Statistical Analysis of the Who’s Who of American Returned Students (1917)

Abstract

This document introduces the data from the Who’s Who of American returned students (Tsing Hua, 1917) and provides basic descriptive statistics on the population as well as more complex multivariate analyses.

Description

This initial aims to introduce the main dataset containing attribute data on the students (one row for each student).

library(readr)
main <- read_delim("data/main.csv", delim = ";", 
    escape_double = FALSE, trim_ws = TRUE)

head(main)

The main dataset includes 401 individuals (rows) and 34 columns (variables) which contains specific information on the students:

Type	Variables	Description
Identification	DocId	Unique identifier of the biography in the imh collection
	name_source	Name of the person, as given in the original source
	FullName_zh	Full name of the person, in Chinese
	FullName_py	Full name of the person, in pinyin transliteration
	FullName_wg	Full name of the person, in Wade-Giles transliteration
	gender	Gender (Male or Female)
	birth_year	Year of birth
	age_in_1917	Age in 1917 (date of publication)
	Birth_Province_zh	Province of birth, in Chinese
	Birth_Province_eng_standard	Province of birth, in English
	Birth_location_zh	Birthplace (city, town, or village), in Chinese
	Birth_Location_eng_standard	Birthplace (city, town, or village), in English
Family	ancestry_province	Province of ancestry (jiguan 籍貫), in Chinese
	ancestry_locality	Locality of ancestry, in Chinese
	married	Marital status (yihun 已婚, married; weihun 未婚, unmarried)
	married_year	Year of marriage
	sons	Number of sons in 1917
	daugthers	Number of daughters in 1917
	total_children	Total number of children in 1917
Education	FundingMain	Source of funding (normalized)
	Funding	Source of funding (from English biography)
	FundingZh	Source of funding (from Chinese biography)
	Number_visit	Number of times the person visited the United States
	estimated_arrived_year	Estimated year of arrival in the USA
	period	Period of study in the USA (periodization 1, based on the First Remission of the Boxer Indemnity: pre-Boxer (1872-1908), post-Boxer (1909-1917)
	period2	Period of study in the USA (periodization 2, based on the abolition of the imperial examinations: before (1872-1905), after (1906-1917)
	arrived_month	Month of arrival in the USA
	estimated_returned_year	Estimated year of return
	returned_month	Month of return
	duration	Duration of stay in the USA
	field_of_study	Field of study (lower level): a synthetic variable that collapses all the disciplines studied. For a detailed list of individual disciplines studied by the individual, see the specific datasets (“education” and “degrees”)
	field_group	Field of study (upper level): based on the typology designed for the Modern China Biographical Database (MCBD)
	Highest	Highest degree obtained (for a detailed list of degrees obtained by the person, refer to the specific dataset (“education”, “discipline”, “degree”)
Metadata	source_name	Source (title)
	source_page	Page number in source

Basic statistics

Gender

How many men and women?

library(dplyr)

main %>% group_by(gender) %>% count()

Timing

Year of Birth

When were they born?

library(ggplot2)
library(hrbrthemes)
library(viridis)

main %>% drop_na(birth_year) %>% 
  ggplot( aes(x=birth_year)) +
  geom_histogram( binwidth=1, fill="#69b3a2", color="#e9ecef", alpha=0.9) +
  ggtitle("Bin size = 1") +
  theme_ipsum() +
  theme(
    plot.title = element_text(size=15)
  ) +
  labs(title = "Early Liumei's Year of Birth",
       x = "Year",
       y = "Frequency", 
       caption = "Source: 遊美同學錄 (1917)")

Year of Arrival

When did they arrive in America?

main %>% drop_na(estimated_arrived_year) %>% 
  ggplot( aes(x=estimated_arrived_year)) +
  geom_histogram( binwidth=1, fill="#69b3a2", color="#e9ecef", alpha=0.9) +
  ggtitle("Bin size = 1") +
  theme_ipsum() +
  theme(
    plot.title = element_text(size=15)
  ) +
  labs(title = "Early Liumei's Year of Arrival",
       x = "Year",
       y = "Frequency", 
       caption = "Source: 遊美同學錄 (1917)")

Year of Return

When did they return to China?

main %>% drop_na(estimated_returned_year) %>% 
  ggplot( aes(x=estimated_returned_year)) +
  geom_histogram( binwidth=1, fill="#69b3a2", color="#e9ecef", alpha=0.9) +
  ggtitle("Bin size = 1") +
  theme_ipsum() +
  theme(
    plot.title = element_text(size=15)
  ) +
  labs(title = "Early Liumei's Year of Return",
       x = "Year",
       y = "Frequency", 
       caption = "Source: 遊美同學錄 (1917)")

Diachronic Comparison

Compare year of birth, arrival, and return:

# Reshape data for comparison 

library(tidyr)
years <- gather(main, key = "Event", value = "Year", birth_year, estimated_arrived_year, estimated_returned_year) %>% drop_na(Year) %>% mutate(Year = as.numeric(Year))

