Comments on: Nootropics Survey Results And Analysis

By: Patrick

Patrick — Wed, 04 Jun 2014 09:11:19 +0000

I agree with your comments about nootropic users tending to ‘outpace’ the research.

We definitely need more research on nootropics. While not a RCT your survey and analysis is a good start.

Awesome in-depth article!
Thanks

By: Douglas Knight

Douglas Knight — Thu, 17 Apr 2014 03:48:07 +0000

But the reason that breastfed babies have high IQs is that their parents have have high IQs – the effect goes away if you control for income and SES. (and I think there was one study that controlled directly for maternal IQ)

Report comment

By: Nootriment

Nootriment — Thu, 17 Apr 2014 02:28:24 +0000

There has been some suggestion that one of the reasons babies who are breastfed have a higher IQ than babies who are raised on formula might have to do with the choline. Specifically, Alpha GPC choline can be found in small amounts of breastmilk while it is not typically included in formula. The brain needs adequate choline to develop new neurons within the first few weeks of gestation, so supplementing with Alpha GPC (or getting enough of it in your diet) could potentially have this effect.

Report comment

By: Michael Mouse

Michael Mouse — Sat, 22 Feb 2014 07:14:10 +0000

I vote this Genuine On-Topic Reasonable-quality Post Most Resembling Comment Spam Of The Year, and consider it a hot contender for ‘Of The Decade’ too.

Report comment

By: Rachael

Rachael — Fri, 21 Feb 2014 10:24:11 +0000

No. Looks interesting. Although it looks (from a casual skim of the Amazon page) like the sleepless in that world are the minority, so quite a different premise.

Added to wishlist

Report comment

By: gwern

gwern — Fri, 21 Feb 2014 00:29:05 +0000

Since I had that analysis of just the ratings done, a thought kept nagging at me: ‘what about the subjects’ age, dose, and history? can’t we throw those in as covariates as well? it can’t be that big a change to the reshape code…’

Well, as usual it wasn’t as easy as I hoped, but I think I got it right (source):

library(reshape) library(lme4) nootropics <- read.csv("http://dl.dropboxusercontent.com/u/182368464/2014-ssc-nootropicssurvey.csv") nootropics$Timestamp <- NULL; nootropics$Directions <- NULL # unnecessary nootropics$CholineHistory <- NA # omitted from original but absence breaks symmetry of the triplet of fields for each nootropic


# it's simpler to grep over the column names than list each and every of the 31 columns by hand

# and we need to sort because having to add 'CholineHistory' screws up the otherwise-sorted-hardwired-columns

experiences <- sort(Filter(function (x) {grepl("Experience", x)}, names(nootropics)))

doses <- sort(Filter(function (x) {grepl("Dose", x)}, names(nootropics)))

histories <- sort(Filter(function (x) {grepl("History", x)}, names(nootropics)))

long <- reshape(nootropics, idvar = "Subject", timevar="Nootropic", v.names=c("Response", "Dose", "History"), times=experiences, varying=list(experiences, doses, histories), direction = "long")
lmr1 <- lmer(Response ~       (1|Subject) + (1               |Nootropic), data=long)

lmr2 <- lmer(Response ~       (1|Subject) + (    Dose        |Nootropic), data=long)

lmr3 <- lmer(Response ~       (1|Subject) + (         History|Nootropic), data=long)

lmr4 <- lmer(Response ~       (1|Subject) + (    Dose+History|Nootropic), data=long)

lmr5 <- lmer(Response ~ Age + (1|Subject) + (    Dose+History|Nootropic), data=long)

lmr6 Chisq)

lmr1  4 5036 5056  -2514     5028

lmr2  6 3555 3583  -1772     3543  1484      2     <2e-16

lmr3  6 3854 3883  -1921     3842     0      0          1

lmr4  9 2893 2933  -1437     2875   967      3     <2e-16

lmr5 10 3006 3050  -1493     2986     0      1          1

lmr6 13 3064 3121  -1519     3038     0      3          1
summary(lmr4)

...

