later years; their pro­duc­tiv­ity re­mained es­sen­tially sta­ble for the next 25 years, with per­haps a slight in­crease be­tween the 26th and 30th years of their ca­reers. Thus, the data for fac­ulty mem­bers at re­search uni­ver­si­ties (or for those at other in­sti­tu­tions) failed to show the pat­tern re­ported in ear­lier stud­ies, in which pro­duc­tiv­ity reached a peak around ages 40 to 45 and then dropped off (Bayer & Dut­ton, 1977; Cole, 1979; Den­nis, 1956; Horner et al., 1986).

This study was based on 6,388 pro­fes­sors and re­searchers who had pub­lished at least one jour­nal ar­ti­cle over the eight-year pe­riod from 2000 to 2007. The study used 10-year age cat­e­gories, rang­ing from age 20 to age 70. Two differ­ent sets of data were used in com­pil­ing av­er­age pro­duc­tiv­i­ty, name­ly, the av­er­age pro­duc­tiv­ity of all pro­fes­sors and that of ac­tive pro­fes­sors who had pub­lished at least one jour­nal ar­ti­cle at the age in ques­tion. Al­though the as­so­ci­a­tion be­tween age and pro­duc­tiv­ity was curvi­lin­ear for both sam­ples, only the to­tal sam­ple showed a de­cline after age 50. For the ac­tive pro­fes­sors, pro­duc­tiv­ity in­creased to age 50 and then stayed at the same level un­til age 70. (There were too few older pro­fes­sors to ex­tend the study be­yond age 70.) Thus, these ac­tive pro­fes­sors sus­tained their pro­duc­tiv­ity at a high level through­out their ca­reers. There was also no de­cline in qual­ity for the group of ac­tive pro­fes­sors. In fact, the av­er­age num­ber of ar­ti­cles they pub­lished in high­-im­pact jour­nals (i.e., the top 1% cited jour­nals) rose steadily to age 70, and so did the av­er­age num­ber of ar­ti­cles that were among the top 10% of highly cited ar­ti­cles. The find­ings of Gin­gras et al. (2008) are dis­crepant with prac­ti­cally all of the early re­search. Given that, as noted above, the province of Que­bec had al­ready abol­ished com­pul­sory re­tire­ment in 1980, this change would offer a plau­si­ble ex­pla­na­tion for the fact that pro­duc­tiv­ity did not de­cline for the older age group.

“Cre­ative ca­reers: the life cy­cles of No­bel lau­re­ates in eco­nom­ics”, Wein­berg & Galen­son 2005

This pa­per stud­ies life cy­cle cre­ativ­ity among No­bel lau­re­ate econ­o­mists. We iden­tify two dis­tinct life cy­cles of schol­arly cre­ativ­i­ty. Ex­per­i­men­tal in­no­va­tors work in­duc­tive­ly, ac­cu­mu­lat­ing knowl­edge from ex­pe­ri­ence. Con­cep­tual in­no­va­tors work de­duc­tive­ly, ap­ply­ing ab­stract prin­ci­ples. We find that con­cep­tual in­no­va­tors do their most im­por­tant work ear­lier in their ca­reers than ex­per­i­men­tal lau­re­ates. For in­stance, our es­ti­mates im­ply that the prob­a­bil­ity that the most con­cep­tual lau­re­ate pub­lishes his sin­gle best work peaks at age 25 com­pared to the mid-50s for the most ex­per­i­men­tal lau­re­ate. Thus while ex­pe­ri­ence ben­e­fits ex­per­i­men­tal in­no­va­tors, new­ness to a field ben­e­fits con­cep­tual in­no­va­tors.

We mea­sure the im­por­tance of work us­ing ci­ta­tions. Ci­ta­tions were col­lected from the Web of Sci­ence, an on-line data­base com­pris­ing the So­cial Sci­ence Ci­ta­tion In­dex, the Sci­ence Ci­ta­tion In­dex, and the Arts and Hu­man­i­ties Ci­ta­tion In­dex.1 We col­lected the num­ber of ci­ta­tions to all works in each year of each lau­re­ate’s ca­reer made be­tween 1980 and 1999 in­clu­sive.2 These data on ci­ta­tions to the works each lau­re­ate pub­lished in each year of his ca­reer are our units of analy­sis. For the pur­pose of the em­pir­i­cal analy­sis, lau­re­ates are in­cluded in our sam­ple from the time they re­ceived their doc­tor­ate or from the time of their first cited pub­li­ca­tion if it pre­ceded their doc­tor­ate or if they never earned a doc­tor­ate.