# Create plot
ggplot(years, aes(x = Year , fill = Event)) +
  geom_histogram(position = "identity", alpha = 0.8) +
  labs(title = "Three Steps in Early Liumei's Trajectories",
       subtitle = "Birth, Arrival, and Return", 
       x = "Year",
       y = "Frequency", 
       caption = "Source: 遊美同學錄 (1917)") +
  scale_fill_manual(values = c("birth_year" = "light green", "estimated_arrived_year" = "steelblue", "estimated_returned_year" = "red"), 
                    label = c("Birth", "Arrival", "Return")) +
  theme_minimal() + theme(legend.position = "bottom")

Birth Place

Where were they born?

Province

main %>% group_by(Birth_Province_zh, Birth_Province_eng_standard) %>% 
  count(sort = TRUE) %>% 
  mutate(percent = round(n/401*100, 2))

main %>% 
  drop_na(Birth_Province_zh) %>% 
  group_by(Birth_Province_zh) %>%
  count()  %>%
  ggplot(aes(reorder(Birth_Province_zh, n), n)) +
  geom_col() +
  coord_flip() + 
  labs(title = "Early Liumei's Province of Birth",
       x = "Province", 
       y = "Number of students",
       caption = "Source: 遊美同學錄 (1917)")

Locality

main %>% group_by(Birth_location_zh, Birth_Location_eng_standard) %>% 
  count(sort = TRUE) %>% 
  mutate(percent = round(n/401*100, 2))

main %>% 
  drop_na(Birth_location_zh) %>% 
  group_by(Birth_location_zh) %>%
  count() %>%
  filter(n>2) %>%
  ggplot(aes(reorder(Birth_location_zh, n), n)) +
  geom_col() +
  coord_flip() + 
  labs(title = "Early Liumei's Place of Birth",
       subtitle = "Places > 2 births",
       x = "Locality", 
       y = "Number of students",
       caption = "Source: 遊美同學錄 (1917)")

Mapping

The following maps show the distribution of birth places:

The maps were made with QGIS using spatial data from the Modern China Geospatial Database (MCGD).

Funding

How did they fund their studies? Did it depend their geographical and family background?

Sources of Funding

main %>% group_by(FundingMain) %>% 
  count(sort = TRUE) %>% 
  mutate(percent = round(n/401*100, 2)) 

Funding and Birthplace

# Count number of births and funding sources in each province
prov_count <- main %>% group_by(Birth_Province_zh) %>% count(sort = TRUE) %>% mutate(percent = n/401*100)  %>% rename(nprov = n)
funding_prov <- main %>% group_by(FundingMain, Birth_Province_zh) %>% count(sort = TRUE) %>% rename(nfund = n)

# Join funding with province count 
funding_prov <- left_join(funding_prov, prov_count)
funding_prov$percent <- NULL

funding_prov <- funding_prov %>% mutate(percent_prov = round(nfund/nprov*100, 2)) 

# What provinces benefited most from government scholarship? 
funding_prov %>% filter(FundingMain == "Government support") %>% filter(percent_prov >50)

# What provinces relied mostly on private funding? 
funding_prov %>% filter(FundingMain == "Private support") %>% filter(percent_prov >50) %>% arrange(desc(percent_prov))

The map was made with QGIS using spatial data from the Modern China Geospatial Database (MCGD).

Family Background

Kinship ties

For this analysis, we need to incorporate kinship data. The “kinship” dataset displayed below lists all the relatives mentioned in the 401 biographies. The dataset provides information on the names of the relatives and the type of relation with the student.There are 5 types of relations: Father, Uncle, Sibling (elder brother 兄, younger brother 弟), and spouse:

library(readr)
kinship <- read_delim("data/kinship.csv", 
    delim = ";", escape_double = FALSE, trim_ws = TRUE)

kinship

The dataset contains 216 rows representing 216 kinship ties, meaning that the same student may have several rows if several relatives are mentioned in his biography. On the opposite, students who did not mention any parents in their biography do not appear in this dataset. 192 biographies mention at least one relative.

# 192 distinct biographies 
kinship %>% distinct(ego) %>% count()

# 118 parents, including 100 fathers (97 unique names), 18 uncles (17 unique)
kinship %>% filter(RelationMain == "Father") %>% distinct(relative) %>% count()

kinship %>% filter(RelationMain == "Uncle") %>% distinct(relative) %>% count()

# 81 distinct siblings
kinship %>% filter(RelationMain == "Sibling") %>% distinct(relative) %>% count()

kinship %>% filter(RelationMain == "Sibling") %>% 
  distinct(relative, Relation2) %>% 
  group_by(Relation2) %>% count() # 55 兄, 26 弟

# 16 spouses (8 couples)
kinship %>% filter(RelationMain == "Spouse") %>% distinct(relative) %>% count()

The 216 ties refer to 118 unique parents, including 97 fathers (100 father-son relations) and 17 uncles. Additionally, there are 80 siblings, including 55 elder brothers, and 26 younger brothers (pairs). The population also contains 16 spouses representing 8 couples.

