Data from Mark and Heapost 2014 ("Soome-ugri rahvaste füüsiline antropoloogia")

[Komintasavalta]

I used data from Heapost (2008) ("Pigmentation variation in Finno-Ugric peoples") to make the plots below (https://www.etis.ee/Portal/Publicati...8-b9b3010eabad). It is based on the same data collected by Karin Mark as the book by Mark and Heapost from 2014.

Eastern Finns were more light-haired than Western Finns and also slightly more light-eyed. Among Finland-Swedes, people from Åland were lighter than people from Närpes (Ostrobothnia) or Liljendal (Uusimaa). Skolt Saami had about as dark hair as Inari Saami but much lighter eyes, even though it could've been because of a small sample size (40).



Among other Finnic peoples, Vepsians were darker than Karelians. Estonians from the Northern Estonian dialect area were lighter than Estonians from the Southern Estonian dialect area.



The plots are based on this data:

(average eye color;average hair color;sample size;location;population)
.49;2.80;100;Haapsalu;Northern Estonians
.28;2.62;98;Lihula;Northern Estonians
.37;2.67;98;Audru;Northern Estonians
.34;2.75;100;Rapla;Northern Estonians
.28;2.71;101;Põltsamaa;Northern Estonians
.28;2.46;100;Rakvere;Northern Estonians
.34;2.42;104;Kohtla-Järve;Northern Estonians
.35;2.95;74;Iisaku;Northern Estonians
.43;2.86;104;Kilingi-Nõmme;Northern Estonians
.52;2.68;100;Karksi;Northern Estonians
.34;2.58;100;Otepää;Southern Estonians
.34;2.81;102;Põlva;Southern Estonians
.44;2.92;100;Võru;Southern Estonians
.40;2.85;63;Karkolye;Izhorians
.29;2.96;100;Soikino;Izhorians
.39;2.90;51;Kurgolovo;Ingrian Finns
.34;2.86;105;Askola;Western Finns
.42;2.87;92;Mynämäki;Western Finns
.41;3.01;85;Kokemäki;Western Finns
.38;3.06;112;Kurikka;Western Finns
.30;2.97;96;Hauho;Western Finns
.24;2.64;140;Keuruu;Eastern Finns
.38;2.76;146;Ristiina;Eastern Finns
.30;2.54;101;Kiuruvesi;Eastern Finns
.29;2.84;112;Kesälahti;Eastern Finns
.27;2.65;116;Ylitornio;Northern Finns
.28;2.79;143;Kuusamo;Eastern Finns
.37;2.95;145;Salla;Northern Finns
.30;2.81;82;Savukoski;Northern Finns
.46;2.80;92;Kalevala;Karelians
.39;2.87;46;Kolatselga;Karelians
.51;2.87;96;Olonets;Karelians
.44;3.12;100;Girva;Karelians
.53;3.06;66;Ozyora;Vepsians
.55;3.10;42;Sidorovo;Vepsians
.97;3.56;63;Kola Saami;Saami
.30;3.19;40;Skolt Saami;Saami
.73;3.19;49;Inari Saami;Saami
1.02;3.31;40;Northern Saami;Saami
.29;2.84;124;Åland;Finland-Swedes
.42;3.08;146;Närpes;Finland-Swedes
.44;3.02;140;Liljendal;Finland-Swedes

[Komintasavalta]

Here's two different clustering methods applied to data by Karin Mark: CLARA and K-means. "/tmp/mark" is this file: https://pastebin.com/raw/iVMLBgse.

library(ggplot2);library(cluster);library(factoextra)
t<-scale(read.csv("/tmp/mark",sep=";",header=T,row.names=1))
fviz_cluster(clara(t,6,samples=50));ggsave("/tmp/clara.png")
fviz_cluster(kmeans(t,6,nstart=25),data=t);ggsave("/tmp/kmeans2.png")

PC1 differentiates populations with higher Mongoloid ancestry from populations with lower Mongoloid ancestry. PC2 differentiates Saami and Northeastern Finns from everyone else, so it has high positive loadings for head width, cephalic index, minimum frontal width, bizygomatic width, nose width, and nasal index, and it has high negative loadings for face height and facial index. For PC3, Kola Saami have the highest value and Estonians have the lowest value. It has the highest loading for the trait "% no inner eyelid fold" and the lowest
loadings for bigonial width, minimum frontal width, head length plus width, and bizygomatic width. On PC4, Votic-Izhorians and Ingrians have the highest value, and the traits "% wavy nose" and "nose height (Soviet)" have the highest loading.