The im­por­tance of schol­ars de­pends pri­mar­ily on their most im­por­tant con­tri­bu­tions. We use two meth­ods to iden­tify the years in which the lau­re­ates made im­por­tant con­tri­bu­tions. One method is to iden­tify all years in which ci­ta­tions are above a thresh­old. To do this, we first es­ti­mate the mean and stan­dard de­vi­a­tion of each lau­re­ate’s an­nual ci­ta­tions. We de­fine years in which a lau­re­ate’s ci­ta­tions were at least 2 of his stan­dard de­vi­a­tions above his mean to be his two stan­dard de­vi­a­tion peaks. To es­ti­mate the year in which each lau­re­ate made his sin­gle most im­por­tant con­tri­bu­tion, we also con­sider the sin­gle year with the most ci­ta­tions for each lau­re­ate. We re­fer to this year as the lau­re­ate’s sin­gle best year.

…Given the range of our in­dex of 201, the im­plied differ­ence in mean age of im­por­tant con­tri­bu­tions be­tween the most ex­per­i­men­tal and most con­cep­tual lau­re­ates is 20.5 years. The sec­ond col­umn shows anal­o­gous re­sults for the sin­gle best years. Here each lau­re­ate ap­pears ex­actly one time and Ageij de­notes the age at which lau­re­ate i had his sin­gle best year. For the sin­gle best years, a 1 point in­crease in the in­dex cor­re­sponds to a .113 year re­duc­tion in the mean age. Given the range of our in­dex, the im­plied differ­ence in mean ages of the sin­gle best years be­tween the most ex­per­i­men­tal and most con­cep­tual lau­re­ates, is 22.7 years.

…The most con­cep­tual lau­re­ate’s prob­a­bil­ity of a two stan­dard de­vi­a­tion peak is 15% in the first year of the ca­reer and it reaches a peak at age 28.8 years. For the most ex­per­i­men­tal lau­re­ate, the prob­a­bil­ity of two stan­dard de­vi­a­tion peak is less than half of a per­cent at the be­gin­ning of the ca­reer, reach­ing a peak at age 56.9, close to dou­ble the age of the most con­cep­tual lau­re­ate. By com­par­ison, the mean lau­re­ate’s pro­file peaks at age 47.1.

The pro­files for the sin­gle best years are be­neath those for the two stan­dard de­vi­a­tion peaks be­cause there are fewer sin­gle best years than two stan­dard de­vi­a­tion peaks. There is lit­tle differ­ence in the shape of the pro­files be­tween the two stan­dard de­vi­a­tion peaks and sin­gle best years for the most ex­per­i­men­tal lau­re­ates - both peak in the mid 50s. For the most con­cep­tual lau­re­ate the prob­a­bil­ity of a sin­gle best year is close to that of an im­por­tant year at the be­gin­ning of the ca­reer, but in­creases less be­fore drop­ping. For the most con­cep­tual lau­re­ate, the prob­a­bil­ity of a sin­gle best year peaks at age at age 24.8.

“Age and Out­stand­ing Achieve­ment: What do We Know After a Cen­tury of Re­search?”, Si­mon­ton 1988

One em­pir­i­cal gen­er­al­iza­tion ap­pears to be fairly se­cure: If one plots cre­ative out­put as a func­tion of age, pro­duc­tiv­ity tends to rise fairly rapidly to a defi­nite peak and there­after de­cline grad­u­ally un­til out­put is about half the rate at the peak (see, e.g., S. Cole, 1979; Den­nis, 1956b, 1966; Lehman, 1953a; but see Di­a­mond, 1986). In crude terms, if one tab­u­lates the num­ber of con­tri­bu­tions (e.g., pub­li­ca­tions, paint­ings, com­po­si­tions) per time unit, the re­sult­ing lon­gi­tu­di­nal fluc­tu­a­tions may be de­scribed by an in­verted back­ward-J curve (Si­mon­ton, 1977a). Ex­pressed more math­e­mat­i­cal­ly, pro­duc­tive out­put, say p(t), over a ca­reer tends to be roughly ap­prox­i­mated by a sec­ond-order poly­no­mial of the form

p(t) = b1 + b2t + b3t^2 (1)