Among the relatives mentioned in the students’ biography, fathers and uncles are the most relevant from the analysis of social-economic backgrounds. In the following steps, therefore, we will focus on fathers and uncles.

Parents’ Occupations

We load the pre-formated dataset that incorporates information on both funding and family background:

library(readr)
kin_funding <- read_delim("data/kin_funding.csv", 
    delim = ";", escape_double = FALSE, trim_ws = TRUE)

kin_funding

The dataset contains the following information:

• Unique identifier of the biography (DocId)
  
• name of the student (ego)
• source of funding (FundingMain, Funding, FundingZh)
• on the relative(s) mentioned in the student’s biography.: name (relative), type of relation (RelationMain, Relation2), and occupation (occup_cat, occupation1, occupation2)

The dataset includes 118 rows (ties) and 11 variables. One row represents one kinship tie, meaning that the same student may have several rows if several relatives are mentioned in his biography. For example, Li Fengzao 李奉藻 (imh-11-61) mentioned not only his father Li Zanchen (李贊辰) but also his uncle Li Enfu (李恩富). On the opposite, students who did not mention any parents in their biography do not appear in this dataset.

First, it is sound to explore the relatives and their occupation before analyzing the correlation with students source of funding:

# relations
kin_funding %>% distinct(ego, relative, RelationMain) %>%
  group_by(RelationMain) %>% 
  count(sort = TRUE) %>% mutate(percent = n / 113*100)

# occupations
kin_funding %>% distinct(relative, occup_cat) %>% 
  group_by(occup_cat) %>% count(sort = TRUE) %>% mutate(percent = n / 113*100)

There are 100 father-son and 18 uncle-nephew relationships in the dataset. Among the 113 unique relatives mentioned, 23% are engaged in business, over 13% are officials, nearly 8% are involved in intellectual activities (such as scholars or teachers), and 5.3% in other professions (such as physicians, religious figures, or rentiers). The occupation is unknown in 50% of cases.

Occupation and Funding

To examine the correlation between family background and source of funding, this study employs correspondence analysis (CA). Correspondence Analysis (CA) is statistical method that examines the relationships between categories within two variables, in this specific case, the source of funding and parents’ occupations. By analyzing the data presented in a contingency table, CA generates insights into the associations between these variables.

# select the two variables

occup_fund <- kin_funding %>% select(occup_cat, FundingMain)  

# create contingency table

occup_fund$Funding <- occup_fund$FundingMain %>% replace_na('Unknown')
occup_fund$occupation_cat <- occup_fund$occup_cat %>% replace_na('Unknown')

occup_fund_table <-
  occup_fund %>% 
  group_by(occup_cat, FundingMain) %>% 
  tally() %>% 
  spread(key = occup_cat, value = n) 

# replace NA with 0
occup_fund_table <- mutate_all(occup_fund_table, ~replace(., is.na(.), 0))

# read first column as row names 

occup_fund_table <- column_to_rownames(occup_fund_table, var = "FundingMain") 

occup_fund_table

The findings are typically visualized in a two-dimensional graphical form known as a biplot, as depicted in the figure below, called biplot. A biplot from correspondence analysis displays both rows (e.g., students’ sources of funding) and columns (e.g., parents’ occupations) as points in a two-dimensional space, where the proximity of points indicates the strength and nature of their relationship. Since correspondence analysis deals with frequency data, there are no negative values, and points close to each other suggest a strong association, while the axes represent the dimensions of variation, helping to interpret the main sources of variation in the data. For this study, we rely on FactoMineR and related packages:

# Correspondance analysis

library(FactoMineR)
library(Factoshiny)
library(factoextra)

res.ca1<-CA(occup_fund_table,graph=FALSE)

plot.CA(res.ca1, title="Source of funding and parents' occupation",
        cex=0.8,cex.main=1.3,cex.axis=1.2,
        caption= "Source: 遊美同學錄, 1917")

The biplot shows the correlation between the source of funding (blue dots) and parents’ occupation (red dots). It captures 90.87% of information. When a particular type of funding is close to a particular occupation, it means that the two are strongly correlated. It is noteworthy that government scholarships mainly benefited students from intellectual backgrounds, likely because they were better prepared to pass the examinations. In contrast, students from business families primarily relied on their own funds, as their economic capital generally spared them the burden of taking competitive exams. Sons of officials relied equally on private and public funding, reflecting the diversity of income situations among late Qing officials, and represented the largest proportion of partial funding.