library(pheatmap);library(RColorBrewer)
t<-prcomp(t(scale(read.csv("/tmp/mark",sep=";",header=T,row.names=1))))
pheatmap(t$rotation[,c(1:4)],cellwidth=10,cellheight=10,cluster_cols=F,rev(colorRampPalette(brewer.pal(5,"RdBu"))(256)),filename="/tmp/heat.png")

library(ggplot2)
t<-prcomp(scale(read.csv("/tmp/mark",sep=";",header=T,row.names=1)))$x
g<-c("Finnic","Finnic","Finnic","Finnic","Finnic","Finnic","Finnic","Finnic","Saami","Saami","Volga-Ural-Komi","Volga-Ural-Komi","Volga-Ural-Komi","Volga-Ural-Komi","Volga-Ural-Komi","Volga-Ural-Komi","Volga-Ural-Komi","Volga-Ural-Komi","Volga-Ural-Komi","Ob-Ugric","Ob-Ugric","Indo-European","Indo-European","Turkic","Turkic","Turkic")
ggplot(t,aes(x=PC3,y=PC4,label=rownames(t),color=g,fill=g))+geom_text(size=4,vjust=0,nudge_y=.1,show.legend=F)+geom_point()+theme(legend.title=element_blank())
ggsave("/tmp/markpc3pc4.png")

You can also make a 3D PCA view by copying the pastebin file linked above to the "SOURCE" tab here (without the header row and with the delimiter changed to comma): https://vahaduo.github.io/custompca/. Then go to the "PCA PLOT" tab, press "RUN PCA", press "PLOT PCA", and press "3D".

[Komintasavalta]

Here's a 3d plot where the color of the label indicates average hair color:

install.packages("plotly")
library(plotly)
download.file("https://pastebin.com/raw/hJb9X7LB","/tmp/mark")
t<-read.csv("/tmp/mark",row.names=1,header=T)
p<-plot_ly(t,x=~head.length,y=~head.width,z=~height,mode="text",text=row.names(t),color=~as.factor(average.hair.color),colors=c("#FFE1A1","#683531"))
layout(p,scene=list(xaxis=list(title="Head length"),yaxis=list(title="Head width"),zaxis=list(title="Body height")),paper_bgcolor="black",showlegend=F)

[Komintasavalta]

Here's a reverse biplot of the data from Mark and Heapost (2014) that was posted on Lucasz's website (https://antropologia-fizyczna.pl/sta...logia-fizyczna). I converted the columns of the table to z-scores before making the PCA. Each point is connected with a line to its three closest neighbors.

PC1 differentiates Mongoloids from Caucasoids. PC2 differentiates Saami and Finnics from VURers. The points of cranial width, bizygomatic width, bizygomatic width, and cephalic index are relatively high on the y-axis, because Saami and Finnics have higher head width than VURers.

The points of nose width and nasal index are between the arrows for Saami and Ob-Ugrics, because nose width was the highest in Kola Saami, followed by Finnish Saami, Khanty, and Mansi.

The bottom left quadrant contains the arrows for IEs and Mordvins. It contains points for some morphologically woggy traits like a high nasal bridge and high facial height.

The bottom right quadrant contains points for traits that are shared by Ob-Ugrics, Maris, Udmurts, and Turkic VURers, like for example dark hair color and dark eye color.

library(tidyverse)
library(colorspace)
library(ggforce)

t=read.table("https://pastebin.com/raw/cA90JaHp",sep=";",header=T,row.names=1,check.names=F)
t=t[,-grep("Western",colnames(t))]
t=t(scale(t))

p=prcomp(t)
p2=as.data.frame(p$x)
pct=paste0(colnames(p$x)," (",sprintf("%.1f",p$sdev/sum(p$sdev)*100),"%)")
k=as.factor(cutree(hclust(dist(t)),k=8))
ranges=apply(p2,2,function(x)abs(max(x)-min(x)))
p2$k=k
load=p$rotation

hue=c(0,30,60,120,180,210,260,310)
pal1=hex(HSV(hue,.5,1))
pal2=hex(HSV(hue,.35,1))

dist=as.data.frame(as.matrix(dist(as.matrix(t))))

for(i in seq(1,7,2)){
  seg=lapply(1:3+1,function(j)apply(dist,1,function(x)unlist(p2[names(sort(x)[j]),c(i,i+1)],use.names=F))%>%t%>%cbind(p2[,c(i,i+1)]))%>%do.call(rbind,.)%>%setNames(paste0("V",1:4))