…In ap­ply­ing this equa­tion, the in­de­pen­dent vari­able, t, is not chrono­log­i­cal age but rather ca­reer or pro­fes­sional age, where t = 0 at the on­set of the ca­reer (see Bayer & Dut­ton, 1977; Lyons, 1968). How­ev­er, in prac­tice, chrono­log­i­cal age is often used in lieu of ca­reer age, a sub­sti­tu­tion jus­ti­fied by their high cor­re­la­tion (e.g., r = .87, ac­cord­ing to Bayer & Dut­ton, 1977). …be­yond a cer­tain value of t, the pre­dicted level of pro­duc­tiv­ity be­comes neg­a­tive, a mean­ing­less out­come if out­put is gauged by sin­gle con­tri­bu­tions or items. 2 In­stead, the curve tends to ap­proach the zero pro­duc­tiv­ity rate more or less as­ymp­tot­i­cal­ly, a ten­dency that im­plies that a third-order poly­no­mial in time may fit the data more pre­cisely (Si­mon­ton, 1984a). The ad­di­tion of fur­ther terms would also serve to re­move an­other fault of a sim­ple qua­drat­ic, name­ly, that it im­plies that the pre- and post­peak slopes are roughly equal, which is sel­dom true in fact (el. Diemer, 1974).

At one ex­treme, some fields are char­ac­ter­ized by rel­a­tively early peaks, usu­ally around the early 30s or even late 20s in chrono­log­i­cal units, with some­what steep de­scents there­after, so that the out­put rate be­comes less than one-quar­ter the max­i­mum. This age­wise pat­tern ap­par­ently holds for such en­deav­ors as lyric po­et­ry, pure math­e­mat­ics, and the­o­ret­i­cal physics, for ex­am­ple (Adams, 1946; Den­nis, 1966; Lehman, 1953a; Moulin, 1955; Roe, 1972b; Si­mon­ton, 1975a; Van Heerin­gen & Dijk­wel, 1987). At the con­trary ex­treme, the typ­i­cal trends in other en­deav­ors may dis­play a leisurely rise to a com­par­a­tively late peak, in the late 40s or even 50s chrono­log­i­cal­ly, with a min­i­mal if not largely ab­sent drop-off after­ward. This more elon­gated curve holds for such do­mains as novel writ­ing, his­to­ry, phi­los­o­phy, med­i­cine, and gen­eral schol­ar­ship, for in­stance (Adams, 1946; Richard A. Davis, 1987; Den­nis, 1966; Lehman, 1953a; Si­mon­ton, 1975a). Of course, many dis­ci­plines ex­hibit age curves some­what be­tween these two outer lim­its, with a max­i­mum out­put rate around chrono­log­i­cal age 40 and a no­table yet mod­er­ate de­cline there­after (see, e.g., Ful­ton & Trow, 1974; Her­mann, 1988; Mc­Dow­ell, 1982; Zhao & Jiang, 1986). Out­put in the last years ap­pears at about half the rate ob­served in the peak years. Pro­duc­tive con­tri­bu­tions in psy­chol­o­gy, as an ex­am­ple, tend to adopt this tem­po­ral pat­tern (Homer et al., 1986; Lehman, 1953b; Over, 1982a, 1982b; Zus­ne, 1976). It must be stressed that these in­ter­dis­ci­pli­nary con­trasts do not ap­pear to be ar­bi­trary but in­stead have been shown to be in­vari­ant across differ­ent cul­tures and dis­tinct his­tor­i­cal pe­ri­ods (Lehman, 1962). As a case in point, the gap be­tween the ex­pected peaks for po­ets and prose au­thors has been found in every ma­jor lit­er­ary tra­di­tion through­out the world and for both liv­ing and dead lan­guages (Si­mon­ton, 1975a). In­deed, be­cause an ear­lier pro­duc­tive op­ti­mum means that a writer can die younger with­out loss to his or her ul­ti­mate rep­u­ta­tion, po­ets ex­hibit a life ex­pectan­cy, across the globe and through his­to­ry, about a half dozen years less than prose writ­ers do (Si­mon­ton, 1975a). This cross-cul­tural and tran­shis­tor­i­cal in­vari­ance strongly sug­gests that the age curves re­flect un­der­ly­ing psy­cho­log­i­cal uni­ver­sals rather than ar­bi­trary so­cio­cul­tural de­ter­mi­nants.