Education

The educational trajectories of early liuMei raise the following questions:

1. Which disciplines did they choose to specialize in? What degrees did they obtain? Is there a relationship between the level of qualification and the field of study? Additionally, is there a connection between their field of specialization and their parents’ occupation?
2. What schools did they attend before going to the United States, and which universities did they study at in the United States?
3. Did their trajectories help establish privileged connections between Chinese and American universities?

Academic Curricula

Field of Study

What did they study in the United States? What was their field of specialization? The following dataset contains the list of all disciplines studied by all students, with two levels of granularity: minor (lower level) and major (upper level)

# load the complete list of disciplines 

library(readr)
disciplines_all_clean <- read_csv("data_old/disciplines_all_clean.csv")

# count distribution (major)

disciplines_all_clean %>% distinct(DocId, major) %>% 
  group_by(major) %>% count(sort = TRUE) %>% 
  mutate(percent = round(n/410*100, 2))

# count distribution (minor)

disciplines_all_clean %>% distinct(DocId, major, minor) %>% 
  group_by(major, minor) %>% count(sort = TRUE) %>% 
  mutate(percent = round(n/420*100, 2))

At the broader level, social sciences ranked first (nearly one third of individual curricula), followed by engineering (30%) and the humanities (13%). Other groups represent less than 10% each, by order of importance: biological sciences (9.5%), health sciences (6%), physical sciences (4.6%), and military sciences (2.2%).
At a finer level, however, civil engineering is most chosen discipline (40 curricula, 9.5%), followed by liberal artd (9.3%), economics (nearly 8%), commerce, and law (nearly 6% each). Agriculture, education, and political sciences (20 curricula, 5.2 % each) followed closely. All other disciplines, led by medicine and mining, amounted for less than 5% each.

Fields Over Time

How did educational choices evolve over time?

# join list of disciplines with period of studies 

periods <- main %>% select(DocId, period, period2)
discipline_period <- left_join(disciplines_all_clean, periods)

# Step 1: Count the distribution of disciplines by period
discipline_count <- discipline_period %>%
  group_by(period, major) %>%
  summarise(count = n()) %>%
  ungroup()

# Step 2: Calculate the percentage each discipline represents during each period
distribution_by_period <- discipline_count %>%
  group_by(period) %>%
  mutate(period_total = sum(count),
         period_percentage = (count / period_total) * 100) %>%
  ungroup()

# Step 3: Calculate the percentage each discipline represents within the discipline across all periods
discipline_total <- discipline_count %>%
  group_by(major) %>%
  mutate(discipline_total = sum(count)) %>%
  ungroup()

discipline_distribution <- distribution_by_period %>%
  left_join(discipline_total, by = c("period", "major", "count")) %>%
  mutate(discipline_percentage = (count / discipline_total) * 100)

head(discipline_distribution)

The table shows the importance of each discipline within each period, as well as the importance of each period within each discipline. Specifically, the “period_percentage” column indicates that the three leading disciplines during the period 1872-1908 (prior to the First Remission of the Boxer Indemnity) were, by order of importance, Social Sciences, Engineering, and the Humanities. During the second period (post-Boxer), Engineering was the most favored choice, before Social Sciences, and Biological Sciences. Within each discipline, the last column (discipline_percentage) reveals that the Humanities and Social Sciences were more popular during the first period, whereas military sciences were more represented after the Boxer remission.

discipline_period %>% 
  drop_na(major) %>% 
  filter(!major == "NA") %>% 
  filter(!period == "NA")%>% 
  group_by(period, major) %>%
  count()  %>%
  ggplot(aes(reorder(major, n), n)) +
  geom_col() +
  coord_flip() + 
  facet_wrap(vars(period), nrow= 2) +
  labs(title = "Early Liumei's Field of Study",
       subtitle = "Field of Study By Period",
       x = "Field of Study (major)", 
       y = "Number of curricula",
       caption = "Source: 遊美同學錄 (1917)")

discipline_period %>% 
  drop_na(minor) %>% 
  filter(!minor == "NA") %>% 
  filter(!period == "NA")%>% 
  group_by(period, minor) %>%
  count()  %>%
  filter(n>2) %>%
  ggplot(aes(reorder(minor, n), n)) +
  geom_col() +
  coord_flip() + 
  facet_wrap(vars(period), scale = "free_x", nrow= 2) +
  theme(axis.text.x = element_text(size = 10))+
  labs(title = "Early Liumei's Field of Study",
       subtitle = "Field of Study By Period",
       x = "Field of Study (minor)" , 
       y = "Number of curricula",
       caption = "Source: 遊美同學錄 (1917)") 

At a lower level, we observe that law was the most chosen discipline during the initial period, followed by civil engineering and liberal arts. After the Remission of the Boxer Indemnity, education and economics became the most preferred disciplines.

Field and Funding

Was the students’ choice of academic discipline affected by their source of funding? As with the previous analysis of funding and parents’ occupation, we will use correspondence analysis to examine the correlation between the source of funding and the field of study.