Random effects:

 Groups    Name        Variance Std.Dev. Corr

 Subject   (Intercept) 6.29e+00 2.507269

 Nootropic (Intercept) 4.29e+00 2.070161

           Dose        6.97e-08 0.000264 -1.000

           History     4.39e-06 0.002096 -0.797  0.797

 Residual              3.18e+00 1.783544

Number of obs: 639, groups: Subject, 121; Nootropic, 27
Fixed effects:

            Estimate Std. Error t value

(Intercept)     5.47       0.34    16.1
rr1 <- ranef(lmr1, postVar=TRUE)

rr4 <- ranef(lmr4, postVar=TRUE)

noots4 <- rr4$Nootropic

noots4[order(noots4$'(Intercept)'), , drop = FALSE]

                          (Intercept)       Dose    History

GingkoExperience             -3.11497  3.973e-04  3.884e-03

DMAEExperience               -2.32433  2.964e-04  1.666e-03

VitaminDExperience           -2.15718  2.751e-04  4.044e-04

ALCARExperience              -1.44127  1.838e-04  1.150e-03

PiracetamExperience          -1.36419  1.740e-04  1.917e-03

VinpocetineExperience        -1.25651  1.603e-04  1.197e-03

BacopaExperience             -1.18497  1.511e-04  1.837e-03

LionapossManeExperience      -1.15578  1.474e-04  9.508e-04

HordenineExperience          -1.15456  1.472e-04  9.246e-04

SulbutiamineExperience       -1.04894  1.338e-04  1.335e-03

HuperzineExperience          -0.76957  9.815e-05  7.428e-04

GinsengExperience            -0.42476  5.417e-05  6.865e-04

AniracetamExperience         -0.30357  3.872e-05  1.129e-03

NoopeptExperience            -0.29668  3.784e-05  7.789e-04

CentrophenoxineExperience    -0.19564  2.495e-05  1.389e-04

OxiracetamExperience         -0.14437  1.841e-05  4.861e-04

CreatineExperience           -0.12480  1.592e-05 -6.631e-04

AshwagandhaExperience        -0.12103  1.544e-05 -5.619e-05

PramiracetamExperience        0.03029 -3.863e-06  3.963e-04

RhodiolaExperience            0.21512 -2.744e-05  1.100e-04

PhenylpiracetamExperience     0.21959 -2.801e-05  1.032e-04

ColuracetamExperience         0.35140 -4.482e-05 -9.800e-05

AdrafinilExperience           0.48123 -6.137e-05 -2.582e-04

TheanineExperience            0.49004 -6.250e-05 -1.659e-05

CaffeineExperience            1.04602 -1.334e-04  1.169e-04

ModafinilExperience           2.47240 -3.153e-04 -5.729e-04

ArmodafinilExperience         2.83790 -3.619e-04 -2.132e-03
qq4 <- qqmath(rr4)

# http://i.imgur.com/aQu8Igv.png

qq4$Nootropic

# Like before, compare simple nootropics analysis with dose+history+nootropics: # what changed the most in absolute terms (least to greatest)? tmp <- merge(rr1$Nootropic, rr4$Nootropic[1], by="row.names", all=T) tmpDelta <- abs(tmp$"(Intercept).x" - tmp$"(Intercept).y") tmp[order(tmpDelta), , drop = FALSE] Row.names (Intercept).x (Intercept).y 20 ModafinilExperience 2.40961 2.47240 17 HuperzineExperience -0.70487 -0.76957 8 CentrophenoxineExperience -0.30221 -0.19564 1 AdrafinilExperience 0.59810 0.48123 28 TheanineExperience 0.64818 0.49004 5 AshwagandhaExperience 0.09079 -0.12103 26 RhodiolaExperience 0.46314 0.21512 22 OxiracetamExperience 0.11794 -0.14437 16 HordenineExperience -0.88016 -1.15456 4 ArmodafinilExperience 2.53205 2.83790 23 PhenylpiracetamExperience 0.52651 0.21959 25 PramiracetamExperience 0.39316 0.03029 11 ColuracetamExperience 0.71994 0.35140 30 VinpocetineExperience -0.87629 -1.25651 3 AniracetamExperience 0.12498 -0.30357 12 CreatineExperience 0.31701 -0.12480 18 LionapossManeExperience -0.70730 -1.15578 21 NoopeptExperience 0.19493 -0.29668 13 DMAEExperience -1.81606 -2.32433 7 CaffeineExperience 1.59909 1.04602 15 GinsengExperience -1.12405 -0.42476 6 BacopaExperience -0.40310 -1.18497 2 ALCARExperience -0.59846 -1.44127 27 SulbutiamineExperience -0.02826 -1.04894 14 GingkoExperience -1.84778 -3.11497 24 PiracetamExperience -0.09529 -1.36419 31 VitaminDExperience -0.47647 -2.15718