In­di­vid­ual differ­ences in life­time out­put are sub­stan­tial (Si­mon­ton, 1984b, chap. 5; 1988b, chap. 4). So skewed is the cross-sec­tional dis­tri­b­u­tion of to­tal con­tri­bu­tions that a small per­cent­age of the work­ers in any given do­main is re­spon­si­ble for the bulk of the work. Gen­er­al­ly, the top 10% of the most pro­lific elite can be cred­ited with around 50% of all con­tri­bu­tions, whereas the bot­tom 50% of the least pro­duc­tive work­ers can claim only 15% of the to­tal work, and the most pro­duc­tive con­trib­u­tor is usu­ally about 100 times more pro­lific than the least (Den­nis, 1954b, 1955; also see Lotka, 1926; Price, 1963, chap. 2). Now from a purely log­i­cal per­spec­tive, there are three dis­tinct ways of achiev­ing an im­pres­sive life­time out­put that en­ables a cre­ator to dom­i­nate an artis­tic or sci­en­tific en­ter­prise. First, the in­di­vid­ual may ex­hibit ex­cep­tional pre­coc­i­ty, be­gin­ning con­tri­bu­tions at an un­com­monly early age. Sec­ond, the in­di­vid­ual may at­tain a no­table life­time to­tal by pro­duc­ing un­til quite late in life, and thereby dis­play pro­duc­tive longevi­ty. Third, the in­di­vid­ual may boast phe­nom­e­nal out­put rates through­out a ca­reer, with­out re­gard to the ca­reer’s on­set and ter­mi­na­tion. These three com­po­nents are math­e­mat­i­cally dis­tinct and so may have al­most any ar­bi­trary cor­re­la­tion what­so­ever with each oth­er, whether pos­i­tive, neg­a­tive, or ze­ro, with­out al­ter­ing their re­spec­tive con­tri­bu­tions to to­tal pro­duc­tiv­i­ty. In pre­cise terms, it is clear that O = R ( L P), where O is life­time out­put, R is the mean rate of out­put through­out the ca­reer, L is the age at which the ca­reer ended (longevi­ty), and P is the age at which the ca­reer be­gan (pre­coc­i­ty). The cor­re­la­tions among these three vari­ables may adopt a wide range of ar­bi­trary val­ues with­out vi­o­lat­ing this iden­ti­ty. For ex­am­ple, the differ­ence L - P, which de­fines the length of a ca­reer, may be more or less con­stant, man­dat­ing that life­time out­put re­sults largely from the av­er­age out­put rate R, given that those who be­gin ear­lier, end ear­lier, and those who be­gin lat­er, end lat­er. Or out­put rates may be more or less con­stant, forc­ing the fi­nal score to be a func­tion solely of pre­coc­ity and longevi­ty, ei­ther singly or in con­junc­tion. In short, R, L, and P, or out­put rate, longevi­ty, and pre­coc­i­ty, com­prise largely or­thog­o­nal com­po­nents of O, the gauge of to­tal con­tri­bu­tions. When we turn to ac­tual em­pir­i­cal data, we can ob­serve two points. First, as might be ex­pect­ed, pre­coc­i­ty, longevi­ty, and out­put rate are each strongly as­so­ci­ated with fi­nal life­time out­put, that is, those who gen­er­ate the most con­tri­bu­tions at the end of a ca­reer also tend to have be­gun their ca­reers at ear­lier ages, ended their ca­reers at later ages, and pro­duced at ex­tra­or­di­nary rates through­out their ca­reers (e.g., Al­bert, 1975; Black­burn et al., 1978; Bloom, 1963; Clemente, 1973; S. Cole, 1979; Richard A. Davis, 1987; Den­nis, 1954a, 1954b; Hel­son & Crutch­field, 1970; Lehman, 1953a; Over, 1982a, 1982b; Rask­in, 1936; Roe, 1965, 1972a, 1972b; Segal, Busse, & Mans­field, 1980; R. J. Si­mon, 1974; Si­mon­ton, 1977c; Zhao & Jiang, 1986). Sec­ond, these three com­po­nents are con­spic­u­ously linked with each oth­er: Those who are pre­co­cious also tend to dis­play longevi­ty, and both pre­coc­ity and longevity are pos­i­tively as­so­ci­ated with high out­put rates per age unit (Black­burn et al., 1978; Den­nis, 1954a, 1954b, 1956b; Horner et al., 1986; Lehman, 1953a, 1958; Lyons, 1968; Roe, 1952; Si­mon­ton, 1977c; Zuck­er­man, 1977). The re­la­tion be­tween longevity and pre­coc­ity be­comes par­tic­u­larly ev­i­dent when care is first taken to con­trol for the im­pact of differ­en­tial life span (Den­nis, 1954b). Be­cause those who are very pro­lific at a pre­co­cious age can afford to die young and still end up with a re­spectable life­time out­put, a neg­a­tive re­la­tion emerges be­tween pre­coc­ity and life span, ne­ces­si­tat­ing that ca­reers be equal­ized on life span be­fore the cor­re­la­tion co­effi­cients are cal­cu­lated (Si­mon­ton, 1977c; Zhao & Jiang, 1986).