# join field and funding 

funding <- main %>% select(DocId, FundingMain)
discipline_funding <- left_join(disciplines_all_clean, funding)

# select the two variables

discipline_funding <- discipline_funding %>% select(major, FundingMain)  

# create contingency table

discipline_funding$FundingMain <- discipline_funding$FundingMain %>% replace_na('Unknown')
discipline_funding$major <- discipline_funding$major %>% replace_na('Unknown')

discipline_funding_table <-
  discipline_funding %>% 
  group_by(major, FundingMain) %>% 
  tally() %>% 
  spread(key = major, value = n) 

# replace NA with 0
discipline_funding_table <- mutate_all(discipline_funding_table, ~replace(., is.na(.), 0))

# read first column as row names 

discipline_funding_table <- column_to_rownames(discipline_funding_table, var = "FundingMain") 

# Correspondance analysis

library(FactoMineR)
library(Factoshiny)
library(factoextra)

res.ca2<-CA(discipline_funding_table,graph=FALSE)

plot.CA(res.ca2, title="Source of funding and field of study",
        cex=0.8,cex.main=1.3,cex.axis=1.2,
        caption= "Source: 遊美同學錄, 1917")

The biplot clearly separates self-funded students on the right, who more frequently engaged in the humanities and health sciences, from government-sponsored students on the left, who more often specialized in engineering and biological sciences.

Influence of Parents’ Occupation

Were the students’ choice of discipline influenced by their parents’ occupation? We use two methods to examine the correlation between the parents’ occupation and the field of study: (1) correspondence analysis (2) sankey diagram

# join field and parents 

parent_occup <- kin_funding %>% select(DocId, occup_cat)
parent_field <- left_join(disciplines_all_clean, parent_occup)
parent_field1 <- parent_field %>% select(occup_cat, major) 
parent_field2 <- parent_field %>% select(occup_cat, minor) 

Correspondence Analysis on Major

parent_field_ca <- parent_field1 %>% select(major, occup_cat) %>% drop_na(occup_cat, major)

# create contingency table

parent_field_table <-
  parent_field_ca %>% 
  group_by(occup_cat, major) %>% 
  tally() %>% 
  spread(key = major, value = n) 


# read first column as row names 

parent_field_table <- column_to_rownames(parent_field_table, var = "occup_cat") 

# replace NA with 0
parent_field_table <- mutate_all(parent_field_table, ~replace(., is.na(.), 0))


res.ca3<-CA(parent_field_table,graph=FALSE)

plot.CA(res.ca3, title="Parents' Occupation and Field of Study",
        cex=0.8,cex.main=1.3,cex.axis=1.2,
        caption= "Source: 遊美同學錄, 1917")

Sankey Chart on Minor

parent_field2 <- parent_field %>% select(occup_cat, minor) %>% drop_na(occup_cat)

parent_field2 <- parent_field2 %>% 
  select(occup_cat, minor) %>% 
  rename(source = occup_cat, target = minor)

# remove NA and compute weight 
link3 <- parent_field2 %>%
  group_by(source, target) %>% 
  count() %>% 
  rename(value = "n") %>% filter(value >1)


# Create node list from edge list ; the node list contains every entity involved in the flow
node3 <- data.frame(
  name=c(as.character(link3$source), 
         as.character(link3$target)) %>% unique()
)

# create unique id for each connection
link3$IDsource <- match(link3$source, node3$name)-1 
link3$IDtarget <- match(link3$target, node3$name)-1


# Make the Network

library(networkD3)
p1 <- sankeyNetwork(Links = link3, Nodes = node3,
                    Source = "IDsource", Target = "IDtarget",
                    Value = "value", NodeID = "name", 
                    fontSize = 12, fontFamily = "Arial", 
                    nodeWidth = 30, 
                    sinksRight=FALSE)
p1

Degrees

What degrees did they obtain?

library(readr)
degrees_all <- read_csv("data/degrees_all.csv")

degrees_all %>% distinct(DocId, Highest) %>% 
  group_by(Highest) %>% count(sort = TRUE)

Did their level of qualification depend on the field of study?

# join degrees with fields

degree_field <- left_join(disciplines_all_clean, degrees_all)

# select variables 

degree_field <- degree_field %>% select(major, Highest) %>% drop_na(Highest, major)

# create contingency table

degree_field_table <-
  degree_field %>% 
  group_by(Highest, major) %>% 
  tally() %>% 
  spread(key = major, value = n) 


# read first column as row names 

degree_field_table <- column_to_rownames(degree_field_table, var = "Highest") 

# replace NA with 0
degree_field_table <- mutate_all(degree_field_table, ~replace(., is.na(.), 0))


res.ca6<-CA(degree_field_table,graph=FALSE)

plot.CA(res.ca6, title="Field of Study and Highest Degree",
        cex=0.8,cex.main=1.3,cex.axis=1.2,
        caption= "Source: 遊美同學錄, 1917")

