#### libraries -------------- library(ggplot2) library(cowplot) library(RColorBrewer) library(reshape2) library(forcats) library(knitr) library(kableExtra) library(rlang) library(wesanderson) library(haven) library(margins)# library(broom) library(stargazer) library(scales) library(dplyr) library(tidyr) #### 4 Estimations #### library(MASS) #MASS:: select clashes with dplyr::select, load late library(mfx) library(lmtest) library(sandwich) library(fastDummies) library(data.table) library(tibble) library(kableExtra) est<-readRDS("est.RDS") ## remove reasons est<-est %>% select(-reason_stud,-reason_comp,-reason_trade,-reason_cust,-reason_inc,- reason_requ,-reason_lab,- reason_partner,-reason_soc,-reason_move_prof,-reason_move_other,-reason_holi,-reason_cult,-reason_inter) table(a$country, a$age) #1. type logit (max and min sample x stepwise introduction X aggregate) #2. type logit (max and min sample x stepwise introduction X country_wise) #1. langjob logit (max and min sample x stepwise introduction X aggregate) #2. langjob logit (max and min sample x stepwise introduction X country_wise) # Info on ind. var.: #1. empstat: ref. cat. empstat=2 table(est$empstat) est <- within(est, empstat <- relevel(factor(empstat), ref = "2")) #2. part (ref cat part=0) est <- within(est, part <- relevel(factor(part), ref = "0")) # part=0 No partner # part=1 partner with German native language # part=2 partner with other native language #3. Degree (ref cat part=2) est <- within(est, degree <- relevel(factor(degree), ref = "2")) # 0 No degree # 1 School leaving certificate which cannot lead to higher education # 2 School leaving certificate which can lead to higher education # 3 University degree # 4 PhD #4. female: ref. cat. female=0 table(est$female) est <- within(est, female <- relevel(factor(female), ref = "0")) stargazer(est, type="text") #Aggregate 0+1+5 and 3+4 est<-est %>% mutate(degree2=case_when(is.na(.$degree) ~ NA, .$degree =="0" ~ FALSE, .$degree =="1" ~ FALSE, .$degree =="2" ~ FALSE, .$degree =="3" ~ TRUE, .$degree =="4" ~ TRUE, .$degree =="5" ~ FALSE), degree=degree2) table(est$degree, est$degree2, useNA = "always") table(est$mreason_lang, useNA="always") #use only those with completed cases table(est$complete) # table(est$type, est$type2, useNA = "always") est$type2[is.na(est$type2)] = "No answer" est<-est %>% mutate(complete=ifelse(complete.cases(est), TRUE, FALSE)) gi<-est[est$complete=="TRUE",] table(gi$type, gi$type2, useNA = "always") table(gi$country, gi$empstat) table(gi$reason_om) #### Sample shrinkage ------------------- indv <- readRDS("indv.RDS") indv <-indv %>% mutate(ling=case_when(.$country=="BIH" ~ 0, .$country=="CZE" ~ 0, .$country=="ESP" ~ 0, .$country=="GBR" ~ 1, .$country=="IDN"~ 0, .$country=="IND"~ 1, .$country=="ITA"~ 0, .$country=="JPN" ~ 0, .$country=="KOR" ~ 0, .$country=="MEX"~ 0, .$country=="NLD"~ 1, .$country=="POL"~ 0, .$country=="ROU"~0, .$country=="UKR"~ 0), europe=case_when(.$country=="BIH" ~ 1, .$country=="CZE" ~ 0, .$country=="ESP" ~ 0, .$country=="GBR" ~ 0, .$country=="IDN"~ 2, .$country=="IND"~ 2, .$country=="ITA"~ 0, .$country=="JPN" ~ 2, .$country=="KOR" ~ 2, .$country=="MEX"~ 2, .$country=="NLD"~ 0, .$country=="POL"~ 0, .$country=="ROU"~0, .$country=="UKR"~ 1), income=case_when(.$country=="BIH" ~ 1, .$country=="CZE" ~ 2, .$country=="ESP" ~ 2, .$country=="GBR" ~ 2, .$country=="IDN"~ 0, .$country=="IND"~ 0, .$country=="ITA"~ 2, .$country=="JPN" ~ 2, .$country=="KOR" ~ 2, .$country=="MEX"~ 1, .$country=="NLD"~ 2, .$country=="POL"~ 2, .$country=="ROU"~1, .$country=="UKR"~ 0)) indv1 <- subset(indv, !is.na(age)) indv2 <- subset(indv1, !is.na(female)) indv3 <- subset(indv2, !is.na(children_d_na)) indv4 <- subset(indv3, !is.na(part)) indv5 <- subset(indv4, !is.na(part)) indv6 <- subset(indv5, !is.na(degree)) indv7 <- subset(indv6, !is.na(risk)) indv8 <- subset(indv7, !is.na(patient)) indv9 <- subset(indv8, !is.na(english)) indv10 <- subset(indv9, !is.na(occupationac)) #### Exclude Education from Main Reason (Uncomment when needed) ----------------- #gi <- as.data.frame(subset(gi, gi$mreason_lang!="Education")) ##### Descriptive table (Table A1: means of individual characteristics by countries) ###### d<-gi d <- dummy_cols(d, select_columns =c("female", "part", "occupationac")) d<-d %>% dplyr::select(country, age, female_0, female_1, female_2, children_d_na, part_0, part_1, part_2, occupationac_0, occupationac_1, occupationac_2, occupationac_3, occupationac_4, degree, risk, patient, english ) %>% mutate(one=1) mean<-aggregate(d[,2:19], list(d$country), mean) x<-colMeans(d[,2:19]) x<-c("Total", x) mean<-rbind(mean, x) mean[,2:19] <- sapply(mean[,2:19], function(x) round(as.numeric(x), digits=2)) mean<- mean %>% rename(country=Group.1) sum<-aggregate(d[,19], list(d$country), sum) x<-sum(d$one) x<-c("Total", x) sum<-rbind(sum, x) sum sum<- sum %>% rename(country=Group.1, obs=x) b<-left_join(mean, sum) b$obs<-as.numeric((b$obs)) b<-b %>% mutate(country=paste0(country, "\n", "n=(", obs, ")")) %>% dplyr::select(-obs) row.names(b)<-b$country # transpose t_b <- transpose(b) # get row and colnames in order colnames(t_b) <- rownames(b) rownames(t_b) <- colnames(b) t_b<-rownames_to_column(t_b, var="Variable") t_b<-filter(t_b, Variable!="country") t_b$Variable<-as.factor(t_b$Variable) t_b$Variable<- fct_recode(t_b$Variable, "Age: under 35 years" = "age" , "Gender: male" = "female_0" , "Gender: female" = "female_1", "Gender: n/a" = "female_2", "No partner" = "part_0" , "Partner (native German)" = "part_1" , "Partner (other native)" = "part_2" , "Children" = "children_d_na" , "University degree" = "degree" , "Occ.: low appl."="occupationac_0", "Occ.: high appl./low comm. skills"="occupationac_1", "Occ.: high appl./high comm. skills"="occupationac_2", "Occ.: in education"="occupationac_3", "Occ.: other occ./no answer"="occupationac_4", "Risk attitude"="risk", "Patience"="patient", "English speaker" = "english", "Intern. applic. of education"="applic_gen_d", "Comm. skills: low" = "comm_skills_0", "Comm. skills: high" = "comm_skills_1", "Comm. skills: n/a" = "comm_skills_2" ) t_b %>% kable(format = "latex", digits=2, booktabs=TRUE, linesep = "\\addlinespace", caption="Descriptive statistics: means of individual characteristics by countries", align = c("l", "c", "c","c","c","c","c","c","c","c","c","c","c","c","c","c")) %>% kable_styling(latex_options = c("hold_position")) %>% landscape() %>% column_spec(1, width = "4.7cm") %>% column_spec(2:16, width= "1cm") ##### Descriptive table (Table 2: means of individual characteristics by country-groups - w/ main reasons) ###### d<-gi d <- dummy_cols(d, select_columns =c("female", "part", "occupationac")) d<-d %>% dplyr::select(country, age, female_0, female_1, female_2, children_d_na, part_0, part_1, part_2, occupationac_0, occupationac_1, occupationac_2, occupationac_3, occupationac_4, degree, risk, patient, english ) %>% mutate(one=1) d<-d %>% mutate(countrygroup=ifelse(country=="ESP" | country=="ITA" | country=="GBR" | country=="NLD" | country=="CZE" | country=="POL" | country=="ROU", "EU", ifelse(country=="UKR" | country=="BIH", "European (non-EU)", "Non-European" ))) mean<-aggregate(d[,2:19], list(d$countrygroup), mean) x<-colMeans(d[,2:19]) x<-c("Total", x) mean<-rbind(mean, x) mean[,2:19] <- sapply(mean[,2:19], function(x) round(as.numeric(x), digits=2)) mean<- mean %>% rename(countrygroup=Group.1) sum<-aggregate(d[,19], list(d$countrygroup), sum) x<-sum(d$one) x<-c("Total", x) sum<-rbind(sum, x) sum sum<- sum %>% rename(countrygroup=Group.1, obs=x) b<-left_join(mean, sum) b$obs<-as.numeric((b$obs)) b<-b %>% mutate(countrygroup=paste0(countrygroup, "\n", "n=(", obs, ")")) %>% dplyr::select(-obs) row.names(b)<-b$countrygroup # transpose t_b <- transpose(b) # get row and colnames in order colnames(t_b) <- rownames(b) rownames(t_b) <- colnames(b) t_b<-rownames_to_column(t_b, var="Variable") t_b<-filter(t_b, Variable!="countrygroup") t_b$Variable<-as.factor(t_b$Variable) t_b$Variable<- fct_recode(t_b$Variable, "Age: under 35 years" = "age" , "Gender: male" = "female_0" , "Gender: female" = "female_1" , "Gender: n/a" = "female_2", "No partner" = "part_0" , "Partner (native German)" = "part_1" , "Partner (other native)" = "part_2" , "Children" = "children_d_na" , "University degree" = "degree" , "Occ.: low appl."="occupationac_0", "Occ.: high appl./low comm. skills"="occupationac_1", "Occ.: high appl./high comm. skills"="occupationac_2", "Occ.: in education"="occupationac_3", "Occ.: other occ./no answer"="occupationac_4", "Risk attitude"="risk", "Patience"="patient", "English speaker" = "english", "Intern. applic. of education"="applic_gen_d", "Comm. skills: low" = "comm_skills_0", "Comm. skills: high" = "comm_skills_1", "Comm. skills: n/a" = "comm_skills_2") t_b %>% kable(format = "latex", digits=2, booktabs=TRUE, linesep = "\\addlinespace", caption="Descriptive statistics: means of individual characteristics by country groups", align = c("l", "c", "c","c","c")) %>% kable_styling(latex_options = c("hold_position")) %>% landscape() %>% column_spec(1, width = "4.7cm") %>% column_spec(2:16, width= "2cm") ##### Descriptive table (Table 2: means of individual characteristics by country-groups - all reasons) ###### d<-est d <- dummy_cols(d, select_columns =c("female", "part", "occupationac")) d<-d %>% dplyr::select(country, age, female_0, female_1, female_2, children_d_na, part_0, part_1, part_2, occupationac_0, occupationac_1, occupationac_2, occupationac_3, occupationac_4, degree,risk, patient, english ) %>% mutate(one=1) d<-d %>% mutate(countrygroup=ifelse(country=="ESP" | country=="ITA" | country=="GBR" | country=="NLD" | country=="CZE" | country=="POL" | country=="ROU", "EU", ifelse(country=="UKR" | country=="BIH", "European (non-EU)", "Non-European" ))) mean<-aggregate(d[,2:17], list(d$countrygroup), mean) x<-colMeans(d[,2:17]) x<-c("Total", x) mean<-rbind(mean, x) mean[,2:17] <- sapply(mean[,2:17], function(x) round(as.numeric(x), digits=2)) mean<- mean %>% rename(countrygroup=Group.1) sum<-aggregate(d[,18], list(d$countrygroup), sum) x<-sum(d$one) x<-c("Total", x) sum<-rbind(sum, x) sum sum<- sum %>% rename(countrygroup=Group.1, obs=x) b<-left_join(mean, sum) b$obs<-as.numeric((b$obs)) b<-b %>% mutate(countrygroup=paste0(countrygroup, "\n", "n=(", obs, ")")) %>% dplyr::select(-obs) row.names(b)<-b$countrygroup # transpose t_b <- transpose(b) # get row and colnames in order colnames(t_b) <- rownames(b) rownames(t_b) <- colnames(b) t_b<-rownames_to_column(t_b, var="Variable") t_b<-filter(t_b, Variable!="countrygroup") t_b$Variable<-as.factor(t_b$Variable) t_b$Variable<- fct_recode(t_b$Variable, "Age: under 35 years" = "age" , "Gender: male" = "female_0" , "Gender: female" = "female_1" , "Gender: n/a" = "female_2", "No partner" = "part_0" , "Partner (native German)" = "part_1" , "Partner (other native)" = "part_2" , "Children" = "children_d_na" , "University degree" = "degree" , "Occ.: in education"="empstat_1", "Occ.: in lab. market"="empstat_2", "Other occ./no answer"="empstat_3", "Risk attitude"="risk", "Patience"="patient", "Intern. applic. of education"="applic_gen") t_b %>% kable(format = "latex", digits=2, booktabs=TRUE, linesep = "\\addlinespace", caption="Descriptive statistics: means of individual characteristics by country groups", align = c("l", "c", "c","c","c")) %>% kable_styling(latex_options = c("hold_position")) %>% landscape() %>% column_spec(1, width = "4.7cm") %>% column_spec(2:16, width= "2cm") ##### Descriptive table (all excluded from 6664) ------------ cleared <- anti_join(indv, gi, by="id_uni") d<-cleared #Aggregate 0+1 and 3+4 d<-d %>% mutate(degree2=case_when(is.na(.$degree) ~ NA, .$degree =="0" ~ FALSE, .$degree =="1" ~ FALSE, .$degree =="2" ~ FALSE, .$degree =="3" ~ TRUE, .$degree =="4" ~ TRUE, .