When Lehman (1953a) com­pared tab­u­la­tions of su­pe­rior con­tri­bu­tions in a wide range of cre­ative ac­tiv­i­ties against those for works of lesser mer­it, he con­cluded that the age curves ob­tained were in­deed con­tin­gent on the qual­ity cri­te­rion uti­lized in con­struct­ing the counts. For the most part, the peak pro­duc­tive age tended to stay rel­a­tively sta­ble, only the peak was far more pro­nounced when only ex­cep­tional works were tab­u­lated (see also Lehman, 1958, 1966a). In con­trast, when the stan­dards of ex­cel­lence were loos­ened, the age curves flat­tened out ap­pre­cia­bly, and the post­peak de­cline was much less con­spic­u­ous. This gen­er­al­iza­tion was largely repli­cated by Den­nis (1966)

When such pre­cau­tions are tak­en, very differ­ent re­sults emerge (Si­mon­ton, 1977a, 1984b, chap. 6, 1985b, 1988b, chap. 4). First, if one cal­cu­lates the age curves sep­a­rately for ma­jor and mi­nor works within ca­reers, the re­sult­ing func­tions are ba­si­cally iden­ti­cal. Both fol­low the same sec­ond-order poly­no­mial (as seen in Equa­tion 1), with roughly equal pa­ra­me­ters. Sec­ond, if the over­all age trend is re­moved from the with­in-ca­reer tab­u­la­tions of both quan­tity and qual­i­ty, mi­nor and ma­jor con­tri­bu­tions still fluc­tu­ate to­geth­er. Those pe­ri­ods in a cre­ator’s life that see the most mas­ter­pieces also wit­ness the great­est num­ber of eas­ily for­got­ten pro­duc­tions, on the av­er­age. An­other way of say­ing the same thing is to note that the “qual­ity ra­tio,” or the pro­por­tion of ma­jor prod­ucts to to­tal out­put per age unit, tends to fluc­tu­ate ran­domly over the course of any ca­reer. The qual­ity ra­tio nei­ther in­creases nor de­creases with age nor does it as­sume some curvi­lin­ear form. These out­comes are valid for both artis­tic (e.g., Si­mon­ton, 1977a) and sci­en­tific (e.g., Si­mon­ton, 1985b) modes of cre­ative con­tri­bu­tion (see also Al­paugh, Ren­ner,& Bir­ren, 1976, p. 28). What these two re­sults sig­nify is that if we se­lect the con­tri­bu­tion rather than the age pe­riod as the unit of analy­sis, then age be­comes ir­rel­e­vant to de­ter­min­ing the suc­cess of a par­tic­u­lar con­tri­bu­tion. For in­stance, the num­ber of ci­ta­tions re­ceived by a sin­gle sci­en­tific ar­ti­cle is not con­tin­gent upon the age of the re­searcher (Oro­man­er, 1977). The lon­gi­tu­di­nal link­age be­tween quan­tity and qual­ity can be sub­sumed un­der the more gen­eral “con­stan­t-prob­a­bil­i­ty-of­suc­cess model” of cre­ative out­put (Si­mon­ton, 1977a, 1984b, 1985b, 1988b, chap. 4). Ac­cord­ing to this hy­poth­e­sis, cre­ativ­ity is a prob­a­bilis­tic