Universities

Universities in China

# load data on pre-us education 

prepa <- read_delim("data/pre_us_edu.csv", 
                         delim = ";", escape_double = FALSE, trim_ws = TRUE)

prepa %>% distinct(DocId, Institution) %>% group_by(Institution) %>% count(sort = TRUE)

American Universities

# load data on us education 

edu_us <- read_csv("data/edu_us.csv", col_types = cols(...1 = col_skip()))

edu_us %>% rename(Institution = Organization) %>% distinct(DocId, Institution) %>% 
  group_by(Institution) %>% count(sort = TRUE)

Transpacific Connections

# filter relevant data 

prepa <- read_delim("data/pre_us_edu.csv", 
                         delim = ";", escape_double = FALSE, trim_ws = TRUE)

prepa <- prepa %>% distinct(DocId, Institution) %>% rename(source = Institution)

# select relevant data

edu_us_tojoin <- edu_us %>% distinct(DocId, Organization) %>% rename(target = Organization)

#  join chinese and us universities

flow <- left_join(prepa, edu_us_tojoin)

# remove DocId column

flow$DocId <- NULL


# compute weight 

link <- flow %>%
  group_by(source, target) %>% 
  count() %>% 
  rename(value = "n")

# save

write.csv(link, "~/youmei-new/data/flow_china_us.csv")

link2 <- link %>% filter(value >2)



# Create node list from edge list ; the node list contains every entity involved in the flow
node <- data.frame(
  name=c(as.character(link2$source), 
         as.character(link2$target)) %>% unique()
)

# create unique id for each connection
link2$IDsource <- match(link2$source, node$name)-1 
link2$IDtarget <- match(link2$target, node$name)-1


# Make the Networks
p2 <- sankeyNetwork(Links = link2, Nodes = node,
                    Source = "IDsource", Target = "IDtarget",
                    Value = "value", NodeID = "name", fontSize = 12, fontFamily = "Arial", nodeWidth = 30,
                    sinksRight=FALSE)
p2

Employment

In which sector were they employed after their return? Did their employment align with their training?

Sector

# load career data

library(readr)
career_post <- read_csv("data/career_post.csv")

career_post %>% group_by(DocId) %>% count(sort = TRUE)

The “career” dataset contains 1096 unique positions held by 358 returnees, ranging from 1 to 17 positions each. The dataset provides information on nature and level of position (occupation, level, position), the timing (year of position taking, life phase and sequence order in the person’s career), the employing institution and the sector of employment (category, code), the geographical location of employment (city, province, country, area).

career_post %>% distinct(DocId, category_final) %>% 
  group_by(category_final) %>% count(sort = TRUE) %>% mutate(percent = round(n/706*100, 2))

Employers

career_post %>% distinct(DocId, Institution) %>% 
  group_by(Institution) %>% count(sort = TRUE) %>% mutate(percent = round(n/943*100, 2))

Matching Training and Employment

# Atomize field of study

library(tidyr)

field_all <- main %>% select(DocId, field_group) %>% separate_rows(field_group, sep = "-")

# recode fields 

field_all <- field_all %>% mutate(field_main = fct_collapse(field_group, 
                                                          "Engineering Sciences" = c("Engineering", "Civil Engineering"), 
                                                          "Physical Sciences" = "Physical", 
                                                          "Biological Sciences" = "Biological", 
                                                          "Earth Sciences" = "Earth", 
                                                          "Health Sciences" = "Health", 
                                                          "Social Sciences" = "Economics", 
                                                          "Military Sciences" = "Military"))


# join with funding 

funding <- main %>% select(DocId, FundingMain)

field_all <- left_join(field_all, funding)

field_all <- field_all %>% distinct(DocId, FundingMain, field_main)

# join with sector of employment

sector_all <- career_post %>% distinct(DocId, category_final)

field_sector_all <- left_join(field_all, sector_all)

field_sector_all <- field_sector_all %>% rename(training = field_main, job = category_final)

Like for analyzing the correlation between parents’ occupation and source of funding, we will employ Correspondence Analysis to examine the correlation between training and employment:

field_sector_CA <- field_sector_all %>% select(training, job)
field_sector_CA <- field_sector_CA %>% mutate_all(as.character)

# create contingency table

field_sector_CA$training <- field_sector_CA$training %>% replace_na('Unknown')
field_sector_CA$job <- field_sector_CA$job %>% replace_na('Unknown')

field_sector_table <-
  field_sector_CA %>% 
  group_by(job, training) %>% 
  tally() %>% 
  spread(key = training, value = n) 

# replace NA with 0
field_sector_table <- mutate_all(field_sector_table, ~replace(., is.na(.), 0))

# read first column as row names 

field_sector_table <- column_to_rownames(field_sector_table, var = "job") 