  mult=min(max(p2[,i])/max(load[,i]),max(p2[,i+1])/max(load[,i+1]))

  maxrange=max(ranges[c(i,i+1)])
  lims=sapply(c(i,i+1),function(x)mean(range(p2[,x]))+c(-.54*maxrange,.54*maxrange))

  xpc=sym(paste0("PC",i))
  ypc=sym(paste0("PC",i+1))

  ggplot(p2,aes(!!xpc,!!ypc))+
  ggforce::geom_mark_hull(aes(group=k),color=pal1[k],fill=pal1[k],concavity=100,radius=unit(.15,"cm"),expand=unit(.15,"cm"),alpha=.2,size=.2)+
  geom_segment(data=seg,aes(x=V1,y=V2,xend=V3,yend=V4),color="black",size=.2)+
  geom_segment(data=load,aes(x=0,y=0,xend=mult*!!xpc,yend=mult*!!ypc),arrow=arrow(length=unit(.3,"lines")),color="gray90",size=.4)+
  annotate("text",x=(mult*load[,i]),y=(mult*load[,i+1]),label=rownames(load),size=2.5,vjust=ifelse(load[,i+1]>0,-.5,1.4),color="gray90")+
  geom_point(aes(x=!!xpc,y=!!ypc),color=pal1[k],size=.3)+
  geom_text_repel(aes(x=!!xpc,y=!!ypc,label=rownames(p2)),color=pal2[k],size=2,max.overlaps=Inf,force=4,force_pull=2,segment.size=.2,min.segment.length=.2,box.padding=.05)+
  # geom_text(aes(x=!!xpc,y=!!ypc,label=rownames(p2)),color=pal2[k],size=2,vjust=-.7)+
  labs(x=pct[i],y=pct[i+1])+
  coord_fixed(xlim=lims[,1],ylim=lims[,2],expand=F)+
  scale_x_continuous(breaks=seq(-10,10,1),expand=expansion(mult=.08))+
  scale_y_continuous(breaks=seq(-10,10,1))+
  theme(
    axis.text=element_text(color="gray80",size=6),
    axis.text.y=element_text(angle=90,vjust=1,hjust=.5),
    axis.ticks=element_blank(),
    axis.ticks.length=unit(0,"pt"),
    axis.title=element_text(color="gray80",size=8),
    legend.position="none",
    panel.background=element_rect(fill="gray25"),
    panel.border=element_rect(color="gray35",fill=NA,size=.4),
    panel.grid.major=element_line(color="gray35",size=.2),
    panel.grid.minor=element_blank(),
    plot.background=element_rect(fill="gray25",color=NA)
  )

  ggsave(paste0(i,".png"),width=7,height=7)
}

Here's also non-reverse biplots, where instead of drawing arrows for the loadings of traits, I only drew a point so there wouldn't be too many overlapping lines of arrows.

[Komintasavalta]

I made a CSV file for Mark's measurements converted to z-scores, and I used it to do two-way models with a modified version of michal3141's convex optimization script:

curl -Ls pastebin.com/raw/vxJ2TBJX|tr -d \\r>mix;chmod +x mix
pip3 install cvxpy
brew install R
Rscript -e 't=read.csv("https://pastebin.com/raw/cA90JaHp",sep=";",row.names=1,check.names=F);write.csv(round(scale(t),4),"markz",quote=F)'
cut -d, -f1 markz|sed 1d|while read l;do ./mix markz <(grep ^"$l", markz) -s -f0 -d1 -m2;done

Result (the number after population name is the distance of the model):