Interpretation gets harder as you throw in more variables, but oh well. I’m also a little dubious about the Dose variable: when you plot the random effects, there seems to be much less variation than I’d expect. I wonder if the reshaping to a long format somehow put them all on the same scale or something?

Report comment

By: gwern

gwern — Thu, 20 Feb 2014 15:50:08 +0000

Have you read https://en.wikipedia.org/wiki/Nancy_Kress#The_Sleepless ?

Report comment

By: Rachael

Rachael — Thu, 20 Feb 2014 09:33:43 +0000

Enjoying all the interesting posts lately!

I thought about filling in the nootropics survey but then didn’t because I thought I wasn’t its target audience, having only tried one substance other than caffeine. I tried modafinil because I wanted to be less tired and need less sleep, but it had no effect on me. My husband also tried it and it had no effect on him either. It looks from your results as though we’re actually quite unusual in this.

I’ve considered writing a dystopia in which use of drugs like modafinil becomes ubiquitous and most people become more productive and need less sleep, and the few people who are resistant to it become an underclass who can’t keep up professionally or socially.

Report comment

By: gwern

gwern — Wed, 19 Feb 2014 21:42:31 +0000

This amazing and super useful… as it allows you to control for individual differences in response patterns (even account for non-linear scale differences – e.g. if someone interprets the scale to be logarithmic rather than linear).

I’ve used lme4 before a few times*, but always in a relatively simple nested design, rather than the ‘wide’ format of this survey…

I spent the last 2 days working on it, and I think I’ve finally got it right. (I took a long detour into using the eRm library to try to apply item-response models directly instead of googling to see how someone would use lme4 for a Rasch model; turns out the models don’t like being fed sparse data with lots of NAs and unrepresented levels, and would have preferred the most obscure nootropics like hordenine to be omitted entirely.) An hour or two of struggling with reshape later… It looks like the subject-specific effects don’t change the nootropics estimates very much, even though subjects did vary a lot in their default rating level. (Nootropic random effects, subject) That said, I don’t understand how I would use lme4 to correct for scale differences as you imply: all the coefficients are the usual intercept+beta stuff, no?

Anyway, my analysis (better formatted copy of the R code):

library(reshape) library(lme4) nootropics <- read.csv("http://dl.dropboxusercontent.com/u/182368464/2014-ssc-nootropicssurvey.csv")


# not going to mess with doses or schedules: just want the ratings of the most-popular nootropics

nootropicsResponses <- with(nootropics, data.frame (AdrafinilExperience=AdrafinilExperience, AniracetamExperience=AniracetamExperience, ArmodafinilExperience=ArmodafinilExperience, AshwagandhaExperience=AshwagandhaExperience, BacopaExperience=BacopaExperience, CILTEPExperience=CILTEPExperience, CaffeineExperience=CaffeineExperience, CentrophenoxineExperience=CentrophenoxineExperience, CholineExperience=CholineExperience, ColuracetamExperience=ColuracetamExperience, CreatineExperience=CreatineExperience, DMAEExperience=DMAEExperience, GingkoExperience=GingkoExperience, GinsengExperience=GinsengExperience, HordenineExperience=HordenineExperience, HuperzineExperience=HuperzineExperience, LionapossManeExperience=LionapossManeExperience, MCTOilExperience=MCTOilExperience, ModafinilExperience=ModafinilExperience, NoopeptExperience=NoopeptExperience, OxiracetamExperience=OxiracetamExperience, PhenylpiracetamExperience=PhenylpiracetamExperience, PiracetamExperience=PiracetamExperience, PramiracetamExperience=PramiracetamExperience, RhodiolaExperience=RhodiolaExperience, SulbutiamineExperience=SulbutiamineExperience, TULIPExperience=TULIPExperience, TheanineExperience=TheanineExperience, VinpocetineExperience=VinpocetineExperience, VitaminDExperience=VitaminDExperience ))