# Correspondance analysis

library(FactoMineR)
library(Factoshiny)

res.ca <- CA(field_sector_table, graph = FALSE)

plot.CA(res.ca, title="Training and Employment" , 
        cex=0.9,cex.main=1.3,cex.axis=1.2, 
        caption= "Source: 遊美同學錄, 1917")

Alternatively, we can visualize the relation as a Sankey chart:

field_sector_sankey <- field_sector_CA %>% 
  rename(source = training, target = job)

# remove NA and compute weight 
link3 <- field_sector_sankey %>%
  group_by(source, target) %>% 
  count() %>% 
  rename(value = "n") %>% filter(value >1)


# Create node list from edge list ; the node list contains every entity involved in the flow
node3 <- data.frame(
  name=c(as.character(link3$source), 
         as.character(link3$target)) %>% unique()
)

# create unique id for each connection
link3$IDsource <- match(link3$source, node3$name)-1 
link3$IDtarget <- match(link3$target, node3$name)-1


# Make the Networks
p3 <- sankeyNetwork(Links = link3, Nodes = node3,
                    Source = "IDsource", Target = "IDtarget",
                    Value = "value", NodeID = "name", fontSize = 12, 
                    sinksRight=FALSE)
p3

Level of Qualification by Sector

Did their level of qualification vary depending on the sector of employment?

# join degrees with sectors

degree_sector <- left_join(sector_all, degrees_all)

# select variables 

degree_sector <- degree_sector %>% distinct(Highest, category_final) %>% drop_na(category_final, Highest) %>% filter(!Highest =="Juren") %>% filter(!Highest =="Jinshi")

# create contingency table

degree_sector_table <-
  degree_sector %>% 
  group_by(Highest, category_final) %>% 
  tally() %>% 
  spread(key = Highest, value = n) 


# read first column as row names 

degree_sector_table <- column_to_rownames(degree_sector_table, var = "category_final") 

# replace NA with 0
degree_sector_table <- mutate_all(degree_sector_table, ~replace(., is.na(.), 0))


res.ca7<-CA(degree_sector_table,graph=FALSE)

plot.CA(res.ca7, title="Level of Qualification and Sector of Employment",
        cex=0.8,cex.main=1.3,cex.axis=1.2,
        caption= "Source: 遊美同學錄, 1917")

Funding and Employment

Did their source of funding influence their sector of employment?

# join funding with sector 

funding_sector <- left_join(funding, sector_all)

funding_sector_ca <- funding_sector %>% select(FundingMain, category_final) %>% drop_na(FundingMain, category_final)

# create contingency table

funding_sector_table <-
  funding_sector_ca %>% 
  group_by(category_final, FundingMain) %>% 
  tally() %>% 
  spread(key = FundingMain, value = n) 


# read first column as row names 

funding_sector_table <- column_to_rownames(funding_sector_table, var = "category_final") 

# replace NA with 0
funding_sector_table <- mutate_all(funding_sector_table, ~replace(., is.na(.), 0))


res.ca4<-CA(funding_sector_table,graph=FALSE)

plot.CA(res.ca4, title="Source of Funding and Sector of Employment",
        cex=0.8,cex.main=1.3,cex.axis=1.2,
        caption= "Source: 遊美同學錄, 1917")

Parents’ Occupation and Students’ Employment

How was their parents’ occupation related to their sector of employment?

# join parents occupation with students' sector of employment 

parent_sector <- left_join(parent_occup, sector_all)
parent_sector <- parent_sector %>% select(occup_cat, category_final) %>% drop_na(occup_cat, category_final)

# create contingency table

parent_sector_table <-
  parent_sector %>% 
  group_by(occup_cat, category_final) %>% 
  tally() %>% 
  spread(key = category_final, value = n) 


# read first column as row names 

parent_sector_table <- column_to_rownames(parent_sector_table, var = "occup_cat") 

# replace NA with 0
parent_sector_table <- mutate_all(parent_sector_table, ~replace(., is.na(.), 0))


res.ca5<-CA(parent_sector_table,graph=FALSE)

plot.CA(res.ca5, title="Parents' Occupation and Sector of Employment",
        cex=0.8,cex.main=1.3,cex.axis=1.2,
        caption= "Source: 遊美同學錄, 1917")

Alternatively, we can visualize the relation as a Sankey chart:

parent_sector_sankey <- parent_sector %>% 
  rename(source = occup_cat, target = category_final)

# remove NA and compute weight 
link4 <- parent_sector_sankey %>%
  group_by(source, target) %>% 
  count() %>% 
  rename(value = "n") %>% filter(value >1)


# Create node list from edge list ; the node list contains every entity involved in the flow
node4 <- data.frame(
  name=c(as.character(link4$source), 
         as.character(link4$target)) %>% unique()
)

# create unique id for each connection
link4$IDsource <- match(link4$source, node4$name)-1 
link4$IDtarget <- match(link4$target, node4$name)-1


# Make the Networks
p4 <- sankeyNetwork(Links = link4, Nodes = node4,
                    Source = "IDsource", Target = "IDtarget",
                    Value = "value", NodeID = "name", fontSize = 12, 
                    sinksRight=FALSE)
p4

Workplace

Where did they work after graduation?