con­se­quence of pro­duc­tiv­i­ty, a re­la­tion­ship that holds both within and across ca­reers. Within sin­gle ca­reers, the count of ma­jor works per age pe­riod will be a pos­i­tive func­tion of to­tal works gen­er­ated each pe­ri­od, yield­ing a qual­ity ra­tio that ex­hibits no sys­tem­atic de­vel­op­men­tal trends. And across ca­reers, those in­di­vid­ual cre­ators who are the most pro­duc­tive will also tend, on the av­er­age, to be the most cre­ative: In­di­vid­ual vari­a­tion in quan­tity is pos­i­tively as­so­ci­ated with vari­a­tion in qual­i­ty. There is abun­dant ev­i­dence for the ap­pli­ca­tion of the con­stan­t-prob­a­bil­i­ty-of-suc­cess model to cross-sec­tional con­trasts in quan­tity and qual­ity of out­put (Richard A. Davis, 1987; Si­mon­ton, 1984b, chap. 6; 1985b, 1988b, chap. 4). In the sci­ences, for ex­am­ple, the rep­u­ta­tion of a nine­teen­th­cen­tury sci­en­tist in the twen­ti­eth cen­tu­ry, as judged by en­tries in stan­dard ref­er­ence works, is pos­i­tively cor­re­lated with the to­tal num­ber of pub­li­ca­tions that can be claimed (Den­nis, 1954a; Si­mon­ton, 1981 a; see also Den­nis, 1954c). Sim­i­lar­ly, the num­ber of ci­ta­tions a sci­en­tist re­ceives, which is a key in­di­ca­tor of achieve­ment, is a pos­i­tive func­tion of to­tal pub­li­ca­tions (Cran­dall, 1978; Richard A. Davis, 1987; My­ers, 1970; Rush­ton, 1984), and to­tal pro­duc­tiv­ity even cor­re­lates pos­i­tively with the ci­ta­tions earned by a sci­en­tist’s three best pub­li­ca­tions (J. R. Cole & S. Cole, 1973, chap. 4). Need­less to say, the cor­re­la­tions be­tween quan­tity and qual­ity are far from per­fect for ei­ther lon­gi­tu­di­nal or cross-sec­tional da­ta.

Last­ly, a long rule means an abun­dance of events from which we can con­struct per­for­mance in­di­ca­tors (Si­mon­ton, 1984d). To il­lus­trate these po­ten­tial as­sets, an in­quiry was made into the ca­reers of 25 Eu­ro­pean kings and queens from over a dozen na­tion­s–­such as Queen Eliz­a­beth I, Fred­er­ick the Great, Ivan the Ter­ri­ble, and Suleiman the Mag­nifi­cen­t–that found that most ob­jec­tive per­for­mance in­di­ca­tors ei­ther de­cline with age or else ex­hibit a curvi­lin­ear in­vert­ed-U func­tion that max­i­mized at the 42nd year of life, this lat­ter curve hold­ing for some mea­sures of mil­i­tary and diplo­matic suc­cess (Si­mon­ton, 1984c). What made this study sen­si­tive to lon­gi­tu­di­nal changes was the fact that none of the sam­pled lead­ers ruled fewer than 36 years, and the av­er­age reign length was 43 years, giv­ing ca­reer du­ra­tions more com­pa­ra­ble to those found in the re­search on dis­tin­guished cre­ativ­i­ty.