# convert each of the '*Experience' columns into a single level of a generic 'Nootropic' factor

long <- reshape(nootropicsResponses, idvar = "Subject", v.names="Response", timevar="Nootropic", times=names(nootropicsResponses), varying=list(names(nootropicsResponses)), direction = "long")

summary(long)

  Nootropic            Response       Subject

 Length:4620        Min.   : 0     Min.   :  1.0

 Class :character   1st Qu.: 3     1st Qu.: 39.0

 Mode  :character   Median : 6     Median : 77.5

                    Mean   : 5     Mean   : 77.5

                    3rd Qu.: 8     3rd Qu.:116.0

                    Max.   :10     Max.   :154.0

                    NA's   :3561
# non-nested design: we're examining a crossover design of Subject x Nootropics, not a nested design like Subject(Nootropics).

# Nootropics is a random effect: each level/nootropics gets its own estimate; and likewise, each Subject gets their own estimate

# see http://www.jstatsoft.org/v20/i02/paper "Estimating the Multilevel Rasch Model: With the lme4 Package" for another example

lmr1 <- lmer(Response ~ (1|Subject) + (1|Nootropic), data=long); summary(lmr1)
...REML criterion at convergence: 4875
Random effects:

 Groups    Name        Variance Std.Dev.

 Subject   (Intercept) 2.98     1.73

 Nootropic (Intercept) 1.24     1.11

 Residual              4.42     2.10

Number of obs: 1059, groups: Subject, 150; Nootropic, 30
Fixed effects:

            Estimate Std. Error t value

(Intercept)    4.995      0.271    18.4
# do Subjects really vary enough to justify 154 random effects rather than 1 fixed effects?

lmr2 Chisq)

lmr1  4 4882 4902  -2437     4874

lmr2  4 5162 5182  -2577     5154     0      0          1
# extract the random effects for both Subjects & Nootropics

rr <- ranef(lmr1, postVar=TRUE)
# visualize the two sets of random-effects

qq <- qqmath(rr)

qq$Nootropic

# http://i.imgur.com/PmrHwqA.png

qq$Subject

# http://i.imgur.com/a0nJiOk.png
# view nootropics sorted by coefficient size

noots <- rr$Nootropic

noots[order(noots$'(Intercept)'), , drop = FALSE]

                          (Intercept)

GingkoExperience             -1.85454

DMAEExperience               -1.84701

GinsengExperience            -1.13313

CholineExperience            -1.08120

HordenineExperience          -0.90128

VinpocetineExperience        -0.87606

HuperzineExperience          -0.72931

LionapossManeExperience      -0.72782

VitaminDExperience           -0.51186

MCTOilExperience             -0.46230

BacopaExperience             -0.41610

CentrophenoxineExperience    -0.33918

CILTEPExperience             -0.26260

PiracetamExperience          -0.10523

SulbutiamineExperience       -0.02941

OxiracetamExperience          0.08505

AshwagandhaExperience         0.09323

AniracetamExperience          0.10354

NoopeptExperience             0.16937

CreatineExperience            0.29738

PramiracetamExperience        0.36177

RhodiolaExperience            0.44556

PhenylpiracetamExperience     0.50122

AdrafinilExperience           0.59758

TheanineExperience            0.63526

ColuracetamExperience         0.69391

TULIPExperience               0.80629

CaffeineExperience            1.58431

ModafinilExperience           2.39321

ArmodafinilExperience         2.50935
# what does the alternate version look like?