Province

# load workplace data 

library(readr)
workplace <- read_delim("data/workplace.csv", 
    delim = ";", escape_double = FALSE, trim_ws = TRUE)

head(workplace)

workplace %>% group_by(province_py_std, work_province) %>% 
  count(sort = TRUE) %>% 
  mutate(percent = round(n/401*100, 2))

workplace %>% 
  drop_na(work_province) %>% 
  group_by(work_province) %>%
  count()  %>%
  filter(n>2) %>%
  ggplot(aes(reorder(work_province, n), n)) +
  geom_col() +
  coord_flip() + 
  labs(title = "Early Liumei's Workplace in 1917",
       subtitle = "Provinces employing more than 2 returnees",
       x = "Province", 
       y = "Number of returnees",
       caption = "Source: 遊美同學錄 (1917)")

Locality

workplace %>% group_by(city_py, work_city) %>% 
  count(sort = TRUE) %>% 
  mutate(percent = round(n/401*100, 2))

workplace %>% 
  drop_na(work_city) %>% 
  group_by(work_city) %>%
  count()  %>%
  filter(n>2) %>%
  ggplot(aes(reorder(work_city, n), n)) +
  geom_col() +
  coord_flip() + 
  labs(title = "Early Liumei's Workplace in 1917",
       subtitle = "Cities employing more than 2 returnees",
       x = "City", 
       y = "Number of returnees",
       caption = "Source: 遊美同學錄 (1917)")

Mapping

The following maps show the distribution of workplaces in 1917. Like the maps above, these were created using QGIS and MCGD data.

The comparative mapping of birthplace and workplace among the early cohorts of liuMei reveals a notable brain gain for the late Qing and Beiyang governments. This analysis highlights a substantial transfer of human and intellectual resources from southeastern to northern China, as well as a shift from private to public sectors. The disproportionate concentration of returned students in the Jiangsu area, particularly in Nanjing and Shanghai, did not manifest until the establishment of the Nationalist government in Nanjing in 1927.

Mobility

Did they return to work in their native place?

# join birthplace and workplace

sankey_birth <- main %>% select(DocId, Birth_Province_zh)
sankey_work <- workplace %>% select(DocId, work_province)
sankey_birth_work <- inner_join(sankey_birth, sankey_work, by = "DocId")

sankey_birth_work <- sankey_birth_work %>% 
  select(Birth_Province_zh, work_province) %>% 
  rename(source = Birth_Province_zh, target = work_province)

head(sankey_birth_work)

# remove NA and compute weight 
link_bw <- sankey_birth_work %>%
  group_by(source, target) %>% 
  count() %>% 
  rename(value = "n")


link_bw_count <- link_bw %>% group_by(source) %>% mutate(total_birth = sum(value))
link_bw_count <- link_bw_count %>% mutate(percent= value/total_birth*100)

# find identical places 

link_bw_count <- link_bw_count %>%
  mutate(Identical = ifelse(source == target, TRUE, FALSE))


head(link_bw_count)

# filter most important flows 

link5 <- link_bw %>% filter(value>1)

# visualize flows as a Sankey chart as a Sankey chart 

# Create node list from edge list ; the node list contains every entity involved in the flow
node5 <- data.frame(
  name=c(as.character(link5$source), 
         as.character(link5$target)) %>% unique()
)

# create unique id for each connection
link5$IDsource <- match(link5$source, node5$name)-1 
link5$IDtarget <- match(link5$target, node5$name)-1

# Create the Network

library(networkD3)
p5 <- sankeyNetwork(Links = link5, Nodes = node5,
                   Source = "IDsource", Target = "IDtarget",
                   Value = "value", NodeID = "name", 
                   fontSize = 14, fontFamily = "Arial", nodeWidth = 30, 
                   sinksRight=FALSE)
p5

# save the widget
library(htmlwidgets)
# saveWidget(p5, file=paste0( getwd(), "/birthwork_prov.html"))

The self-loops indicate that a significant number of early liuMei returned to work in their native provinces, particularly those from Hebei (17, 74%), Jiangsu (77, 56%), and Guangdong (37, 36%). However, there was also significant brain drain in favor of Hebei from certain provinces, especially Zhejiang (59%), Guangdong (27.5%), and peripheral provinces like Gansu. Conversely, while 22% of Jiangsu natives were working in Hebei in 1917, 26% of Hebei natives were employed in Jiangsu. Additionally, China benefited from American-born Chinese who exemplified the “American exodus” studied by Charlotte Brooks (Brooks, 2019), with 43% (6) employed in Hebei, 21% (3) in Guangdong, and 14% (2) in Jiangsu.