$degree =="5" ~ FALSE), degree=degree2) d <- dummy_cols(d, select_columns =c("female", "part", "occupationac")) d<-d %>% dplyr::select(country, age, female_0, female_1, female_2, children_d_na, part_0, part_1, part_2, occupationac_0, occupationac_1, occupationac_2, occupationac_3, occupationac_4, degree, risk, patient, english ) %>% mutate(one=1) d<-d %>% mutate(countrygroup=ifelse(country=="ESP" | country=="ITA" | country=="GBR" | country=="NLD" | country=="CZE" | country=="POL" | country=="ROU", "EU", ifelse(country=="UKR" | country=="BIH", "European (non-EU)", "Non-European" ))) mean<-aggregate(d[,2:19], na.action = na.omit, list(d$countrygroup), mean, na.rm=TRUE) x<-colMeans(d[,2:19], na.rm=TRUE) x<-c("Total", x) mean<-rbind(mean, x) mean[,2:19] <- sapply(mean[,2:19], function(x) round(as.numeric(x), digits=2)) mean<- mean %>% rename(countrygroup=Group.1) sum<-aggregate(d[,19], list(d$countrygroup), sum) x<-sum(d$one) x<-c("Total", x) sum<-rbind(sum, x) sum sum<- sum %>% rename(countrygroup=Group.1, obs=x) b<-left_join(mean, sum) b$obs<-as.numeric((b$obs)) b<-b %>% mutate(countrygroup=paste0(countrygroup, "\n", "n=(", obs, ")")) %>% dplyr::select(-obs) row.names(b)<-b$countrygroup # transpose t_b <- transpose(b) # get row and colnames in order colnames(t_b) <- rownames(b) rownames(t_b) <- colnames(b) t_b<-rownames_to_column(t_b, var="Variable") t_b<-filter(t_b, Variable!="countrygroup") t_b$Variable<-as.factor(t_b$Variable) t_b$Variable<- fct_recode(t_b$Variable, "Age: under 35 years" = "age" , "Gender: male" = "female_0" , "Gender: female" = "female_1" , "Gender: n/a" = "female_2", "No partner" = "part_0" , "Partner (native German)" = "part_1" , "Partner (other native)" = "part_2" , "Children" = "children_d_na" , "University degree" = "degree" , "Occ.: low appl."="occupationac_0", "Occ.: high appl./low comm. skills"="occupationac_1", "Occ.: high appl./high comm. skills"="occupationac_2", "Occ.: in education"="occupationac_3", "Occ.: other occ./no answer"="occupationac_4", "Risk attitude"="risk", "Patience"="patient", "English speaker" = "english") t_b %>% kable(format = "latex", digits=2, booktabs=TRUE, linesep = "\\addlinespace", caption="Descriptive statistics: means of individual characteristics by country groups", align = c("l", "c", "c","c","c")) %>% kable_styling(latex_options = c("hold_position")) %>% landscape() %>% column_spec(1, width = "4.7cm") %>% column_spec(2:16, width= "2cm") ##### Descriptive table (all 6664) ------------ cleared <- indv d<-cleared #Aggregate 0+1 and 3+4 d<-d %>% mutate(degree2=case_when(is.na(.$degree) ~ NA, .$degree =="0" ~ FALSE, .$degree =="1" ~ FALSE, .$degree =="2" ~ FALSE, .$degree =="3" ~ TRUE, .$degree =="4" ~ TRUE, .$degree =="5" ~ FALSE), degree=degree2) d <- dummy_cols(d, select_columns =c("female", "part", "occupationac")) d<-d %>% dplyr::select(country, age, female_0, female_1, female_2, children_d_na, part_0, part_1, part_2, occupationac_0, occupationac_1, occupationac_2, occupationac_3, occupationac_4, degree, risk, patient, english ) %>% mutate(one=1) d<-d %>% mutate(countrygroup=ifelse(country=="ESP" | country=="ITA" | country=="GBR" | country=="NLD" | country=="CZE" | country=="POL" | country=="ROU", "EU", ifelse(country=="UKR" | country=="BIH", "European (non-EU)", "Non-European" ))) mean<-aggregate(d[,2:19], na.action = na.omit, list(d$countrygroup), mean, na.rm=TRUE) x<-colMeans(d[,2:19], na.rm=TRUE) x<-c("Total", x) mean<-rbind(mean, x) mean[,2:19] <- sapply(mean[,2:19], function(x) round(as.numeric(x), digits=2)) mean<- mean %>% rename(countrygroup=Group.1) sum<-aggregate(d[,19], list(d$countrygroup), sum) x<-sum(d$one) x<-c("Total", x) sum<-rbind(sum, x) sum sum<- sum %>% rename(countrygroup=Group.1, obs=x) b<-left_join(mean, sum) b$obs<-as.numeric((b$obs)) # b<-b %>% mutate(age=paste0(sum_age, " (", round(mean_age, digits = 2) , ")" ), # female=paste0(sum_female, " (", round(mean_female, digits = 2) , ")" ), # part_0=paste0(sum_part_0, " (", round(mean_part_0, digits = 2) , ")" ), # part_1=paste0(sum_part_1, " (", round(mean_part_1, digits = 2) , ")" ), # part_2=paste0(sum_part_2, " (", round(mean_part_2, digits = 2) , ")" ), # children_d_na=paste0(sum_children_d_na, " (", round(mean_children_d_na, digits = 2) , ")" ), # degree=paste0(sum_degree, " (", round(mean_degree, digits = 2) , ")" ), # empstat_1=paste0(sum_empstat_1, " (", round(mean_empstat_1, digits = 2) , ")" ), # empstat_2=paste0(sum_empstat_2, " (", round(mean_empstat_2, digits = 2) , ")" ), # empstat_3=paste0(sum_empstat_3, " (", round(mean_empstat_3, digits = 2) , ")" ), # empstat_4=paste0(sum_empstat_4, " (", round(mean_empstat_4, digits = 2) , ")" ), # risk=paste0(" ", " (", round(mean_risk, digits = 2) , ")" ), # patient=paste0(" ", " (", round(mean_patient, digits = 2) , ")" ), # applic_gen=paste0(" ", " (", round(mean_applic_gen, digits = 2) , ")" ), # mig_phase=paste0(sum_mig_phase, " (", round(mean_mig_phase, digits = 2) , ")" ) # ) %>% # rename(obs=sum_one) %>% # dplyr::select(-starts_with("mean_"), -starts_with("sum_")) %>% b<-b %>% mutate(countrygroup=paste0(countrygroup, "\n", "n=(", obs, ")")) %>% dplyr::select(-obs) row.names(b)<-b$countrygroup # transpose t_b <- transpose(b) # get row and colnames in order colnames(t_b) <- rownames(b) rownames(t_b) <- colnames(b) t_b<-rownames_to_column(t_b, var="Variable") t_b<-filter(t_b, Variable!="countrygroup") t_b$Variable<-as.factor(t_b$Variable) t_b$Variable<- fct_recode(t_b$Variable, "Age: under 35 years" = "age" , "Gender: male" = "female_0" , "Gender: female" = "female_1" , "Gender: n/a" = "female_2", "No partner" = "part_0" , "Partner (native German)" = "part_1" , "Partner (other native)" = "part_2" , "Children" = "children_d_na" , "University degree" = "degree" , "Occ.: low appl."="occupationac_0", "Occ.: high appl./low comm. skills"="occupationac_1", "Occ.: high appl./high comm. skills"="occupationac_2", "Occ.: in education"="occupationac_3", "Occ.: other occ./no answer"="occupationac_4", "Risk attitude"="risk", "Patience"="patient", "English speaker" = "english") t_b %>% kable(format = "latex", digits=2, booktabs=TRUE, linesep = "\\addlinespace", caption="Descriptive statistics: means of individual characteristics by country groups", align = c("l", "c", "c","c","c")) %>% kable_styling(latex_options = c("hold_position")) %>% landscape() %>% column_spec(1, width = "4.7cm") %>% column_spec(2:16, width= "2cm") ##### T-test to compare full and cleared sample -------------------------------- gitest <- gi %>% mutate(female1 = case_when(female == 1 ~ 1, female == 0 ~ 0, female == 2 ~ 0), female0 = case_when(female == 1 ~ 0, female == 0 ~ 1, female == 2 ~ 0), part0 = case_when(part == 0 ~ 1, part == 1 ~ 0, part == 2 ~ 0), part1 = case_when(part == 1 ~ 1, part == 0 ~ 0, part == 2 ~ 0), part2 = case_when(part == 2 ~ 1, part == 0 ~ 0, part == 1 ~ 0), occupationac0 = case_when(occupationac == 0 ~ 1, occupationac == 1 ~ 0, occupationac == 2 ~ 0, occupationac == 3 ~ 0, occupationac == 4 ~ 0), occupationac1 = case_when(occupationac == 0 ~ 0, occupationac == 1 ~ 1, occupationac == 2 ~ 0, occupationac == 3 ~ 0, occupationac == 4 ~ 0), occupationac2 = case_when(occupationac == 0 ~ 0, occupationac == 1 ~ 0, occupationac == 2 ~ 1, occupationac == 3 ~ 0, occupationac == 4 ~ 0), occupationac3 = case_when(occupationac == 0 ~ 0, occupationac == 1 ~ 0, occupationac == 2 ~ 0, occupationac == 3 ~ 1, occupationac == 4 ~ 0), occupationac4 = case_when(occupationac == 0 ~ 0, occupationac == 1 ~ 0, occupationac == 2 ~ 0, occupationac == 3 ~ 0, occupationac == 4 ~ 1), degree1 = case_when(degree2 == 0 ~ 0, degree2 == 1 ~ 1), english1 = case_when(english == 0 ~ 0, english == 1 ~ 1) ) indvtest <- indv %>% mutate(female1 = case_when(female == 1 ~ 1, female == 0 ~ 0, female == 2 ~ 0), female0 = case_when(female == 1 ~ 0, female == 0 ~ 1, female == 2 ~ 0), part0 = case_when(part == 0 ~ 1, part == 1 ~ 0, part == 2 ~ 0), part1 = case_when(part == 1 ~ 1, part == 0 ~ 0, part == 2 ~ 0), part2 = case_when(part == 2 ~ 1, part == 0 ~ 0, part == 1 ~ 0), occupationac0 = case_when(occupationac == 0 ~ 1, occupationac == 1 ~ 0, occupationac == 2 ~ 0, occupationac == 3 ~ 0, occupationac == 4 ~ 0), occupationac1 = case_when(occupationac == 0 ~ 0, occupationac == 1 ~ 1, occupationac == 2 ~ 0, occupationac == 3 ~ 0, occupationac == 4 ~ 0), occupationac2 = case_when(occupationac == 0 ~ 0, occupationac == 1 ~ 0, occupationac == 2 ~ 1, occupationac == 3 ~ 0, occupationac == 4 ~ 0), occupationac3 = case_when(occupationac == 0 ~ 0, occupationac == 1 ~ 0, occupationac == 2 ~ 0, occupationac == 3 ~ 1, occupationac == 4 ~ 0), occupationac4 = case_when(occupationac == 0 ~ 0, occupationac == 1 ~ 0, occupationac == 2 ~ 0, occupationac == 3 ~ 0, occupationac == 4 ~ 1), degree1 = case_when(degree == 0 ~ 0, degree == 1 ~ 0, degree == 2 ~ 0, degree == 3 ~ 1, degree == 4 ~ 1, degree == 5 ~ 0), english1 = case_when(english == 0 ~ 0, english == 1 ~ 1) ) t.test(indvtest$degree1, gitest$degree1, var.equal = FALSE) ##### Analyze missings in mreason_lang ----------------------------------------- mlangmiss <- as.data.frame(subset(gi, gi$mreason_lang=="No answer")) mlangcomp <- as.data.frame(subset(gi, gi$mreason_lang!="No answer")) t.test(mlangcomp$age, mlangmiss$age, var.equal = FALSE) ##### #### Analyze correlation between reason for migration and language learning motive ------ # gimig <- as.data.frame(subset(gi, mreason_mig1 >= 0)) # # table(gimig$mainrea_educlab) # # educlab <- table(gimig$mreason_mig1[gimig$mreason_lang=="Other migration"]) # library(openxlsx) # write.xlsx(educlab, 'tables/2.xlsx') # educlab <- table(gimig$mainrea_other, gimig$mreason_lang) # educlab <- as.data.frame(educlab) # educlab <- reshape(educlab, idvar = "Var1", timevar = "Var2", direction = "wide") # names(educlab) <- sub(".*\\.", "", names(educlab)) # library(openxlsx) # write.xlsx(educlab, 'tables/educ.xlsx') ######## Binary Probit ######################################### #### 1 Binomial (logit) type #### #### Exclude edcuation from motives -------------------------------------------- giwo <- as.data.frame(subset(gi, gi$mreason_lang!="Education")) #### 1 a1 w/o FE #### type_full<-as.formula("type ~ factor(age)+female+children_d_na+part+occupationac+degree+risk+patient+english") #+mig_phase") type<-glm(formula = type_full, family = binomial(link = "probit"), data = gi) type_het <- print(coeftest(type, vcovCL), digits=6) nobs(type) # Marginal effects (Average marginal effects) type_ame <- probitmfx(formula = type_full, data=gi, atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) # average marginal effects with heteroscedasticity-robust White standard errors print(type_ame) # Percent correctly predicted values type_predict<-(table(true = gi$type, pred = round(fitted(type)))[1,1]+table(true = gi$type, pred = round(fitted(type)))[2,2])/nobs(type)*100 # McFadden's Pseudo R-squared type_baseline <- update(type, formula= type ~ 1) type_McFadden <- 1 - as.vector(logLik(type)/logLik(type_baseline)) type_McFadden #### 1 a2 w FE #### table(gi$age) type_full_fe<-as.formula("type ~ factor(age)+female+children_d_na+part+occupationac+degree+risk+patient+english+country")#+mig_phase") type_fe<-glm(formula = type_full_fe, family = binomial(link = "probit"), data = gi) type_fe_het <- print(coeftest(type_fe, vcovCL), digits=6) nobs(type_fe) # Marginal effects (Average marginal effects) type_fe_ame <- probitmfx(formula = type_full_fe, data=gi, atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) # average marginal effects with heteroscedasticity-robust White standard errors print(type_fe_ame) # Percent correctly predicted values type_fe_predict<-(table(true = gi$type, pred = round(fitted(type_fe)))[1,1]+table(true = gi$type, pred = round(fitted(type_fe)))[2,2])/nobs(type_fe)*100 # McFadden's Pseudo R-squared type_fe_baseline <- update(type_fe, formula= type ~ 1) type_fe_McFadden <- 1 - as.vector(logLik(type_fe)/logLik(type_fe_baseline)) type_fe_McFadden #### 1 a2 w FE w/o education #### type_full_fe<-as.formula("type ~ factor(age)+female+children_d_na+part+occupationac+degree+risk+patient+english+country")#+mig_phase") type_fe_wo<-glm(formula = type_full_fe, family = binomial(link = "probit"), data = giwo) type_fe_het_wo <- print(coeftest(type_fe_wo, vcovCL), digits=6) nobs(type_fe_wo) # Marginal effects (Average marginal effects) type_fe_ame_wo <- probitmfx(formula = type_full_fe, data=giwo, atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) # average marginal effects with heteroscedasticity-robust White standard errors print(type_fe_ame_wo) # Percent correctly predicted values type_fe_predict_wo <- (table(true = giwo$type, pred = round(fitted(type_fe_wo)))[1,1]+table(true = giwo$type, pred = round(fitted(type_fe_wo)))[2,2])/nobs(type_fe_wo)*100 # McFadden's Pseudo R-squared type_fe_baseline_wo <- update(type_fe_wo, formula= type ~ 1) type_fe_McFadden_wo <- 1 - as.vector(logLik(type_fe_wo)/logLik(type_fe_baseline_wo)) type_fe_McFadden_wo #### 1 a3 w/o FE + c#### type_full_c<-as.formula("type ~ factor(age)+female+children_d_na+part+occupationac+degree+risk+patient+english+factor(ling)+factor(europe)+factor(income)+ltodist+ivrdist") #+mig_phase") type_c<-glm(formula = type_full_c, family = binomial(link = "probit"), data = gi) type_c_het <- print(coeftest(type_c, vcovCL), digits=6) nobs(type_c) # Marginal effects (Average marginal effects) type_c_ame <- probitmfx(formula = type_full_c, data=gi, atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) # average marginal effects with heteroscedasticity-robust White standard errors print(type_c_ame) # Percent correctly predicted values type_c_predict<-(table(true = gi$type, pred = round(fitted(type_c)))[1,1]+table(true = gi$type, pred = round(fitted(type_c)))[2,2])/nobs(type_c)*100 # McFadden's Pseudo R-squared type_c_baseline <- update(type_c, formula= type ~ 1) type_c_McFadden <- 1 - as.vector(logLik(type_c)/logLik(type_c_baseline)) type_c_McFadden #### 1 a3 w/o FE + c w/o education #### type_full_c<-as.formula("type ~ factor(age)+female+children_d_na+part+occupationac+degree+risk+patient+english+factor(ling)+factor(europe)+factor(income)+ltodist+ivrdist") #+mig_phase") type_c_wo<-glm(formula = type_full_c, family = binomial(link = "probit"), data = giwo) type_c_het_wo <- print(coeftest(type_c_wo, vcovCL), digits=6) nobs(type_c_wo) # Marginal effects (Average marginal effects) type_c_ame_wo <- probitmfx(formula = type_full_c, data=giwo, atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) # average marginal effects with heteroscedasticity-robust White standard errors print(type_c_ame_wo) # Percent correctly predicted values type_c_predict_wo <- (table(true = giwo$type, pred = round(fitted(type_c_wo)))[1,1]+table(true = giwo$type, pred = round(fitted(type_c_wo)))[2,2])/nobs(type_c_wo)*100 # McFadden's Pseudo R-squared type_c_baseline_wo <- update(type_c_wo, formula= type ~ 1) type_c_McFadden_wo <- 1 - as.vector(logLik(type_c_wo)/logLik(type_c_baseline_wo)) type_c_McFadden_wo #### 1 a3 Output (Average Marginal Effects) #### stargazer(type_ame$fit, type_fe_ame$fit, type_c_ame$fit, coef = list(type_ame$mfxest[,1], type_fe_ame$mfxest[,1], type_c_ame$mfxest[,1]), se = list(type_ame$mfxest[,2], type_fe_ame$mfxest[,2], type_c_ame$mfxest[,2]), header=TRUE, p.auto=FALSE, style="default", nobs=TRUE, align=TRUE, no.space=TRUE, single.row=FALSE, model.names = FALSE, dep.var.labels = "Human capital investment", covariate.labels= c( "Age: under 35 years", "Gender: female", "Gender: n/a", "Children", "Partner (native German)", "Partner (other native)", "Occ.: high appl./low comm. skills", "Occ.: high appl./high comm. skills", "Occ.: in education", "Occ.: other occ./no answer", "University degree", "Risk attitude", "Patience", "English speaker", "Germanic lang.", "European (Non-EU)", "Non-European", "Upper-middle income", "High income", "Cultural distance: LTO", "Cultural distance: IVR"), omit = c("Constant", "country", "empstat4"), font.size = "small", add.lines = list(c("Country FE", "No", "Yes", "No"), c("McFadden Pseudo R2", round(type_McFadden, digits=2), round(type_fe_McFadden, digits=2), round(type_c_McFadden, digits=2)), c("Percent. correctly predicted", round(type_predict, digits=1), round(type_fe_predict, digits=1), round(type_c_predict, digits=1) )), omit.table.layout = "n", omit.stat = c("aic", "ll")) #### 1 a3 Output (Average Marginal Effects) #### stargazer(type_fe_ame_wo$fit, type_c_ame_wo$fit, type_fe_ame$fit, type_c_ame$fit, coef = list(type_fe_ame_wo$mfxest[,1], type_c_ame_wo$mfxest[,1], type_fe_ame$mfxest[,1], type_c_ame$mfxest[,1]), se = list(type_fe_ame_wo$mfxest[,2], type_c_ame_wo$mfxest[,2], type_fe_ame$mfxest[,2], type_c_ame$mfxest[,2]), header=TRUE, p.auto=FALSE, style="default", nobs=TRUE, align=TRUE, no.space=TRUE, single.row=FALSE, model.names = FALSE, dep.var.labels = "Human capital investment", covariate.labels= c( "Age: under 35 years", "Gender: female", "Gender: n/a", "Children", "Partner (native German)", "Partner (other native)", "Occ.: high appl./low comm. skills", "Occ.: high appl./high comm. skills", "Occ.: in education", "Occ.: other occ./no answer", "University degree", "Risk attitude", "Patience", "English speaker", "Germanic lang.", "European (Non-EU)", "Non-European", "Upper-middle income", "High income", "Cultural distance: LTO", "Cultural distance: IVR"), omit = c("Constant", "country", "empstat4"), font.size = "small", add.lines = list(c("Country FE", "Yes", "No", "Yes", "No"), c("McFadden Pseudo R2", round(type_fe_McFadden_wo, digits=2), round(type_c_McFadden_wo, digits=2), round(type_fe_McFadden, digits=2), round(type_c_McFadden, digits=2)), c("Percent. correctly predicted", round(type_fe_predict_wo, digits=1), round(type_c_predict_wo, digits=1), round(type_fe_predict, digits=1), round(type_c_predict, digits=1))), omit.table.layout = "n", omit.stat = c("aic", "ll")) #### 1 b subsamples ### #### 1 a3 Output (Average Marginal Effects) #### stargazer(type_fe_ame$fit, type_c_ame_nodeg$fit, type_c_ame_deg$fit, type_ame_nodeg$fit, type_ame_deg$fit, coef = list(type_fe_ame$mfxest[,1], type_c_ame_nodeg$mfxest[,1], type_c_ame_deg$mfxest[,1], type_ame_nodeg$mfxest[,1], type_ame_deg$mfxest[,1] ), se = list(type_fe_ame$mfxest[,2], type_c_ame_nodeg$mfxest[,2], type_c_ame_deg$mfxest[,2], type_ame_nodeg$mfxest[,2], type_ame_deg$mfxest[,2] ), header=TRUE, style="default", dep.var.labels = "Human capital investment", covariate.labels= c( "Age: under 35 years", "Gender: female", "Gender: n/a", "Children", "Partner (native German)", "Partner (other native)", "Occ.: in education", "Other occ./no answer", "University degree", "Risk attitude", "Patience", "English speaker", "Intern. applic. of education: high", "Comm. skills in prof. life: high", "Comm. skills in prof. life: n/a"), column.labels = c("Whole sample", "No uni. deg.", "Uni. deg.", "No uni. deg.", "Uni deg."), nobs=TRUE, no.space=TRUE, single.row=TRUE, model.names = FALSE, omit = c("Constant", "country", "empstat4"), font.size = "small", add.lines = list(c("Country FE", "Yes", "Yes", "Yes", "Yes", "Yes"), c("McFadden Pseudo R2", round(type_fe_McFadden, digits=2), round(type_c_McFadden_nodeg, digits=2), round(type_c_McFadden_deg, digits=2), round(type_McFadden_nodeg, digits=2), round(type_McFadden_deg, digits=2)), c("Percent. correctly predicted", round(type_fe_predict, digits=1), round(type_c_predict_nodeg, digits=1), round(type_c_predict_deg, digits=1), round(type_predict_nodeg, digits=1), round(type_predict_deg, digits=1) )), omit.table.layout = "n", omit.stat = c("aic", "ll")) #### 1 a2 w FE + FE:age #### table(gi$age) type_full_fea<-as.formula("type ~ factor(age)+female+children_d_na+part+occupationac+degree+risk+patient+english+country+country:factor(age)")#+mig_phase") type_fea<-glm(formula = type_full_fea, family = binomial(link = "probit"), data = gi) type_fe_heta <- print(coeftest(type_fea, vcovCL), digits=6) nobs(type_fea) # Marginal effects (Average marginal effects) type_fe_amea <- probitmfx(formula = type_full_fea, data=gi, atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) # average marginal effects with heteroscedasticity-robust White standard errors print(type_fe_amea) # Percent correctly predicted values type_fe_predicta<-(table(true = gi$type, pred = round(fitted(type_fea)))[1,1]+table(true = gi$type, pred = round(fitted(type_fea)))[2,2])/nobs(type_fea)*100 # McFadden's Pseudo R-squared type_fe_baselinea <- update(type_fea, formula= type ~ 1) type_fe_McFaddena <- 1 - as.vector(logLik(type_fea)/logLik(type_fe_baselinea)) type_fe_McFaddena #### 1 a2 w FE + FE:age + FE:gender #### table(gi$age) type_full_feag <-as.formula("type ~ factor(age)+female+children_d_na+part+occupationac+degree+risk+patient+english+country+country:factor(age)+country:female")#+mig_phase") type_feag<-glm(formula = type_full_feag, family = binomial(link = "probit"), data = gi) type_fe_hetag <- print(coeftest(type_feag, vcovCL), digits=6) nobs(type_feag) # Marginal effects (Average marginal effects) type_fe_ameag <- probitmfx(formula = type_full_feag, data=gi, atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) # average marginal effects with heteroscedasticity-robust White standard errors print(type_fe_ameag) # Percent correctly predicted values type_fe_predictag<-(table(true = gi$type, pred = round(fitted(type_feag)))[1,1]+table(true = gi$type, pred = round(fitted(type_feag)))[2,2])/nobs(type_feag)*100 # McFadden's Pseudo R-squared type_fe_baselineag <- update(type_feag, formula= type ~ 1) type_fe_McFaddenag <- 1 - as.vector(logLik(type_feag)/logLik(type_fe_baselineag)) type_fe_McFaddenag #### 1 a3 Output (Average Marginal Effects) #### stargazer(type_fe_ame$fit, type_fe_amea$fit, type_fe_ameag$fit, coef = list(type_fe_ame$mfxest[,1], type_fe_amea$mfxest[,1], type_fe_ameag$mfxest[,1]), se = list(type_fe_ame$mfxest[,2], type_fe_amea$mfxest[,2], type_fe_ameag$mfxest[,2]), header=TRUE, p.auto=FALSE, style="default", nobs=TRUE, align=TRUE, no.space=TRUE, single.row=FALSE, model.names = FALSE, dep.var.labels = "Human capital investment", covariate.labels= c( "Age: under 35 years", "Gender: female", "Gender: n/a", "Children", "Partner (native German)", "Partner (other native)", "Occ.: high appl./low comm. skills", "Occ.: high appl./high comm. skills", "Occ.: in education", "Occ.: other occ./no answer", "University degree", "Risk attitude", "Patience", "English speaker" ), omit = c("Constant", "country"), font.size = "small", add.lines = list(c("Country FE", "Yes", "Yes", "Yes"), c("McFadden Pseudo R2", round(type_fe_McFadden, digits=2), round(type_fe_McFaddena, digits=2), round(type_fe_McFaddenag, digits=2)), c("Percent. correctly predicted", round(type_fe_predict, digits=1), round(type_fe_predicta, digits=1), round(type_fe_predictag, digits=1))), omit.table.layout = "n", omit.stat = c("aic", "ll")) #### 1 b1 Female w FE #### gi_female<-gi[gi$female==1,] type_full_fe_gender<-as.formula("type ~ factor(age)+children_d_na+part+occupationac+degree+risk+patient+english+country") #+mig_phase type_female<-glm(formula = type_full_fe_gender, family = binomial(link = "probit"), data = gi_female) type_female_het <- print(coeftest(type_female, vcovCL), digits=6) nobs(type_female) # Marginal effects (Average marginal effects) type_female_ame <- probitmfx(formula = type_full_fe_gender, data=gi_female, atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) # average marginal effects with heteroscedasticity-robust White standard errors print(type_female_ame) # Percent correctly predicted values type_female_predict <-(table(true = gi_female$type, pred = round(fitted(type_female)))[1,1]+table(true = gi_female$type, pred = round(fitted(type_female)))[2,2])/nobs(type_female)*100 # McFadden's Pseudo R-squared type_female_baseline <- update(type_female, formula= type ~ 1) type_female_McFadden <- 1 - as.vector(logLik(type_female)/logLik(type_female_baseline)) type_female_McFadden #### 1 b2 Male w FE #### gi_male<-gi[gi$female==0,] type_full_fe_gender<-as.formula("type ~ factor(age)+children_d_na+part+occupationac+degree+risk+patient+english+country") #+mig_phase type_male<-glm(formula = type_full_fe_gender, family = binomial(link = "probit"), data = gi_male) type_male_het <- print(coeftest(type_male, vcovCL), digits=6) nobs(type_male) # Marginal effects (Average marginal effects) type_male_ame <- probitmfx(formula = type_full_fe_gender, data=gi_male, atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) # average marginal effects with heteroscedasticity-robust White standard errors print(type_male_ame) # Percent correctly predicted values type_male_predict <-(table(true = gi_male$type, pred = round(fitted(type_male)))[1,1]+table(true = gi_male$type, pred = round(fitted(type_male)))[2,2])/nobs(type_male)*100 # McFadden's Pseudo R-squared type_male_baseline <- update(type_male, formula= type ~ 1) type_male_McFadden <- 1 - as.vector(logLik(type_male)/logLik(type_male_baseline)) type_male_McFadden #### 1 b3 Age young w FE #### table(gi$age) gi_agey<-gi[gi$age=="TRUE",] type_full_fe_age<-as.formula("type ~ female+children_d_na+part+occupationac+degree+risk+patient+english+country") #+mig_phase type_agey<-glm(formula = type_full_fe_age, family = binomial(link = "probit"), data = gi_agey) type_agey_het <- print(coeftest(type_agey, vcovCL), digits=6) nobs(type_agey) # Marginal effects (Average marginal effects) type_agey_ame <- probitmfx(formula = type_full_fe_age, data=gi_agey, atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) # average marginal effects with heteroscedasticity-robust White standard errors print(type_agey_ame) # Percent correctly predicted values type_agey_predict <-(table(true = gi_agey$type, pred = round(fitted(type_agey)))[1,1]+table(true = gi_agey$type, pred = round(fitted(type_agey)))[2,2])/nobs(type_agey)*100 # McFadden's Pseudo R-squared type_agey_baseline <- update(type_agey, formula= type ~ 1) type_agey_McFadden <- 1 - as.vector(logLik(type_agey)/logLik(type_agey_baseline)) type_agey_McFadden #### 1 b4 Age older w FE #### table(gi$age) gi_ageo<-gi[gi$age!