Estonian (4.6): 73% Finnish + 27% Ingrian
Votic-Izhorian (5.1): 67% Ingrian + 33% Izhorian
Izhorian (5.3): 56% Karelian + 44% Vepsian
Ingrian (5.3): 73% Votic-Izhorian + 27% Estonian
Finnish (4.5): 64% Estonian + 36% Karelian
Finnish (Northeast) (6.8): 50% Estonian + 50% Finnish Saami
Karelian (4.2): 59% Izhorian + 41% Moksha
Vepsian (4.8): 54% Izhorian + 46% Erzya
Finnish Saami (3.3): 66% Kola Saami + 34% Finnish (Northeast)
Kola Saami (5.4): 93% Finnish Saami + 7% Khanty
Erzya (3.0): 85% Russian (Middle Volga) + 15% Estonian
Moksha (4.5): 87% Russian (Middle Volga) + 13% Khanty
Teryukhan (6.9): 69% Russian (Middle Volga) + 31% Votic-Izhorian
Karatai (5.8): 70% Russian (Middle Volga) + 30% Udmurt
Mari (4.4): 59% Bashkir + 41% Karatai
Udmurt (4.5): 61% Komi-Permyak + 39% Besermyan
Besermyan (5.6): 51% Udmurt + 49% Mari
Komi-Permyak (3.7): 54% Udmurt + 46% Karelian
Komi-Zyrian (3.6): 64% Komi-Permyak + 36% Estonian
Mansi (4.3): 89% Khanty + 11% Finnish (Northeast)
Khanty (4.5): 90% Mansi + 10% Bashkir
Finland-Swede (6.0): 70% Estonian + 30% Russian (Middle Volga)
Russian (Middle Volga) (3.1): 81% Erzya + 19% Volga Tatar
Chuvash (4.0): 52% Bashkir + 48% Udmurt
Volga Tatar (3.5): 53% Bashkir + 47% Russian (Middle Volga)
Bashkir (4.6): 76% Volga Tatar + 24% Khanty

The Northeastern Finns are from Kuusamo, Salla, and Savukoski. Based on Mark's measurements, they seem to be heavily influenced by Saami, and even in the model above, they got 50% Saami ancestry.

I haven't seen genetic results Komi-Permyaks separated from Komi-Zyrians, but based on Mark's measurements, Komi-Permyaks seem to be more Mongoloid than Komi-Zyrians.

Besermyans are Muslim Udmurts whose dialect has influences from the Tatar language. In G25, Besermyans are slightly more western and southern than Udmurts, or they are shifted towards Tatars, but in the models above, Besermyans appear more to be Mongoloid than Udmurts.

There are four different groups of Mordvins, but all of them got at least 69% of ancestry from Russians in the Middle-Volga region. Mark's dataset includes measurements of Russians from 6 localities, which all had a Mordvin, Mari, or Udmurt population.

Izhorians from Soikino (Soikkola) got about half Karelian and half Vepsian ancestry, so they seem to be surprisingly eastern relative to their geographic location. The Votic-Izhorian group from the village of Krakolye (Joenperä) got 2/3 Ingrian ancestry and 1/3 Izhorian ancestry.

In G25, Kola Saami are less Mongoloid than other Saami, but based on the models above and the PCA in my previous post, Kola Saami appear more Mongoloid than Finnish Saami. I think it's because the Finnish Saami include 49 Inari Saami, 40 Skolts, and 40 Northern Saami. In Heapost (2008) which includes locality-level data for eye color and hair color, Inari Saami were much lighter than Northern Saami, and Skolts were much lighter than Inari Saami. However only about 5% of Finnish Saami are Skolts, so Skolts are overrepresented in Mark's dataset.

[travv]

Besermyans are Muslim Udmurts whose dialect has influences from the Tatar language.

Besermyans are Orthodox Christians but this word in the Russian chronicles meant Muslims.

[Veslan]

Interesting data. Cephalic index means surprised me, I thought Finno-Ugrian peoples were more strongly brachycephalic, but they seem to be nearly all in subbrachy range.

[Lucas]

Here's a reverse biplot of the data from Mark and Heapost (2014) that was posted on Lucasz's website (https://antropologia-fizyczna.pl/sta...logia-fizyczna). I converted the columns of the table to z-scores before making the PCA. Each point is connected with a line to its three closest neighbors.

I see you are still interested in this. You inspired me to add some more data for those region. Interested in your visualization later.

[Komintasavalta]

Interesting data. Cephalic index means surprised me, I thought Finno-Ugrian peoples were more strongly brachycephalic, but they seem to be nearly all in subbrachy range.

Here's traits sorted by their correlation coefficient with cephalic index. Mouth width has an even higher correlation than head width (maximum cranial breadth). It might be partially because some tall populations had high head width, or for example Estonians had the second highest head width (156.9), even though it's not necessarily that wide relative to their overall body size. The height of both lips had the third lowest correlation, which seems to fit in with the stereotype that long-skulled Nordids have thin lips, even though actually the height of both lips was even lower in Saami and most Finnic populations than in Finland-Swedes. The absolute value of the correlation coefficient was higher for face height (-.51) than bizygomatic width (.30).