An anal­o­gous age gap ap­pears be­tween rev­o­lu­tion­ar­ies and lead­ers of long-estab­lished po­lit­i­cal in­sti­tu­tions. Al­though as many as half of the no­table rev­o­lu­tion­ar­ies were younger than 35 (Re­jai & Phillips, 1979), very few of the world’s po­lit­i­cal lead­ers at­tained power be­fore age 40 (Blondel, 1980). In­deed, just as po­ets can die younger than prose writ­ers and still achieve a durable rep­u­ta­tion (Si­mon­ton, 1975a), so the pre­dom­i­nant youth­ful­ness of rev­o­lu­tion­ar­ies be­trays it­self in a lower life ex­pectan­cy. In the Cox (1926) sam­ple of 301 ge­nius­es, who had an over­all life span mean of 66 years, the rev­o­lu­tion­ar­ies av­er­aged only 51 years, not one liv­ing to be 80 and more than 44% dy­ing prior to age 50. These fig­ures con­trast dra­mat­i­cally with the states­men in Cox’s sam­ple who op­er­ated un­der more sta­tus quo con­di­tions; their life ex­pectancy was 70, only about 5% lived fewer than 50 years, and fully 30% sur­vived to their 80th birth­day. Fur­ther­more, these re­sults are en­larged by the find­ing that as po­lit­i­cal in­sti­tu­tions ma­ture, the age of their lead­ers in­creases as well (Lehman, 1953a, chap. 17). In the United States, for ex­am­ple, mem­bers of the House of Rep­re­sen­ta­tives and the Sen­ate, House speak­ers, cab­i­net offi­cers, Supreme Court jus­tices, am­bas­sadors, and army com­man­ders have all got­ten older and older since the na­tion’s found­ing, trends that can­not be ex­plained by cor­re­spond­ing en­hance­ments in gen­eral life ex­pectancy (see also Si­mon­ton, 1985c, 1987d, chap. 4). In­deed, tran­shis­tor­i­cal data have con­sis­tently shown that the mean life span has not sig­nifi­cantly changed over the cen­turies but rather has stayed close to around 65 years (see, e.g., Si­mon­ton, 1975a, 1977c; Zhao & Jiang, 1986), a fig­ure close to the “three­-s­core years and ten” said by Solon to be the nor­mal term of a hu­man life way back in an­cient Greece.

What­ever the spe­cific pre­cau­tions tak­en, once the in­tru­sion of the com­po­si­tional fal­lacy has been de­nied, the em­pir­i­cal re­suits dis­cussed ear­lier in this re­view yet per­sist, al­beit the de­cline may not be so pro­nounced as it some­times looks in many pub­lished da­ta. The lo­ca­tion of the age peak is sin­gu­larly im­mune from this con­sid­er­a­tion, and for good cause (Lehman, 1962). The num­ber of in­di­vid­u­als who died be­fore they would be ex­pected to reach their peak age for achieve­ment is quite small (Zhao & Jiang, 1986; cf. Bul­lough et al., 1978). Only 11% of Cox’s (1926) sam­ple failed to at­tain the 50th year, which comes about a decade after the ex­pected peak for most ac­tiv­i­ties. To be sure, po­ets die young, yet their age op­ti­mum is cor­re­spond­ingly younger. And even if the peak age for lead­er­ship some­times oc­curs after the 50th year, the life ex­pectancy of lead­ers is older in rough pro­por­tion. Al­though the con­cern of most re­searchers has been on how the com­po­si­tional fal­lacy may in­tro­duce an ar­ti­fac­tual de­cline, it is clear that it may im­pede ac­cu­rate in­fer­ences in other ways as well. Most no­tably, those stud­ies men­tioned ear­lier that claim to have di­vulged sad­dle-shaped age func­tions may ac­tu­ally have failed to seg­re­gate dis­tinc­tive achieve­ment do­mains that har­bor dis­crepant peaks (Si­mon­ton, 1984a). For ex­am­ple, if achieve­ment in pure math­e­mat­ics peaks at an ear­lier age than that in ap­plied math­e­mat­ics, then ag­gre­gat­ing across both types of con­tri­bu­tions will per­force gen­er­ate a dou­ble-peak age curve (of. Den­nis, 1966). Hence, the er­rors of ag­gre­ga­tion can be very per­va­sive.

Many in­ves­ti­ga­tors pin­pointed a de­cline in in­tel­lec­tual power in the later years of life (or at least a drop in “fluid” as op­posed to “crys­tal­lized” in­tel­li­gence) (e.g., Horn, 1982), and oth­ers re­ported sin­gle-peak func­tions and neg­a­tive age slopes for cer­tain cre­ativ­ity mea­sures as well (Al­paugh & Bir­ren, 1977; Al­paugh, Parham, Cole, &Bir­ren, 1982; Brom­ley, 1956; Cor­nelius & Caspi, 1987; Eisen­man, 1970; Mc­Crae et al., 1987; Ruth & Bir­ren, 1985; cf. Jaquish & Rip­ple, 1981). Yet the de­fen­sive­ness noted twice ear­lier in this es­say may have pro­voked the de­bate that fol­lowed these pub­lished re­sults, a con­tro­versy about whether the de­creases with age were real or sim­ply re­flected some per­ni­cious age bias. Some of the is­sues in this de­bate were the same re­cur­rent method­olog­i­cal ques­tions that plague life span de­vel­op­men­tal re­search, es­pe­cially the po­ten­tial ar­ti­fact in­tro­duced by de­pend­ing on cross-sec­tional data when in­fer­ring lon­gi­tu­di­nal trends (Ko­gan, 1973; Ro­manuik & Ro­manuik, 1981; Schaie & Strother, 1968).