="TRUE",] type_full_fe_age<-as.formula("type ~ female+children_d_na+part+occupationac+degree+risk+patient+english+country") #+mig_phase type_ageo<-glm(formula = type_full_fe_age, family = binomial(link = "probit"), data = gi_ageo) type_ageo_het <- print(coeftest(type_ageo, vcovCL), digits=6) nobs(type_ageo) # Marginal effects (Average marginal effects) type_ageo_ame <- probitmfx(formula = type_full_fe_age, data=gi_ageo, atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) # average marginal effects with heteroscedasticity-robust White standard errors print(type_ageo_ame) # Percent correctly predicted values type_ageo_predict <-(table(true = gi_ageo$type, pred = round(fitted(type_ageo)))[1,1]+table(true = gi_ageo$type, pred = round(fitted(type_ageo)))[2,2])/nobs(type_ageo)*100 # McFadden's Pseudo R-squared type_ageo_baseline <- update(type_ageo, formula= type ~ 1) type_ageo_McFadden <- 1 - as.vector(logLik(type_ageo)/logLik(type_ageo_baseline)) type_ageo_McFadden #### 1 b5 no Uni_deg w FE #### gi_nodeg <- as.data.frame(subset(gi, degree==FALSE)) type_full<-as.formula("type ~ factor(age)+female+children_d_na+part+occupationac+risk+patient+english+country") #+mig_phase") type_nodeg<-glm(formula = type_full, family = binomial(link = "probit"), data = gi_nodeg) type_hetnodeg <- print(coeftest(type_nodeg, vcovCL), digits=6) nobs(type_nodeg) # Marginal effects (Average marginal effects) type_ame_nodeg <- probitmfx(formula = type_full, data=gi_nodeg, atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) # average marginal effects with heteroscedasticity-robust White standard errors print(type_ame_nodeg) # Percent correctly predicted values type_predict_nodeg<-(table(true = gi_nodeg$type, pred = round(fitted(type_nodeg)))[1,1]+table(true = gi_nodeg$type, pred = round(fitted(type_nodeg)))[2,2])/nobs(type_nodeg)*100 # McFadden's Pseudo R-squared type_baseline_nodeg <- update(type_nodeg, formula= type ~ 1) type_McFadden_nodeg <- 1 - as.vector(logLik(type_nodeg)/logLik(type_baseline_nodeg)) type_McFadden_nodeg #### 1 b6 Uni_deg w FE #### gi_deg <- as.data.frame(subset(gi, degree==TRUE)) type_full<-as.formula("type ~ factor(age)+female+children_d_na+part+occupationac+risk+patient+english+country") #+mig_phase") type_deg<-glm(formula = type_full, family = binomial(link = "probit"), data = gi_deg) type_hetdeg <- print(coeftest(type_deg, vcovCL), digits=6) nobs(type_deg) # Marginal effects (Average marginal effects) type_ame_deg <- probitmfx(formula = type_full, data=gi_deg, atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) # average marginal effects with heteroscedasticity-robust White standard errors print(type_ame_deg) # Percent correctly predicted values type_predict_deg<-(table(true = gi_deg$type, pred = round(fitted(type_deg)))[1,1]+table(true = gi_deg$type, pred = round(fitted(type_deg)))[2,2])/nobs(type_deg)*100 # McFadden's Pseudo R-squared type_baseline_deg <- update(type_deg, formula= type ~ 1) type_McFadden_deg <- 1 - as.vector(logLik(type_deg)/logLik(type_baseline_deg)) type_McFadden_deg #### 1 b6 Output AME FE Gender/ age / degree #### stargazer(type_female_ame$fit, type_male_ame$fit, type_agey_ame$fit, type_ageo_ame$fit, type_ame_nodeg$fit, type_ame_deg$fit, coef = list(type_female_ame$mfxest[,1], type_male_ame$mfxest[,1], type_agey_ame$mfxest[,1], type_ageo_ame$mfxest[,1], type_ame_nodeg$mfxest[,1], type_ame_deg$mfxest[,1]), se = list(type_female_ame$mfxest[,2], type_male_ame$mfxest[,2], type_agey_ame$mfxest[,2], type_ageo_ame$mfxest[,2], type_ame_nodeg$mfxest[,2], type_ame_deg$mfxest[,2]), header=TRUE, p.auto=FALSE, style="default", nobs=TRUE, align=TRUE, no.space=TRUE, single.row=FALSE, model.names = FALSE, dep.var.labels = "Humancapital investment", column.labels = c("female", "male", "Age: under 35 years", "Age: 35 and more years", "no Uni. deg.", "Uni. deg."), covariate.labels= c( "Age: under 35 years", "Gender: female", "Gender: n/a", "Children", "Partner (native German)", "Partner (other native)", "Occ.: high appl./low comm. skills", "Occ.: high appl./high comm. skills", "Occ.: in education", "Occ.: other occ./no answer", "University degree", "Risk attitude", "Patience", "English speaker"), omit = c("Constant", "country"), font.size = "small", add.lines = list(c("Country FE", "Yes", "Yes","Yes", "Yes", "Yes", "Yes"), c("McFadden Pseudo R2", round(type_female_McFadden, digits=2), round(type_male_McFadden, digits=2), round(type_agey_McFadden, digits=2), round(type_ageo_McFadden, digits=2), round(type_McFadden_nodeg, digits=2), round(type_McFadden_deg, digits=2)), c("Percent. correctly predicted", round(type_female_predict, digits=1), round(type_male_predict, digits=1), round(type_agey_predict, digits=1), round(type_ageo_predict, digits=1), round(type_predict_nodeg, digits=1), round(type_predict_deg, digits=1))), omit.stat = c("aic", "ll"), omit.table.layout = "n") #### 1 c Country-wise #### #### 1 c1 Countries #### TOT<-probitmfx(formula = type_full_fe, data=gi, atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) BIH<-probitmfx(formula = type_full, data=gi[gi$country=="BIH",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) CZE<-probitmfx(formula = type_full, data=gi[gi$country=="CZE",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) ESP<-probitmfx(formula = type_full, data=gi[gi$country=="ESP",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) GBR<-probitmfx(formula = type_full, data=gi[gi$country=="GBR",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) ITA<-probitmfx(formula = type_full, data=gi[gi$country=="ITA",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) JPN<-probitmfx(formula = type_full, data=gi[gi$country=="JPN",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) KOR<-probitmfx(formula = type_full, data=gi[gi$country=="KOR",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) NLD<-probitmfx(formula = type_full, data=gi[gi$country=="NLD",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) POL<-probitmfx(formula = type_full, data=gi[gi$country=="POL",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) ROU<-probitmfx(formula = type_full, data=gi[gi$country=="ROU",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) IND<-probitmfx(formula = type_full, data=gi[gi$country=="IND",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) IDN<-probitmfx(formula = type_full, data=gi[gi$country=="IDN",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) UKR<-probitmfx(formula = type_full, data=gi[gi$country=="UKR",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) MEX<-probitmfx(formula = type_full, data=gi[gi$country=="MEX",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) stargazer(TOT$fit, BIH$fit, CZE$fit, ESP$fit, GBR$fit, ITA$fit, JPN$fit, KOR$fit, NLD$fit, POL$fit, ROU$fit, IND$fit, IDN$fit, UKR$fit, MEX$fit, coef = list(TOT$mfxest[,1], BIH$mfxest[,1], CZE$mfxest[,1], ESP$mfxest[,1], GBR$mfxest[,1], ITA$mfxest[,1], JPN$mfxest[,1], KOR$mfxest[,1], NLD$mfxest[,1], POL$mfxest[,1], ROU$mfxest[,1], IND$mfxest[,1], IDN$mfxest[,1], UKR$mfxest[,1], MEX$mfxest[,1]), se = list(TOT$mfxest[,2], BIH$mfxest[,2], CZE$mfxest[,2], ESP$mfxest[,2], GBR$mfxest[,2], ITA$mfxest[,2], JPN$mfxest[,2], KOR$mfxest[,2], NLD$mfxest[,2], POL$mfxest[,2], ROU$mfxest[,2], IND$mfxest[,2], IDN$mfxest[,2], UKR$mfxest[,2], MEX$mfxest[,2]), column.labels = c("TOT", "BIH", "CZE", "ESP", "GBR", "ITA", "JPN", "KOR", "NLD", "POL", "ROU", "IND", "IDN", "UKR", "MEX"), header=FALSE, style="default", dep.var.labels = "Humancapital investment", covariate.labels= c( "Age: under 35 years", "Gender: female", "Gender: n/a", "Children", "Partner (native German)", "Partner (other native)", "Occ.: in education", "Other occ./no answer", "University degree", "Risk attitude", "Patience", "Intern. applic. of education" ), nobs=TRUE, align=FALSE, no.space=TRUE, single.row=TRUE, model.names = FALSE, omit = c("Constant", "country"), font.size = "small", add.lines = list(c("Country FE", "Yes", "No", "No","No","No","No","No","No","No","No","No","No","No","No","No")), omit.stat = c("aic", "ll")) TOT_type<-tidy(TOT) %>% filter(term != "(Intercept)") %>% mutate(model = "Total") BIH_type<-tidy(BIH) %>% filter(term != "(Intercept)") %>% mutate(model = "BIH (n=126)") CZE_type<-tidy(CZE) %>% filter(term != "(Intercept)") %>% mutate(model = "CZE (n=311)") ESP_type<-tidy(ESP) %>% filter(term != "(Intercept)") %>% mutate(model = "ESP (n=467)") GBR_type<-tidy(GBR) %>% filter(term != "(Intercept)") %>% mutate(model = "GBR (n=339)") ITA_type<-tidy(ITA) %>% filter(term != "(Intercept)") %>% mutate(model = "ITA (n=218)") JPN_type<-tidy(JPN) %>% filter(term != "(Intercept)") %>% mutate(model = "JPN (n=196)") KOR_type<-tidy(KOR) %>% filter(term != "(Intercept)") %>% mutate(model = "KOR (n=215)") NLD_type<-tidy(NLD) %>% filter(term != "(Intercept)") %>% mutate(model = "NLD (n=68)") POL_type<-tidy(POL) %>% filter(term != "(Intercept)") %>% mutate(model = "POL (n=138)") ROU_type<-tidy(ROU) %>% filter(term != "(Intercept)") %>% mutate(model = "ROU (n=223)") IND_type<-tidy(IND) %>% filter(term != "(Intercept)") %>% mutate(model = "IND (n=406)") IDN_type<-tidy(IDN) %>% filter(term != "(Intercept)") %>% mutate(model = "IDN (n=539)") UKR_type<-tidy(UKR) %>% filter(term != "(Intercept)") %>% mutate(model = "UKR (n=512)") MEX_type<-tidy(MEX) %>% filter(term != "(Intercept)") %>% mutate(model = "MEX (n=316)") gi %>% count(country) %>% arrange(-n) country_l_type<-rbind(TOT_type, ESP_type, GBR_type, CZE_type, ITA_type, POL_type, ROU_type, NLD_type, BIH_type, UKR_type, MEX_type, IND_type, IDN_type, KOR_type, JPN_type) country_l_type<-country_l_type %>% mutate(sign=ifelse(model=="Total" & p.value<0.05, TRUE, FALSE)) %>% group_by(term) %>% mutate(sign_5=max(sign), sign=ifelse(sign_5==0, "P>0.05 (aggr. model)", "P<0.05 (aggr. model)"), sign=as.factor(sign)) %>% filter(model!="Total") #### 1 c2a Country groups #### eu_sw<-probitmfx(formula = type_full_fe, data=gi[gi$country=="ESP" | gi$country=="ITA"| gi$country=="GBR" | gi$country=="NLD" ,], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) eu_e<-probitmfx(formula = type_full_fe, data=gi[gi$country=="CZE" | gi$country=="POL" | gi$country=="ROU",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) eur<- probitmfx(formula = type_full_fe, data=gi[gi$country=="UKR" | gi$country=="BIH",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) noneu_d<-probitmfx(formula = type_full_fe, data=gi[gi$country=="JPN" | gi$country=="KOR" | gi$country=="MEX",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) noneu_ld<-probitmfx(formula = type_full_fe, data=gi[gi$country=="IDN" | gi$country=="IND" ,], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) eu<-probitmfx(formula = type_full_fe, data=gi[gi$country=="ESP" | gi$country=="ITA"|gi$country=="NLD" | gi$country=="GBR"|gi$country=="CZE" | gi$country=="POL" | gi$country=="ROU",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) noneu<- probitmfx(formula = type_full_fe, data=gi[gi$country=="MEX" | gi$country=="IDN" | gi$country=="IND"|gi$country=="JPN" | gi$country=="KOR"|gi$country=="UKR" | gi$country=="BIH",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) stargazer(eu_sw$fit, eu_e$fit, eur$fit, noneu_d$fit, noneu_ld$fit, eu$fit, noneu$fit, coef = list(eu_sw$mfxest[,1], eu_e$mfxest[,1], eur$mfxest[,1], noneu_d$mfxest[,1], noneu_ld$mfxest[,1], eu$mfxest[,1], noneu$mfxest[,1]), se = list(eu_sw$mfxest[,2], eu_e$mfxest[,2], eur$mfxest[,2], noneu_d$mfxest[,2], noneu_ld$mfxest[,2], eu$mfxest[,2], noneu$mfxest[,2]), column.labels = c("ESP, ITA, GBR, NLD", "CZE, POL, ROU", "UKR, BIH", "JPN, KOR, MEX", "IND, IDN", "EU", "Non-EU"), header=FALSE, style="default", dep.var.labels = "Human capital investment", covariate.labels= c( "Age: under 35 years", "Gender: female", "Gender: n/a", "Children", "Partner (native German)", "Partner (other native)", "Occ.: in education", "Other occ./no answer", "University degree", "Risk attitude", "Patience", "Intern. applic. of