rr2 <- ranef(lmr2, postVar=TRUE)

noots2 <- rr2$Nootropic

noots2[order(noots2$'(Intercept)'), , drop = FALSE]

                          (Intercept)

GingkoExperience            -1.399810

GinsengExperience           -1.188626

CholineExperience           -0.978048

DMAEExperience              -0.954214

VitaminDExperience          -0.901817

LionapossManeExperience     -0.662886

VinpocetineExperience       -0.611086

MCTOilExperience            -0.457433

HuperzineExperience         -0.437104

BacopaExperience            -0.320166

HordenineExperience         -0.241617

CentrophenoxineExperience   -0.233822

PiracetamExperience         -0.148546

AshwagandhaExperience       -0.135738

SulbutiamineExperience      -0.015350

CILTEPExperience             0.003327

CreatineExperience           0.012964

TULIPExperience              0.144545

RhodiolaExperience           0.207083

AniracetamExperience         0.326791

PramiracetamExperience       0.354544

OxiracetamExperience         0.374706

NoopeptExperience            0.407759

TheanineExperience           0.515546

PhenylpiracetamExperience    0.568206

AdrafinilExperience          0.594847

ColuracetamExperience        0.778954

CaffeineExperience           1.234636

ArmodafinilExperience        1.456277

ModafinilExperience          1.706078

# What did correcting for subject-specific ratings buy us? # How much did nootropic rankings or estimates change from # the alternative? subjectDelta <- rr$Nootropic - rr2$Nootropic sum(subjectDelta$'(Intercept)') [1] 1.767e-12 subjectDelta[order(subjectDelta$'(Intercept)'), , drop = FALSE] (Intercept) DMAEExperience -0.892796 HordenineExperience -0.659664 GingkoExperience -0.454732 HuperzineExperience -0.292204 OxiracetamExperience -0.289652 CILTEPExperience -0.265925 VinpocetineExperience -0.264976 NoopeptExperience -0.238387 AniracetamExperience -0.223253 CentrophenoxineExperience -0.105358 CholineExperience -0.103155 BacopaExperience -0.095930 ColuracetamExperience -0.085046 PhenylpiracetamExperience -0.066984 LionapossManeExperience -0.064930 SulbutiamineExperience -0.014057 MCTOilExperience -0.004862 AdrafinilExperience 0.002729 PramiracetamExperience 0.007223 PiracetamExperience 0.043313 GinsengExperience 0.055499 TheanineExperience 0.119715 AshwagandhaExperience 0.228967 RhodiolaExperience 0.238472 CreatineExperience 0.284411 CaffeineExperience 0.349678 VitaminDExperience 0.389960 TULIPExperience 0.661741 ModafinilExperience 0.687129 ArmodafinilExperience 1.053072 # so the biggest difference was DMAE falls by +1 rating # and Armodafinil looks better by +1 (and Modafinil improves too)

Sort of an odd setup, interpretation-wise: a negative coefficient here means simply that something was below the average (the intercept of 4.995 or 5), not that it was useless or harmful.

* in http://www.gwern.net/Google%20Alerts / http://www.gwern.net/Lewis%20meditation / http://www.gwern.net/LSD%20microdosing / http://www.gwern.net/Lunar%20sleep / http://www.gwern.net/Weather

Report comment

By: Dave

Dave — Wed, 19 Feb 2014 05:31:09 +0000

I am surprised none of the amphetamines seem to be included here. I assume you decided to leave them out of the survey because nootropic enthusiasts are not posting about them. But that seems odd to me.

I have read that 20%+ of students at some schools have used prescription stimulants as performance enhancers. That figure is probably unrepresentative, but even a much more conservative guess would make these drugs the most widely used nootropics outside of coffee and nicotine. Are people not discussing them because they are too hard to get, or too illegal? Drug policy does make it very inconvenient to fill your Adderall prescriptions.

And while we’re at it, what about nicotine?

Report comment