To be­gin with, even ifa min­i­mal level of in­tel­li­gence is req­ui­site for achieve­ment, be­yond a thresh­old of around IQ 120 (the ac­tual amount vary­ing across field­s), in­tel­lec­tual prowess be­comes largely ir­rel­e­vant in pre­dict­ing in­di­vid­ual differ­ences in ei­ther cre­ativ­ity or lead­er­ship (Si­mon­ton, 1985a).

The spe­cific re­la­tion be­tween age and out­stand­ing achieve­ment is by no means a purely aca­d­e­mic is­sue. Yuasa (1974) ar­gued that by the year 2000 a de­cline in sci­ence in the United States is in­evitable be­cause of the shift­ing age struc­ture of Amer­i­can sci­en­tists (see also Oro­man­er, 1981). More specifi­cal­ly, when the mean age of U.S. sci­en­tists at­tains the 50th year, the United States will soon be re­placed by some other na­tion as the cen­ter of sci­en­tific ac­tiv­ity (Zhao & Jiang, 1985). An anal­o­gous “Yuasa phe­nom­e­non” may at­tend achieve­ment in other do­mains as well. Yet this fore­cast is pred­i­cated on the no­tion that the slope of the age func­tion is neg­a­tive after some peak in the late 30s or early 40s. De­spite the con­sid­er­able em­pir­i­cal and the­o­ret­i­cal cor­rob­o­ra­tion this pos­tu­late pos­sess­es, more doc­u­men­ta­tion is nec­es­sary be­fore this prog­no­sis of doom (for U.S. cit­i­zens anx­ious for No­bel prizes) projects full force, It bears re­peat­ing that the age struc­ture of Amer­i­can so­ci­ety very much hinges on the baby boomers, and this gen­er­a­tion is only about a decade away from the crit­i­cal age when the United States may wit­ness it­self sup­planted by some up­start na­tion. Ad­mit­ted­ly, en­hanced knowl­edge may give us no means to re­verse in­ex­orable his­tor­i­cal trends (cf. Al­paugh et al., 1976), yet we can at least have the con­so­la­tion of un­der­stand­ing why the lo­cus in out­stand­ing achieve­ment strayed from our own shores (Si­mon­ton, 1984b, chap. 10).

  1. An ex­am­ple: Sin­ga­pore’s gov­ern­ment re­port­edly ex­pects half the pop­u­la­tion to hold at least a bach­e­lor’s de­gree by 2020. This is surely doable, just as the US could ex­pect half its pop­u­la­tion to grad­u­ate high school, just like in­dus­tri­al­iz­ing coun­tries can shift their pro­le­tariat to the cities & fac­to­ries. But this is a trick you can do only on­ce! You can’t have half your pop­u­la­tion join the first half in the cities, leav­ing 1% to han­dle the mech­a­nized agri­cul­ture - for tremen­dous eco­nomic growth - and then have an­other half move into the cities to keep the growth go­ing, be­cause there is no third half. Sim­i­larly for de­grees. Bach­e­lor’s, per­haps; mas­ter’s, may­be; but PhD? Not with ex­ist­ing pop­u­la­tions or gene pools.↩︎

  2. One es­ti­mate of the in­crease from Jones 2006 is the fac­tor is 19x.↩︎

  3. “Tri­als and Er­rors: Why Sci­ence Is Fail­ing Us”, Wired 2011↩︎

  4. “The Truly Stag­ger­ing Cost Of In­vent­ing New Drugs”, 2012↩︎

  5. “In­side Pfiz­er’s palace coup”, For­tune↩︎

  6. “Drugs That Are as Smart as Our Dis­eases”, WSJ↩︎