education" ), nobs=TRUE, align=FALSE, no.space=TRUE, single.row=FALSE, model.names = FALSE, omit = c("Constant", "country"), font.size = "small", add.lines = list(c("Country FE", "Yes", "Yes","Yes","Yes","Yes","Yes","Yes")), omit.stat = c("aic", "ll"), omit.table.layout = "n") eu_sw_type<-tidy(eu_sw) %>% filter(term != "(Intercept)") %>% mutate(model = "ESP, ITA, GBR, NLD (n=1118)") nrow(gi[gi$country=="ESP" | gi$country=="ITA"| gi$country=="GBR" | gi$country=="NLD" ,]) eu_e_type<-tidy(eu_e) %>% filter(term != "(Intercept)") %>% mutate(model = "CZE, POL, ROU (n=704)") nrow(gi[gi$country=="CZE" | gi$country=="POL" | gi$country=="ROU",]) eur_type<-tidy(eur) %>% filter(term != "(Intercept)") %>% mutate(model = "UKR, BIH (n=683)") nrow(gi[gi$country=="UKR" | gi$country=="BIH",]) noneu_d_type<-tidy(noneu_d) %>% filter(term != "(Intercept)") %>% mutate(model = "JPN, KOR, MEX (n=848)") nrow(gi[gi$country=="JPN" | gi$country=="KOR" | gi$country=="MEX",]) noneu_ld_type<-tidy(noneu_ld) %>% filter(term != "(Intercept)") %>% mutate(model = "IND, IDN (n=1003)") nrow(gi[gi$country=="IDN" | gi$country=="IND",]) eu_type<-tidy(eu) %>% filter(term != "(Intercept)") %>% mutate(model = "EU (n=1822)") nrow(gi[gi$country=="ESP" | gi$country=="ITA"|gi$country=="NLD" | gi$country=="GBR"|gi$country=="CZE" | gi$country=="POL" | gi$country=="ROU",]) noneu_type<-tidy(noneu) %>% filter(term != "(Intercept)") %>% mutate(model = "Non-EU (n=2534)") nrow(gi[gi$country=="MEX" | gi$country=="IDN" | gi$country=="IND"|gi$country=="JPN" | gi$country=="KOR"|gi$country=="UKR" | gi$country=="BIH",]) countrygr_l_type<-rbind(TOT_type, eu_sw_type, eu_e_type, eur_type, noneu_d_type, noneu_ld_type, eu_type, noneu_type) countrygr_l_type<-countrygr_l_type %>% mutate(sign=ifelse(model=="Total" & p.value<0.05, TRUE, FALSE)) %>% group_by(term) %>% mutate(sign_5=max(sign), sign=ifelse(sign_5==0, "P>0.05 (aggr. model)", "P<0.05 (aggr. model)"), sign=as.factor(sign)) %>% filter(model!="Total") #### 1 c2b Country characteristics #### germ<-probitmfx(formula = type_full_fe, data=gi[gi$country=="NLD" | gi$country=="GBR" | gi$country=="IND",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) ngerm<-probitmfx(formula = type_full_fe, data=gi[gi$country=="ESP" | gi$country=="ITA" | gi$country=="CZE" |gi$country=="ROU" | gi$country=="BIH" | gi$country=="MEX"| gi$country=="IND"| gi$country=="UKR" | gi$country=="JPN" | gi$country=="KOR" | gi$country=="POL", ], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) hi<-probitmfx(formula = type_full_fe, data=gi[gi$country=="ESP" | gi$country=="ITA"| gi$country=="GBR" | gi$country=="NLD" | gi$country=="CZE" | gi$country=="POL" | gi$country=="JPN"| gi$country=="KOR" ,], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) umi<-probitmfx(formula = type_full_fe, data=gi[gi$country=="ROU" | gi$country=="BIH" | gi$country=="MEX",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) lmi<-probitmfx(formula = type_full_fe, data=gi[gi$country=="IDN" | gi$country=="IND"| gi$country=="UKR" ,], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) mi<-probitmfx(formula = type_full_fe, data=gi[gi$country=="IDN" | gi$country=="IND"| gi$country=="UKR" | gi$country=="ROU" | gi$country=="BIH" | gi$country=="MEX" ,], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) eur<- probitmfx(formula = type_full_fe, data=gi[gi$country=="UKR" | gi$country=="BIH",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) eu<-probitmfx(formula = type_full_fe, data=gi[gi$country=="ESP" | gi$country=="ITA"|gi$country=="NLD" | gi$country=="GBR"|gi$country=="CZE" | gi$country=="POL" | gi$country=="ROU",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) noneur<- probitmfx(formula = type_full_fe, data=gi[gi$country=="MEX" | gi$country=="IDN" | gi$country=="IND"|gi$country=="JPN" | gi$country=="KOR",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) noneu<- probitmfx(formula = type_full_fe, data=gi[gi$country=="MEX" | gi$country=="IDN" | gi$country=="IND"|gi$country=="JPN" | gi$country=="KOR"|gi$country=="UKR" | gi$country=="BIH",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) cultclose<- probitmfx(formula = type_full_fe, data=gi[gi$country=="JPN" | gi$country=="BIH" | gi$country=="CZE"|gi$country=="IDN" | gi$country=="UKR"|gi$country=="KOR" | gi$country=="ITA",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) cultdist<- probitmfx(formula = type_full_fe, data=gi[gi$country=="MEX" | gi$country=="IND" | gi$country=="ESP"|gi$country=="NLD" | gi$country=="ROU"|gi$country=="POL" | gi$country=="GBR",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) stargazer(germ$fit, ngerm$fit, umi$fit, lmi$fit, hi$fit, eu$fit, eur$fit, noneur$fit, cultclose$fit, cultdist$fit, coef = list(germ$mfxest[,1], ngerm$mfxest[,1], lmi$mfxest[,1], umi$mfxest[,1], mi$mfxest[,1], hi$mfxest[,1], eu$mfxest[,1], eur$mfxest[,1], noneur$mfxest[,1], cultclose$mfxest[,1], cultdist$mfxest[,1] ), se = list(germ$mfxest[,2], ngerm$mfxest[,2], lmi$mfxest[,2], umi$mfxest[,2], hi$mfxest[,2], eu$mfxest[,2], eur$mfxest[,2], noneur$mfxest[,2], cultclose$mfxest[,2], cultdist$mfxest[,2] ), column.labels = c("Germanic", "Non-Germanic", "Lower-middle income", "Upper-middle income", "High income", "EU", "European (non-EU)", "Non-European", "Culturally close", "Culturally far" ), header=TRUE, p.auto=FALSE, style="default", nobs=TRUE, align=TRUE, no.space=TRUE, single.row=FALSE, model.names = FALSE, dep.var.labels = "Human capital investment", covariate.labels= c( "Age: under 35 years", "Gender: female", "Gender: n/a", "Children", "Partner (native German)", "Partner (other native)", "Occ.: high appl./low comm. skills", "Occ.: high appl./high comm. skills", "Occ.: in education", "Occ.: other occ./no answer", "University degree", "Risk attitude", "Patience", "English speaker"), omit = c("Constant", "country"), font.size = "small", add.lines = list(c("Country FE", "Yes", "Yes","Yes","Yes","Yes","Yes","Yes","Yes","Yes","Yes")), omit.stat = c("aic", "ll"), omit.table.layout = "n") germ_type<-tidy(germ) %>% filter(term != "(Intercept)") %>% mutate(model = "Germanic") ngerm_type<-tidy(ngerm) %>% filter(term != "(Intercept)") %>% mutate(model = "Non-Germanic") umi_type<-tidy(umi) %>% filter(term != "(Intercept)") %>% mutate(model = "Upper-middle income") lmi_type<-tidy(lmi) %>% filter(term != "(Intercept)") %>% mutate(model = "Lower-middle income") hi_type<-tidy(hi) %>% filter(term != "(Intercept)") %>% mutate(model = "High income") mi_type<-tidy(mi) %>% filter(term != "(Intercept)") %>% mutate(model = "Middle income") eu_type<-tidy(eu) %>% filter(term != "(Intercept)") %>% mutate(model = "EU") noneu_type<-tidy(noneu) %>% filter(term != "(Intercept)") %>% mutate(model = "Non-EU (European)") eur_type<-tidy(eur) %>% filter(term != "(Intercept)") %>% mutate(model = "Non-EU (European)") noneur_type<-tidy(noneur) %>% filter(term != "(Intercept)") %>% mutate(model = "Non-EU (non-European)") cultclose_type<-tidy(cultclose) %>% filter(term != "(Intercept)") %>% mutate(model = "Cuturally close") cultdist_type<-tidy(cultdist) %>% filter(term != "(Intercept)") %>% mutate(model = "Culturally dist.") countrygr_l_type<-rbind(TOT_type, ngerm_type, germ_type, lmi_type, umi_type, hi_type, eu_type, eur_type, noneur_type, cultclose_type, cultdist_type) countrygr_l_type<-countrygr_l_type %>% mutate(sign=ifelse(model=="Total" & p.value<0.05, TRUE, FALSE)) %>% group_by(term) %>% mutate(sign_5=max(sign), sign=ifelse(sign_5==0, "P>0.05 (aggr. model)", "P<0.05 (aggr. model)"), sign=as.factor(sign)) %>% filter(model!="Total") countrygr_l_type<-rbind(TOT_type, ngerm_type, germ_type, lmi_type, umi_type, hi_type, eu_type, eur_type, noneur_type, cultclose_type, cultdist_type) countrygr_l_type<-countrygr_l_type %>% mutate(sign=ifelse(model=="Total" & p.value<0.05, TRUE, FALSE)) %>% group_by(term) %>% mutate(sign_5=max(sign), sign=ifelse(sign_5==0, "P>0.05 (aggr. model)", "P<0.05 (aggr. model)"), sign=as.factor(sign)) %>% filter(model!="Total") p_countrygroup<-dwplot(subset(countrygr_l_type, (countrygr_l_type$term=="factor(age)TRUE" | countrygr_l_type$term=="female1" | countrygr_l_type$term=="children_d_naTRUE" | countrygr_l_type$term=="part1" | countrygr_l_type$term=="part2" | countrygr_l_type$term=="occupationac1" | countrygr_l_type$term=="occupationac2" | countrygr_l_type$term=="occupationac3" | countrygr_l_type$term=="degreeTRUE" | countrygr_l_type$term=="risk" | countrygr_l_type$term=="patient" | countrygr_l_type$term=="englishTRUE" )), vline = geom_vline(xintercept = 0, colour = "grey60", linetype = 2), dodge_size= 0.7, dot_args = list(aes(shape =as.factor(sign)), size=4)) + scale_x_continuous(limits = c(-0.7, 0.35))+ scale_y_discrete( labels=c("factor(age)TRUE" = "Age: under 35 years", female1 = "Gender: female", children_d_naTRUE = "Children", "part1" = "Partner (native German)", "part2" = "Partner (other native)", degreeTRUE= "University degree", "occupationac1" = "Occ.: high appl./low comm. skills", "occupationac2" = "Occ.: high appl./high comm. skills", "occupationac3" = "Occ.: in education", risk = "Risk attitude", patient = "Patience", "englishTRUE" = "English speaker" ))+ xlab("AME")+ theme(legend.title = element_blank())+ scale_color_grey(guide=guide_legend(reverse = TRUE)) pdf("../graphics/invest_countrychar.pdf", paper="a4", width = 12, height = 14) p_countrygroup dev.off() #### 1 c3 leave one country out #### TOT<-probitmfx(formula = type_full_fe, data=gi, atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) BIH<-probitmfx(formula = type_full_fe, data=gi[gi$country!="BIH",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) CZE<-probitmfx(formula = type_full_fe, data=gi[gi$country!="CZE",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) ESP<-probitmfx(formula = type_full_fe, data=gi[gi$country!="ESP",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) GBR<-probitmfx(formula = type_full_fe, data=gi[gi$country!="GBR",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) ITA<-probitmfx(formula = type_full_fe, data=gi[gi$country!="ITA",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) JPN<-probitmfx(formula = type_full_fe, data=gi[gi$country!="JPN",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) KOR<-probitmfx(formula = type_full_fe, data=gi[gi$country!="KOR",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) NLD<-probitmfx(formula = type_full_fe, data=gi[gi$country!="NLD",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) POL<-probitmfx(formula = type_full_fe, data=gi[gi$country!="POL",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) ROU<-probitmfx(formula = type_full_fe, data=gi[gi$country!="ROU",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) IND<-probitmfx(formula = type_full_fe, data=gi[gi$country!="IND",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) IDN<-probitmfx(formula = type_full_fe, data=gi[gi$country!="IDN",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) UKR<-probitmfx(formula = type_full_fe, data=gi[gi$country!="UKR",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) MEX<-probitmfx(formula = type_full_fe, data=gi[gi$country!