> t=read.csv("https://pastebin.com/raw/cA90JaHp",sep=";",row.names=1,check.names=F)
> t2=scale(t)
> s=sort(cor(t2)[,"cephalic index"],decreasing=T)
> cat(paste(sprintf("%.2f",s),names(s)),sep="\n")
1.00 cephalic index
0.58 mouth width
0.55 head width
0.53 prominent vs receding chin
0.42 protrusion of cheekbones
0.41 downturned vs upturned nose
0.31 % no Mongolian fold
0.31 % strongly protruding cheekbones
0.30 bizygomatic width
0.29 % no inner eyelid fold
0.26 % light eyes
0.25 % brown hair
0.25 oval vs round nostrils
0.25 rectracted vs prominent upper lip
0.24 % concave nose
0.22 % narrow palpebral fissure width
0.20 minimum frontal width
0.19 bigonial width
0.19 % dark blonde hair
0.18 nose width
0.17 nasal index (Western)
0.17 nasal index (Soviet)
0.15 eye tilt
0.10 % blonde hair
0.09 % eyes tilted upwards
0.09 % light blonde hair
0.08 % wavy nose
0.04 beard density
0.04 parallel vs perpendicular nostrils
0.04 protruding vs flat nose
-0.00 straight vs tilted forehead
-0.04 height of nasal bridge
-0.04 nose height (Soviet)
-0.11 earlobe shape
-0.11 profiled vs flat face
-0.11 hair shape
-0.13 % flat nose
-0.14 outer eyelid fold
-0.15 % red hair
-0.18 height
-0.18 % flat facial profile
-0.21 average hair color
-0.21 head length + width
-0.21 % mixed eyes
-0.23 % dark brown and black hair
-0.24 palpebral fissure width
-0.24 strength of inner eyelid fold
-0.25 upper lip height
-0.26 % almost perpendicular nostrils
-0.26 % dark eyes
-0.27 average eye color
-0.29 % straight nose
-0.30 convex vs concave nose
-0.33 browridge prominence
-0.37 % convex nose
-0.45 eyebrow thickness
-0.51 face height (Soviet)
-0.58 facial index (converted to Western method)
-0.58 facial index (Soviet method measured from eyebrows)
-0.62 height of both lips
-0.66 mouth width-height index
-0.68 head length

Cephalic index has a surprisingly weak correlation with height, because it's only about -.18. I wonder if it would be stronger based on individual-level data, or for example based on all Estonian individuals in the dataset.

library(tidyverse)
library(ggrepel)

t=read.csv("https://pastebin.com/raw/cA90JaHp",sep=";",row.names=1,check.names=F)

xvar="cephalic index"
yvar="height"

xy=as.data.frame(cor(t)[,c(xvar,yvar)])
colnames(xy)=c("x","y")
xy$k=as.factor(cutree(hclust(dist(t(scale(t)))),12))

ggplot(xy,aes(x,y))+
geom_abline(linetype="dashed",color="gray80",size=.3,slope=-1)+
geom_polygon(data=xy%>%group_by(k)%>%slice(chull(x,y)),aes(color=k,fill=k),alpha=.15,size=.15)+
geom_point(aes(color=k),size=.5)+
geom_text_repel(aes(color=k),label=rownames(xy),size=2,max.overlaps=Inf,force=2,segment.size=.2,min.segment.length=.2,box.padding=.05)+
# geom_text(aes(color=k),label=rownames(xy),size=2,vjust=-.7)+
coord_cartesian(xlim=c(-1.1,1.1),ylim=c(-1.1,1.1),expand=F)+
scale_x_continuous(breaks=seq(-1,1,.2))+
scale_y_continuous(breaks=seq(-1,1,.2))+
labs(x=paste("Correlation with",xvar),y=paste("Correlation with",yvar))+
theme(
  axis.text=element_text(size=6),
  axis.text.y=element_text(angle=90,vjust=1,hjust=.5),
  axis.ticks.length=unit(0,"cm"),
  axis.ticks=element_blank(),
  axis.title=element_text(size=8),
  legend.position="none",
  panel.background=element_rect(fill="white"),
  panel.border=element_rect(color="gray80",fill=NA,size=.6),
  panel.grid.major=element_line(color="gray85",size=.2),
  panel.grid.minor=element_blank()
)

ggsave("a.png",width=6,height=6)

[travv]

I like her data.
Mari: brown, dark brown and black hair 92,4%.
Chuvash: brown, dark brown and black hair 95,3%.

And still lighter than most so called "blondes".

Spoiler!