="MEX",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) stargazer(TOT$fit, BIH$fit, CZE$fit, ESP$fit, GBR$fit, ITA$fit, JPN$fit, KOR$fit, NLD$fit, POL$fit, ROU$fit, IND$fit, IDN$fit, UKR$fit, MEX$fit, coef = list(TOT$mfxest[,1], BIH$mfxest[,1], CZE$mfxest[,1], ESP$mfxest[,1], GBR$mfxest[,1], ITA$mfxest[,1], JPN$mfxest[,1], KOR$mfxest[,1], NLD$mfxest[,1], POL$mfxest[,1], ROU$mfxest[,1], IND$mfxest[,1], IDN$mfxest[,1], UKR$mfxest[,1], MEX$mfxest[,1]), se = list(TOT$mfxest[,2], BIH$mfxest[,2], CZE$mfxest[,2], ESP$mfxest[,2], GBR$mfxest[,2], ITA$mfxest[,2], JPN$mfxest[,2], KOR$mfxest[,2], NLD$mfxest[,2], POL$mfxest[,2], ROU$mfxest[,2], IND$mfxest[,2], IDN$mfxest[,2], UKR$mfxest[,2], MEX$mfxest[,2]), column.labels = c("TOT", "BIH", "CZE", "ESP", "GBR", "ITA", "JPN", "KOR", "NLD", "POL", "ROU", "IND", "IDN", "UKR", "MEX"), header=FALSE, style="default", dep.var.labels = "Humancapital investment", covariate.labels= c( "Age: under 35 years", "Gender: female", "Gender: n/a", "Children", "Partner (native German)", "Partner (other native)", "Occ.: in education", "Occ.: not in lab. market", "University degree", "Risk attitude", "Patience", "Intern. applic. of education" ), nobs=TRUE, align=TRUE, no.space=TRUE, single.row=TRUE, model.names = FALSE, omit = c("Constant", "country", "empstat4"), font.size = "small", add.lines = list(c("Country FE", "Yes", "Yes", "Yes","Yes","Yes","Yes","Yes","Yes","Yes","Yes","Yes","Yes","Yes","Yes","Yes")), notes= "Average marginal effects. Heteroscedasticity robust White standard errors in parentheses.", omit.stat = c("aic", "ll")) TOT_type<-tidy(TOT) %>% filter(term != "(Intercept)") %>% mutate(model = "Total") BIH_type<-tidy(BIH) %>% filter(term != "(Intercept)") %>% mutate(model = "BIH (n=110)") CZE_type<-tidy(CZE) %>% filter(term != "(Intercept)") %>% mutate(model = "CZE (n=280)") ESP_type<-tidy(ESP) %>% filter(term != "(Intercept)") %>% mutate(model = "ESP (n=395)") GBR_type<-tidy(GBR) %>% filter(term != "(Intercept)") %>% mutate(model = "GBR (n=274)") ITA_type<-tidy(ITA) %>% filter(term != "(Intercept)") %>% mutate(model = "ITA (n=186)") JPN_type<-tidy(JPN) %>% filter(term != "(Intercept)") %>% mutate(model = "JPN (n=183)") KOR_type<-tidy(KOR) %>% filter(term != "(Intercept)") %>% mutate(model = "KOR (n=189)") NLD_type<-tidy(NLD) %>% filter(term != "(Intercept)") %>% mutate(model = "NLD (n=60)") POL_type<-tidy(POL) %>% filter(term != "(Intercept)") %>% mutate(model = "POL (n=121)") ROU_type<-tidy(ROU) %>% filter(term != "(Intercept)") %>% mutate(model = "ROU (n=197)") IND_type<-tidy(IND) %>% filter(term != "(Intercept)") %>% mutate(model = "IND (n=373)") IDN_type<-tidy(IDN) %>% filter(term != "(Intercept)") %>% mutate(model = "IDN (n=485)") UKR_type<-tidy(UKR) %>% filter(term != "(Intercept)") %>% mutate(model = "UKR (n=458)") MEX_type<-tidy(MEX) %>% filter(term != "(Intercept)") %>% mutate(model = "MEX (n=281)") gi %>% count(country) %>% arrange(-n) country_l_type<-rbind(TOT_type, ESP_type, GBR_type, CZE_type, ITA_type, POL_type, ROU_type, NLD_type, BIH_type, UKR_type, MEX_type, IND_type, IDN_type, KOR_type, JPN_type) country_l_type<-country_l_type %>% mutate(sign=ifelse(model=="Total" & p.value<0.05, TRUE, FALSE)) %>% group_by(term) %>% mutate(sign_5=max(sign), sign=ifelse(sign_5==0, "P>0.05 (aggr. model)", "P<0.05 (aggr. model)"), sign=as.factor(sign)) #### 2 Binary logit langjob #### gi<-est[est$complete=="TRUE",] gi$langjob<-ifelse(!is.na(gi$langjob), (gi$langjob-6)*(-1), NA) table(gi$langjob, gi$country) table(applic=gi[gi$type==TRUE,]$applic_gen, langjob=gi[gi$type==TRUE,]$langjob) mean(gi$langjob) median(gi$langjob) table(gi$langjob) gi<-gi %>% mutate(langjob_d=ifelse(langjob>3, TRUE, ifelse(is.na(langjob), NA, FALSE))) mean(gi$langjob_d) langjob_full_fe<-as.formula("langjob_d ~ factor(age)+female+children_d_na+part+occupationac+degree+risk+patient+english+country") #+mig_phase langjob_full_type_fe<-as.formula("langjob_d ~ type + factor(age)+female+children_d_na+part+occupationac+degree+risk+patient+english+country") #+mig_phase langjob_full_type_fe_g<-as.formula("langjob_d ~ type + factor(age)+children_d_na+part+occupationac+degree+risk+patient+english+country") #+mig_phase langjob_full_type2_fe<-as.formula("langjob_d ~ type2+factor(age)+female+children_d_na+part+occupationac+degree+risk+patient+english+country") #+mig_phase langjob_full_type2_fe_g<-as.formula("langjob_d ~ type2+factor(age)+female+children_d_na+part+occupationac+degree+risk+patient+english+country") #+mig_phase langjob_full_type2_c<-as.formula("langjob_d ~ type+factor(age)+female+children_d_na+part+occupationac+degree+risk+patient+english+factor(ling)+factor(europe)+factor(income)+ltodist+ivrdist") #+mig_phase langjob_full_type2_c_g<-as.formula("langjob_d ~ type+factor(age)+female+children_d_na+part+occupationac+degree+risk+patient+english+factor(ling)+factor(europe)+factor(income)+ltodist+ivrdist") #+mig_phase langjob_full_type2<-as.formula("langjob_d ~ type2+factor(age)+female+children_d_na+part+occupationac+degree+risk+patient+english") #+mig_phase langjob_full_type2_g<-as.formula("langjob_d ~ type2+factor(age)+female+children_d_na+part+occupationac+degree+risk+patient+english") #+mig_phase #### 2 a1 base langjob #### langjob_fe<-glm(formula = langjob_full_fe, family = binomial(link = "probit"), data = gi) langjob_fe_het <- print(coeftest(langjob_fe, vcovCL), digits=6) nobs(langjob_fe) # Marginal effects (Average marginal effects) langjob_fe_ame <- probitmfx(formula = langjob_full_fe, data=gi, atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) # average marginal effects with heteroscedasticity-robust White standard errors print(langjob_fe_ame) # Percent correctly predicted values langjob_fe_predict<- (table(true = gi$langjob_d, pred = round(fitted(langjob_fe)))[1,1]+table(true = gi$langjob_d, pred = round(fitted(langjob_fe)))[2,2] ) /nobs(langjob_fe) *100 # 71,4 % correctly predicted values # McFadden's Pseudo R-squared langjob_fe_baseline <- update(langjob_fe, formula= langjob_d ~ 1) langjob_fe_McFadden <- 1 - as.vector(logLik(langjob_fe)/logLik(langjob_fe_baseline)) langjob_fe_McFadden #### 2 a2 type #### langjob_type_fe<-glm(formula = langjob_full_type_fe, family = binomial(link = "probit"), data = gi) langjob_type_fe_het <- print(coeftest(langjob_type_fe, vcovCL), digits=6) nobs(langjob_type_fe) # Marginal effects (Average marginal effects) langjob_type_fe_ame <- probitmfx(formula = langjob_full_type_fe, data=gi, atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) # average marginal effects with heteroscedasticity-robust White standard errors print(langjob_type_fe_ame) # Percent correctly predicted values langjob_type_fe_predict<- (table(true = gi$langjob_d, pred = round(fitted(langjob_type_fe)))[1,1]+table(true = gi$langjob_d, pred = round(fitted(langjob_type_fe)))[2,2] ) /nobs(langjob_type_fe) *100# 71,4 % correctly predicted values # McFadden's Pseudo R-squared langjob_type_fe_baseline <- update(langjob_type_fe, formula= langjob_d ~ 1) langjob_type_fe_McFadden <- 1 - as.vector(logLik(langjob_type_fe)/logLik(langjob_type_fe_baseline)) langjob_type_fe_McFadden #### 2 a2 type w/o education #### langjob_type_fe_wo<-glm(formula = langjob_full_type_fe, family = binomial(link = "probit"), data = giwo) langjob_type_fe_het_wo <- print(coeftest(langjob_type_fe_wo, vcovCL), digits=6) nobs(langjob_type_fe_wo) # Marginal effects (Average marginal effects) langjob_type_fe_ame_wo <- probitmfx(formula = langjob_full_type_fe, data=giwo, atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) # average marginal effects with heteroscedasticity-robust White standard errors print(langjob_type_fe_ame_wo) # Percent correctly predicted values langjob_type_fe_predict_wo <- (table(true = giwo$langjob_d, pred = round(fitted(langjob_type_fe_wo)))[1,1]+table(true = giwo$langjob_d, pred = round(fitted(langjob_type_fe_wo)))[2,2] ) /nobs(langjob_type_fe_wo) *100# 71,4 % correctly predicted values # McFadden's Pseudo R-squared langjob_type_fe_baseline_wo <- update(langjob_type_fe_wo, formula= langjob_d ~ 1) langjob_type_fe_McFadden_wo <- 1 - as.vector(logLik(langjob_type_fe_wo)/logLik(langjob_type_fe_baseline_wo)) langjob_type_fe_McFadden_wo #### 2 a3 type2 #### table(gi$type2) langjob_type2_fe<-glm(formula = langjob_full_type2_fe, family = binomial(link = "probit"), data = gi[gi$type2!="No answer",]) langjob_type2_fe_het <- print(coeftest(langjob_type2_fe, vcovCL), digits=6) nobs(langjob_type2_fe) # Marginal effects (Average marginal effects) langjob_type2_fe_ame <- probitmfx(formula = langjob_full_type2_fe, data = gi[gi$type2!="No answer",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) # average marginal effects with heteroscedasticity-robust White standard errors print(langjob_type2_fe_ame) # Percent correctly predicted values langjob_type2_fe_predict<- (table(true = gi[gi$type2!="No answer",]$langjob_d, pred = round(fitted(langjob_type2_fe)))[1,1]+table(true =gi[gi$type2!="No answer",]$langjob_d, pred = round(fitted(langjob_type2_fe)))[2,2] ) /nobs(langjob_type2_fe) *100 # 71,4 % correctly predicted values # McFadden's Pseudo R-squared langjob_type2_fe_baseline <- update(langjob_type2_fe, formula= langjob_d ~ 1) langjob_type2_fe_McFadden <- 1 - as.vector(logLik(langjob_type2_fe)/logLik(langjob_type2_fe_baseline)) langjob_type2_fe_McFadden #### 2 a4 type2 + c #### table(gi$type2) langjob_type2_c<-glm(formula = langjob_full_type2_c, family = binomial(link = "probit"), data = gi) langjob_type2_c_het <- print(coeftest(langjob_type2_c, vcovCL), digits=6) nobs(langjob_type2_c) # Marginal effects (Average marginal effects) langjob_type2_c_ame <- probitmfx(formula = langjob_full_type2_c, data = gi, atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) # average marginal effects with heteroscedasticity-robust White standard errors print(langjob_type2_c_ame) # Percent correctly predicted values langjob_type2_c_predict<- (table(true = gi$langjob_d, pred = round(fitted(langjob_type2_c)))[1,1]+table(true =gi$langjob_d, pred = round(fitted(langjob_type2_c)))[2,2] ) /nobs(langjob_type2_c) *100 # 71,4 % correctly predicted values # McFadden's Pseudo R-squared langjob_type2_c_baseline <- update(langjob_type2_c, formula= langjob_d ~ 1) langjob_type2_c_McFadden <- 1 - as.vector(logLik(langjob_type2_c)/logLik(langjob_type2_c_baseline)) langjob_type2_c_McFadden # #### 2 a4 type2 + c w/o education and w/education as consumption motive (uncomment if neccessary) #### # table(giwo$type2) # langjob_type2_c_wo<-glm(formula = langjob_full_type2_c, family = binomial(link = "probit"), data = giwo) # langjob_type2_c_het_wo <- print(coeftest(langjob_type2_c_wo, vcovCL), digits=6) # nobs(langjob_type2_c_wo) # # # # Marginal effects (Average marginal effects) # langjob_type2_c_ame_wo <- probitmfx(formula = langjob_full_type2_c, # data = giwo, # atmean = FALSE, # robust = TRUE, # clustervar1 = NULL, clustervar2 = NULL, # start = NULL, control = list()) # # average marginal effects with heteroscedasticity-robust White standard errors # print(langjob_type2_c_ame_wo) # # # # Percent correctly predicted values # langjob_type2_c_predict_wo<- (table(true = giwo$langjob_d, pred = round(fitted(langjob_type2_c_wo)))[1,1]+table(true =giwo$langjob_d, pred = round(fitted(langjob_type2_c_wo)))[2,2] ) /nobs(langjob_type2_c_wo) *100 # 71,4 % correctly predicted values # # # # McFadden's Pseudo R-squared # langjob_type2_c_baseline_wo <- update(langjob_type2_c_wo, formula= langjob_d ~ 1) # langjob_type2_c_McFadden_wo <- 1 - as.vector(logLik(langjob_type2_c_wo)/logLik(langjob_type2_c_baseline_wo)) # langjob_type2_c_McFadden_wo # # #### 2 a4 Output #### stargazer(langjob_fe_ame$fit, langjob_type_fe_ame$fit, langjob_type2_fe_ame$fit, langjob_type2_c_ame$fit, coef = list(langjob_fe_ame$mfxest[,1], langjob_type_fe_ame$mfxest[,1], langjob_type2_fe_ame$mfxest[,1], langjob_type2_c_ame$mfxest[,1]), se = list(langjob_fe_ame$mfxest[,2], langjob_type_fe_ame$mfxest[,2], langjob_type2_fe_ame$mfxest[,2], langjob_type2_c_ame$mfxest[,2]), header=TRUE, p.auto=FALSE, style="default", nobs=TRUE, align=TRUE, no.space=TRUE, single.row=FALSE, model.names = FALSE, dep.var.labels = "Professional use of German", covariate.labels= c( "Investment", "Domestic labour market", "Educational and labour migration", "Personal reasons", "Cultural interest", "Age: under 35 years", "Gender: female", "Gender: n/a", "Children", "Partner (native German)", "Partner (other native)", "Occ.: high appl./low comm. skills", "Occ.: high appl./high comm. skills", "Occ.: in education", "Occ.: other occ./no answer", "University degree", "Risk attitude", "Patience", "English speaker", "Germanic lang.", "European (Non-EU)", "Non-European", "Upper-middle income", "High income", "Cultural distance: LTO", "Cultural distance: IVR"), omit = c("Constant", "country", "empstat4"), font.size = "small", add.lines = list(c("Country FE", "Yes", "Yes", "Yes", "No"), c("McFadden Pseudo R2", round(langjob_fe_McFadden , digits=2), round(langjob_type_fe_McFadden , digits=2) , round(langjob_type2_fe_McFadden , digits=2), round(langjob_type2_c_McFadden , digits=2) ), c("Percent. correctly predicted", round(langjob_fe_predict, digits=1) , round(langjob_type_fe_predict, digits=1) , round(langjob_type2_fe_predict, digits=1) , round(langjob_type2_c_predict, digits=1) )), omit.stat = c("aic", "ll"), omit.table.layout = "n") # #### 2 a5 Output w/o education and w/education as consumption motive (uncomment if neccessary) ----------- # stargazer(langjob_type_fe_ame_wo$fit, # langjob_type2_c_ame_wo$fit, # langjob_type_fe_ame$fit, # langjob_type2_c_ame$fit, # coef = list(langjob_type_fe_ame_wo$mfxest[,1], # langjob_type2_c_ame_wo$mfxest[,1], # langjob_type_fe_ame$mfxest[,1], # langjob_type2_c_ame$mfxest[,1]), # se = list(langjob_type_fe_ame_wo$mfxest[,2], # langjob_type2_c_ame_wo$mfxest[,2], # langjob_type_fe_ame$mfxest[,2], # langjob_type2_c_ame$mfxest[,2]), # header=TRUE, # p.auto=FALSE, # style="default", # nobs=TRUE, # align=TRUE, # no.space=TRUE, # single.row=FALSE, # model.names = FALSE, # dep.var.labels = "Professional use of German", # covariate.labels= c( "Investment", # "Age: under 35 years", # "Gender: female", # "Gender: n/a", # "Children", # "Partner (native German)", # "Partner (other native)", # "Occ.: high appl./low comm. skills", # "Occ.: high appl./high comm. skills", # "Occ.: in education", # "Occ.: other occ./no answer", # "University degree", # "Risk attitude", # "Patience", # "English speaker", # "Germanic lang.", # "Non-EU (European)", # "Non-EU (Non-European)", # "Upper-middle income", # "High income", # "Cultural distance: LTO", # "Cultural distance: IVR"), # omit = c("Constant", "country", "empstat4"), # font.size = "small", # add.lines = list(c("Country FE", "Yes", "No", "Yes", "No"), # c("McFadden Pseudo R2", # round(langjob_type_fe_McFadden_wo, digits=2), # round(langjob_type2_c_McFadden_wo, digits=2), # round(langjob_type_fe_McFadden, digits=2), # round(langjob_type2_c_McFadden, digits=2) # ), # c("Percent. correctly predicted", # round(langjob_type_fe_predict_wo, digits=1), # round(langjob_type2_c_predict_wo, digits=1), # round(langjob_type_fe_predict, digits=1), # round(langjob_type2_c_predict, digits=1) # )), # omit.stat = c("aic", "ll"), # omit.table.layout = "n") # #### 2 b1a base langjob female #### gi_female<-gi[gi$female==1,] langjob_female_full_fe<-as.formula("langjob_d ~ factor(age)+children_d_na+part+occupationac+degree+risk+patient+english+country") #+mig_phase langjob_female_full_type_fe<-as.formula("langjob_d ~ type + factor(age)+children_d_na+part+occupationac+degree+risk+patient+english+country") #+mig_phase langjob_female_full_type2_fe<-as.formula("langjob_d ~ type2+factor(age)+children_d_na+part+occupationac+degree+risk+patient+english+country") #+mig_phase langjob_female_fe<-glm(formula = langjob_female_full_fe, family = binomial(link = "probit"), data = gi_female) langjob_female_fe_het <- print(coeftest(langjob_female_fe, vcovCL), digits=6) nobs(langjob_female_fe) # Marginal effects (Average marginal effects) langjob_female_fe_ame <- probitmfx(formula = langjob_female_full_fe, data=gi_female, atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) # average marginal effects with heteroscedasticity-robust White standard errors print(langjob_female_fe_ame) # Percent correctly predicted values langjob_female_fe_predict<- (table(true = gi_female$langjob_d, pred = round(fitted(langjob_female_fe)))[1,1]+table(true = gi_female$langjob_d, pred = round(fitted(langjob_female_fe)))[2,2] ) /nobs(langjob_female_fe) *100 # 71,4 % correctly predicted values # McFadden's Pseudo R-squared langjob_female_fe_baseline <- update(langjob_female_fe, formula= langjob_d ~ 1) langjob_female_fe_McFadden <- 1 - as.vector(logLik(langjob_female_fe)/logLik(langjob_female_fe_baseline)) langjob_female_fe_McFadden #### 2 b1b type female#### langjob_female_type_fe<-glm(formula = langjob_full_type_fe_g, family = binomial(link = "probit"), data = gi_female) langjob_female_type_fe_het <- print(coeftest(langjob_female_type_fe, vcovCL), digits=6) nobs(langjob_female_type_fe) # Marginal effects (Average marginal effects) langjob_female_type_fe_ame <- probitmfx(formula = langjob_female_full_type_fe, data=gi_female, atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) # average marginal effects with heteroscedasticity-robust White standard errors print(langjob_female_type_fe_ame) # Percent correctly predicted values langjob_female_type_fe_predict<- (table(true = gi_female$langjob_d, pred = round(fitted(langjob_female_type_fe)))[1,1]+table(true = gi_female$langjob_d, pred = round(fitted(langjob_female_type_fe)))[2,2] ) /nobs(langjob_female_type_fe) *100# 71,4 % correctly predicted values # McFadden's Pseudo R-squared langjob_female_type_fe_baseline <- update(langjob_female_type_fe, formula= langjob_d ~ 1) langjob_female_type_fe_McFadden <- 1 - as.vector(logLik(langjob_female_type_fe)/logLik(langjob_female_type_fe_baseline)) langjob_female_type_fe_McFadden #### 2 b1c type2 female #### langjob_female_type2_fe<-glm(formula = langjob_female_full_type2_fe, family = binomial(link = "probit"), data = gi_female[gi_female$type2!="No answer",]) langjob_female_type2_fe_het <- print(coeftest(langjob_female_type2_fe, vcovCL), digits=6) nobs(langjob_female_type2_fe) # Marginal effects (Average marginal effects) langjob_female_type2_fe_ame <- probitmfx(formula = langjob_female_full_type2_fe, data=gi_female[gi_female$type2!="No answer",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) # average marginal effects with heteroscedasticity-robust White standard errors print(langjob_female_type2_fe_ame) # Percent correctly predicted values langjob_female_type2_fe_predict<- (table(true = gi_female[gi_female$type2!="No answer",]$langjob_d, pred = round(fitted(langjob_female_type2_fe)))[1,1]+table(true = gi_female[gi_female$type2!="No answer",]$langjob_d, pred = round(fitted(langjob_female_type2_fe)))[2,2] ) /nobs(langjob_female_type2_fe) *100 # 71,4 % correctly predicted values # McFadden's Pseudo R-squared langjob_female_type2_fe_baseline <- update(langjob_female_type2_fe, formula= langjob_d ~ 1) langjob_female_type2_fe_McFadden <- 1 - as.vector(logLik(langjob_female_type2_fe)/logLik(langjob_female_type2_fe_baseline)) langjob_female_type2_fe_McFadden # #### 2 b1d base langjob male #### gi_male<-gi[gi$female==0,] langjob_male_full_fe<-as.formula("langjob_d ~ factor(age)+children_d_na+part+occupationac+degree+risk+patient+english+country") #+mig_phase langjob_male_full_type_fe<-as.formula("langjob_d ~ type + factor(age)+children_d_na+part+occupationac+degree+risk+patient+english+country") #+mig_phase langjob_male_full_type2_fe<-as.formula("langjob_d ~ type2+factor(age)+children_d_na+part+occupationac+degree+risk+patient+english+country") #+mig_phase langjob_male_fe<-glm(formula = langjob_male_full_fe, family = binomial(link = "probit"), data = gi_male) langjob_male_fe_het <- print(coeftest(langjob_male_fe, vcovCL), digits=6) nobs(langjob_male_fe) # Marginal effects (Average marginal effects) langjob_male_fe_ame <- probitmfx(formula = langjob_male_full_fe, data=gi_male, atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) # average marginal effects with heteroscedasticity-robust White standard errors print(langjob_male_fe_ame) # Percent correctly predicted values langjob_male_fe_predict<- (table(true = gi_male$langjob_d, pred = round(fitted(langjob_male_fe)))[1,1]+table(true = gi_male$langjob_d, pred = round(fitted(langjob_male_fe)))[2,2] ) /nobs(langjob_male_fe) *100 # 71,4 % correctly predicted values # McFadden's Pseudo R-squared langjob_male_fe_baseline <- update(langjob_male_fe, formula= langjob_d ~ 1) langjob_male_fe_McFadden <- 1 - as.vector(logLik(langjob_male_fe)/logLik(langjob_male_fe_baseline)) langjob_male_fe_McFadden #### 2 b1e type male#### langjob_male_type_fe<-glm(formula = langjob_full_type_fe_g, family = binomial(link = "probit"), data = gi_male) langjob_male_type_fe_het <- print(coeftest(langjob_male_type_fe, vcovCL), digits=6) nobs(langjob_male_type_fe) # Marginal effects (Average marginal effects) langjob_male_type_fe_ame <- probitmfx(formula = langjob_male_full_type_fe, data=gi_male, atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) # average marginal effects with heteroscedasticity-robust White standard errors print(langjob_male_type_fe_ame) # Percent correctly predicted values langjob_male_type_fe_predict<- (table(true = gi_male$langjob_d, pred = round(fitted(langjob_male_type_fe)))[1,1]+table(true = gi_male$langjob_d, pred = round(fitted(langjob_male_type_fe)))[2,2] ) /nobs(langjob_male_type_fe) *100# 71,4 % correctly predicted values # McFadden's Pseudo R-squared langjob_male_type_fe_baseline <- update(langjob_male_type_fe, formula= langjob_d ~ 1) langjob_male_type_fe_McFadden <- 1 - as.vector(logLik(langjob_male_type_fe)/logLik(langjob_male_type_fe_baseline)) langjob_male_type_fe_McFadden #### 2 b1f type2 male #### langjob_male_type2_fe<-glm(formula = langjob_male_full_type2_fe, family = binomial(link = "probit"), data = gi_male[gi_male$type2!="No answer",]) langjob_male_type2_fe_het <- print(coeftest(langjob_male_type2_fe, vcovCL), digits=6) nobs(langjob_male_type2_fe) # Marginal effects (Average marginal effects) langjob_male_type2_fe_ame <- probitmfx(formula = langjob_male_full_type2_fe, data=gi_male[gi_male$type2!="No answer",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) # average marginal effects with heteroscedasticity-robust White standard errors print(langjob_male_type2_fe_ame) # Percent correctly predicted values langjob_male_type2_fe_predict<- (table(true = gi_male[gi_male$type2!="No answer",]$langjob_d, pred = round(fitted(langjob_male_type2_fe)))[1,1]+table(true = gi_male[gi_male$type2!="No answer",]$langjob_d, pred = round(fitted(langjob_male_type2_fe)))[2,2] ) /nobs(langjob_male_type2_fe) *100 # 71,4 % correctly predicted values # McFadden's Pseudo R-squared langjob_male_type2_fe_baseline <- update(langjob_male_type2_fe, formula= langjob_d ~ 1) langjob_male_type2_fe_McFadden <- 1 - as.vector(logLik(langjob_male_type2_fe)/logLik(langjob_male_type2_fe_baseline)) langjob_male_type2_fe_McFadden #### 2 b1g type no uni deg#### gi_nodeg<-gi[gi$degree==FALSE,] langjob_nodeg_full_fe<-as.formula("langjob_d ~ type + factor(age)+female+children_d_na+part+occupationac+risk+patient+english+country") #+mig_phase langjob_nodeg_type_fe<-glm(formula = langjob_nodeg_full_fe, family = binomial(link = "probit"), data = gi_nodeg) langjob_nodeg_type_fe_het <- print(coeftest(langjob_nodeg_type_fe, vcovCL), digits=6) nobs(langjob_nodeg_type_fe) # Marginal effects (Average marginal effects) langjob_nodeg_type_fe_ame <- probitmfx(formula = langjob_nodeg_full_fe, data=gi_nodeg, atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) # average marginal effects with heteroscedasticity-robust White standard errors print(langjob_nodeg_type_fe_ame) # Percent correctly predicted values langjob_nodeg_type_fe_predict<- (table(true = gi_nodeg$langjob_d, pred = round(fitted(langjob_nodeg_type_fe)))[1,1]+table(true = gi_nodeg$langjob_d, pred = round(fitted(langjob_nodeg_type_fe)))[2,2] ) /nobs(langjob_nodeg_type_fe) *100# 71,4 % correctly predicted values # McFadden's Pseudo R-squared langjob_nodeg_type_fe_baseline <- update(langjob_nodeg_type_fe, formula= langjob_d ~ 1) langjob_nodeg_type_fe_McFadden <- 1 - as.vector(logLik(langjob_nodeg_type_fe)/logLik(langjob_nodeg_type_fe_baseline)) langjob_nodeg_type_fe_McFadden #### 2 b1g type uni deg#### gi_deg<-gi[gi$degree==TRUE,] langjob_deg_full_fe<-as.formula("langjob_d ~ type + factor(age)+female+children_d_na+part+occupationac+risk+patient+english+country") #+mig_phase langjob_deg_type_fe<-glm(formula = langjob_deg_full_fe, family = binomial(link = "probit"), data = gi_deg) langjob_deg_type_fe_het <- print(coeftest(langjob_deg_type_fe, vcovCL), digits=6) nobs(langjob_deg_type_fe) # Marginal effects (Average marginal effects) langjob_deg_type_fe_ame <- probitmfx(formula = langjob_deg_full_fe, data=gi_deg, atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) # average marginal effects with heteroscedasticity-robust White standard errors print(langjob_deg_type_fe_ame) # Percent correctly predicted values langjob_deg_type_fe_predict<- (table(true = gi_deg$langjob_d, pred = round(fitted(langjob_deg_type_fe)))[1,1]+table(true = gi_deg$langjob_d, pred = round(fitted(langjob_deg_type_fe)))[2,2] ) /nobs(langjob_deg_type_fe) *100# 71,4 % correctly predicted values # McFadden's Pseudo R-squared langjob_deg_type_fe_baseline <- update(langjob_deg_type_fe, formula= langjob_d ~ 1) langjob_deg_type_fe_McFadden <- 1 - as.vector(logLik(langjob_deg_type_fe)/logLik(langjob_deg_type_fe_baseline)) langjob_deg_type_fe_McFadden #### 2 b2a base langjob young #### gi_young<-gi[gi$age==TRUE,] langjob_young_full_fe<-as.formula("langjob_d ~ female+children_d_na+part+occupationac+degree+risk+patient+english+country") #+mig_phase langjob_young_full_type_fe<-as.formula("langjob_d ~ type + female+children_d_na+part+occupationac+degree+risk+patient+english+country") #+mig_phase langjob_young_full_type2_fe<-as.formula("langjob_d ~ type2+female++children_d_na+part+occupationac+degree+risk+patient+english+country") #+mig_phase langjob_young_fe<-glm(formula = langjob_young_full_fe, family = binomial(link = "probit"), data = gi_young) langjob_young_fe_het <- print(coeftest(langjob_young_fe, vcovCL), digits=6) nobs(langjob_young_fe) # Marginal effects (Average marginal effects) langjob_young_fe_ame <- probitmfx(formula = langjob_young_full_fe, data=gi_young, atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) # average marginal effects with heteroscedasticity-robust White standard errors print(langjob_young_fe_ame) # Percent correctly predicted values langjob_young_fe_predict<- (table(true = gi_young$langjob_d, pred = round(fitted(langjob_young_fe)))[1,1]+table(true = gi_young$langjob_d, pred = round(fitted(langjob_young_fe)))[2,2] ) /nobs(langjob_young_fe) *100 # 71,4 % correctly predicted values # McFadden's Pseudo R-squared langjob_young_fe_baseline <- update(langjob_young_fe, formula= langjob_d ~ 1) langjob_young_fe_McFadden <- 1 - as.vector(logLik(langjob_young_fe)/logLik(langjob_young_fe_baseline)) langjob_young_fe_McFadden #### 2 b2b type young#### langjob_young_type_fe<-glm(formula = langjob_young_full_type_fe, family = binomial(link = "probit"), data = gi_young) langjob_young_type_fe_het <- print(coeftest(langjob_young_type_fe, vcovCL), digits=6) nobs(langjob_young_type_fe) # Marginal effects (Average marginal effects) langjob_young_type_fe_ame <- probitmfx(formula = langjob_young_full_type_fe, data=gi_young, atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) # average marginal effects with heteroscedasticity-robust White standard errors print(langjob_young_type_fe_ame) # Percent correctly predicted values langjob_young_type_fe_predict<- (table(true = gi_young$langjob_d, pred = round(fitted(langjob_young_type_fe)))[1,1]+table(true = gi_young$langjob_d, pred = round(fitted(langjob_young_type_fe)))[2,2] ) /nobs(langjob_young_type_fe) *100# 71,4 % correctly predicted values # McFadden's Pseudo R-squared langjob_young_type_fe_baseline <- update(langjob_young_type_fe, formula= langjob_d ~ 1) langjob_young_type_fe_McFadden <- 1 - as.vector(logLik(langjob_young_type_fe)/logLik(langjob_young_type_fe_baseline)) langjob_young_type_fe_McFadden #### 2 b2d base langjob old #### gi_old<-gi[gi$age==FALSE,] langjob_old_full_fe<-as.formula("langjob_d ~ female+children_d_na+part+occupationac+degree+risk+patient+english+country") #+mig_phase langjob_old_full_type_fe<-as.formula("langjob_d ~ type + female+children_d_na+part+occupationac+degree+risk+patient+english+country") #+mig_phase langjob_old_full_type2_fe<-as.formula("langjob_d ~ type2+female++children_d_na+part+occupationac+degree+risk+patient+english+country") #+mig_phase langjob_old_fe<-glm(formula = langjob_old_full_fe, family = binomial(link = "probit"), data = gi_old) langjob_old_fe_het <- print(coeftest(langjob_old_fe, vcovCL), digits=6) nobs(langjob_old_fe) # Marginal effects (Average marginal effects) langjob_old_fe_ame <- probitmfx(formula = langjob_old_full_fe, data=gi_old, atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) # average marginal effects with heteroscedasticity-robust White standard errors print(langjob_old_fe_ame) # Percent correctly predicted values langjob_old_fe_predict<- (table(true = gi_old$langjob_d, pred = round(fitted(langjob_old_fe)))[1,1]+table(true = gi_old$langjob_d, pred = round(fitted(langjob_old_fe)))[2,2] ) /nobs(langjob_old_fe) *100 # 71,4 % correctly predicted values # McFadden's Pseudo R-squared langjob_old_fe_baseline <- update(langjob_old_fe, formula= langjob_d ~ 1) langjob_old_fe_McFadden <- 1 - as.vector(logLik(langjob_old_fe)/logLik(langjob_old_fe_baseline)) langjob_old_fe_McFadden #### 2 b2e type old#### langjob_old_type_fe<-glm(formula = langjob_old_full_type_fe, family = binomial(link = "probit"), data = gi_old) langjob_old_type_fe_het <- print(coeftest(langjob_old_type_fe, vcovCL), digits=6) nobs(langjob_old_type_fe) # Marginal effects (Average marginal effects) langjob_old_type_fe_ame <- probitmfx(formula = langjob_old_full_type_fe, data=gi_old, atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) # average marginal effects with heteroscedasticity-robust White standard errors print(langjob_old_type_fe_ame) # Percent correctly predicted values langjob_old_type_fe_predict<- (table(true = gi_old$langjob_d, pred = round(fitted(langjob_old_type_fe)))[1,1]+table(true = gi_old$langjob_d, pred = round(fitted(langjob_old_type_fe)))[2,2] ) /nobs(langjob_old_type_fe) *100# 71,4 % correctly predicted values # McFadden's Pseudo R-squared langjob_old_type_fe_baseline <- update(langjob_old_type_fe, formula= langjob_d ~ 1) langjob_old_type_fe_McFadden <- 1 - as.vector(logLik(langjob_old_type_fe)/logLik(langjob_old_type_fe_baseline)) langjob_old_type_fe_McFadden #### 2 b4 Output subsamples age gender #### stargazer(langjob_female_type_fe_ame$fit, langjob_male_type_fe_ame$fit, langjob_young_type_fe_ame$fit, langjob_old_type_fe_ame$fit, langjob_deg_type_fe_ame$fit, langjob_nodeg_type_fe_ame$fit, coef = list(langjob_female_type_fe_ame$mfxest[,1], langjob_male_type_fe_ame$mfxest[,1], langjob_young_type_fe_ame$mfxest[,1], langjob_old_type_fe_ame$mfxest[,1], langjob_nodeg_type_fe_ame$mfxest[,1], langjob_deg_type_fe_ame$mfxest[,1] ), se = list(langjob_female_type_fe_ame$mfxest[,2], langjob_male_type_fe_ame$mfxest[,2], langjob_young_type_fe_ame$mfxest[,2], langjob_old_type_fe_ame$mfxest[,2], langjob_nodeg_type_fe_ame$mfxest[,2], langjob_deg_type_fe_ame$mfxest[,2] ), header=TRUE, p.auto=FALSE, style="default", nobs=TRUE, align=TRUE, no.space=TRUE, single.row=FALSE, model.names = FALSE, dep.var.labels = "Professional use of German", column.labels = c("female", "male","age: <35", "age: >=35", "no uni. deg.", "uni. deg." ), covariate.labels= c( "Investment", "Age: under 35 years", "Gender: female", "Gender: n/a", "Children", "Partner (native German)", "Partner (other native)", "Occ.: high appl./low comm. skills", "Occ.: high appl./high comm. skills", "Occ.: in education", "Occ.: other occ./no answer", "University degree", "Risk attitude", "Patience", "English speaker"), omit = c("Constant", "country", "empstat4"), font.size = "small", add.lines = list(c("Country FE", "Yes", "Yes", "Yes", "Yes", "Yes", "Yes", "Yes"), c("McFadden Pseudo R2", round(langjob_female_type_fe_McFadden , digits=2) , round(langjob_male_type_fe_McFadden , digits=2) , round(langjob_young_type_fe_McFadden , digits=2) , round(langjob_old_type_fe_McFadden , digits=2), round(langjob_nodeg_type_fe_McFadden , digits=2), round(langjob_deg_type_fe_McFadden , digits=2)), c("% correctly predicted", round(langjob_female_type_fe_predict, digits=1) , round(langjob_male_type_fe_predict, digits=1), round(langjob_young_type_fe_predict, digits=1) , round(langjob_old_type_fe_predict, digits=1), round(langjob_nodeg_type_fe_predict, digits=1) , round(langjob_deg_type_fe_predict, digits=1))), omit.stat = c("aic", "ll"), omit.table.layout = "n") #### 2 b3 langjobsubsamble #### #### 2 b3a langjob inv #### gi_inv<-gi[gi$type==TRUE,] langjob_full_coninv_type2_fe<-as.formula("langjob_d ~ type2+factor(age)+female+children_d_na+part+occupationac+degree+risk+patient+english+country") #+mig_phase langjob_inv_type2_fe<-glm(formula = langjob_full_coninv_type2_fe, family = binomial(link = "probit"), data = gi_inv[gi_inv$type2!="No answer",]) langjob_inv_type2_fe_het <- print(coeftest(langjob_inv_type2_fe, vcovCL), digits=6) nobs(langjob_inv_type2_fe) # Marginal effects (Average marginal effects) langjob_inv_type2_fe_ame <- probitmfx(formula = langjob_full_coninv_type2_fe, data=gi_inv[gi_inv$type2!="No answer",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) # average marginal effects with heteroscedasticity-robust White standard errors print(langjob_inv_type2_fe_ame) # Percent correctly predicted values langjob_inv_type2_fe_predict<- (table(true = gi_inv[gi_inv$type2!="No answer",]$langjob_d, pred = round(fitted(langjob_inv_type2_fe)))[1,1]+table(true = gi_inv[gi_inv$type2!="No answer",]$langjob_d, pred = round(fitted(langjob_inv_type2_fe)))[2,2] ) /nobs(langjob_inv_type2_fe) *100 # 71,4 % correctly predicted values # McFadden's Pseudo R-squared langjob_inv_type2_fe_baseline <- update(langjob_inv_type2_fe, formula= langjob_d ~ 1) langjob_inv_type2_fe_McFadden <- 1 - as.vector(logLik(langjob_inv_type2_fe)/logLik(langjob_inv_type2_fe_baseline)) langjob_inv_type2_fe_McFadden #### 2 b3b langjob con #### gi_con<-gi[gi$type==FALSE,] langjob_full_coninv_type2_fe<-as.formula("langjob_d ~ type2+factor(age)+female+children_d_na+part+occupationac+degree+risk+patient+english+country") #+mig_phase langjob_con_type2_fe<-glm(formula = langjob_full_coninv_type2_fe, family = binomial(link = "probit"), data = gi_con[gi_con$type2!="No answer",]) langjob_con_type2_fe_het <- print(coeftest(langjob_con_type2_fe, vcovCL), digits=6) nobs(langjob_con_type2_fe) # Marginal effects (Average marginal effects) langjob_con_type2_fe_ame <- probitmfx(formula = langjob_full_coninv_type2_fe, data=gi_con[gi_con$type2!="No answer",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) # average marginal effects with heteroscedasticity-robust White standard errors print(langjob_con_type2_fe_ame) # Percent correctly predicted values langjob_con_type2_fe_predict<- (table(true = gi_con[gi_con$type2!="No answer",]$langjob_d, pred = round(fitted(langjob_con_type2_fe)))[1,1]+table(true = gi_con[gi_con$type2!="No answer",]$langjob_d, pred = round(fitted(langjob_con_type2_fe)))[2,2] ) /nobs(langjob_con_type2_fe) *100 # 71,4 % correctly predicted values # McFadden's Pseudo R-squared langjob_con_type2_fe_baseline <- update(langjob_con_type2_fe, formula= langjob_d ~ 1) langjob_con_type2_fe_McFadden <- 1 - as.vector(logLik(langjob_con_type2_fe)/logLik(langjob_con_type2_fe_baseline)) langjob_con_type2_fe_McFadden #### 2 b3c langjob inv #### gi_inv<-gi[gi$type==TRUE,] langjob_full_coninv_type_fe<-as.formula("langjob_d ~ factor(age)+female+children_d_na+part+occupationac+degree+risk+patient+english+country") #+mig_phase langjob_inv_type_fe<-glm(formula = langjob_full_coninv_type_fe, family = binomial(link = "probit"), data = gi_inv) langjob_inv_type_fe_het <- print(coeftest(langjob_inv_type_fe, vcovCL), digits=6) nobs(langjob_inv_type_fe) # Marginal effects (Average marginal effects) langjob_inv_type_fe_ame <- probitmfx(formula = langjob_full_coninv_type_fe, data=gi_inv, atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) # average marginal effects with heteroscedasticity-robust White standard errors print(langjob_inv_type_fe_ame) # Percent correctly predicted values langjob_inv_type_fe_predict<- (table(true = gi_inv$langjob_d, pred = round(fitted(langjob_inv_type_fe)))[1,1]+table(true = gi_inv$langjob_d, pred = round(fitted(langjob_inv_type_fe)))[2,2] ) /nobs(langjob_inv_type_fe) *100 # 71,4 % correctly predicted values # McFadden's Pseudo R-squared langjob_inv_type_fe_baseline <- update(langjob_inv_type_fe, formula= langjob_d ~ 1) langjob_inv_type_fe_McFadden <- 1 - as.vector(logLik(langjob_inv_type_fe)/logLik(langjob_inv_type_fe_baseline)) langjob_inv_type_fe_McFadden #### 2 b3d langjob con #### gi_con<-gi[gi$type==FALSE,] langjob_full_coninv_type_fe<-as.formula("langjob_d ~ factor(age)+female+children_d_na+part+occupationac+degree+risk+patient+english+country") #+mig_phase langjob_con_type_fe<-glm(formula = langjob_full_coninv_type_fe, family = binomial(link = "probit"), data = gi_con) langjob_con_type_fe_het <- print(coeftest(langjob_con_type_fe, vcovCL), digits=6) nobs(langjob_con_type_fe) # Marginal effects (Average marginal effects) langjob_con_type_fe_ame <- probitmfx(formula = langjob_full_coninv_type_fe, data=gi_con, atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) # average marginal effects with heteroscedasticity-robust White standard errors print(langjob_con_type_fe_ame) # Percent correctly predicted values langjob_con_type_fe_predict<- (table(true = gi_con$langjob_d, pred = round(fitted(langjob_con_type_fe)))[1,1]+table(true = gi_con$langjob_d, pred = round(fitted(langjob_con_type_fe)))[2,2] ) /nobs(langjob_con_type_fe) *100 # 71,4 % correctly predicted values # McFadden's Pseudo R-squared langjob_con_type_fe_baseline <- update(langjob_con_type_fe, formula= langjob_d ~ 1) langjob_con_type_fe_McFadden <- 1 - as.vector(logLik(langjob_con_type_fe)/logLik(langjob_con_type_fe_baseline)) langjob_con_type_fe_McFadden #### 2 b4 langjobsubsamble +reason_om #### #### 2 b3a langjob inv #### gi_inv<-gi[gi$type==TRUE,] langjob_full_coninv_type2_fe<-as.formula("langjob_d ~ type2+factor(age)+female+children_d_na+part+occupationac+degree+risk+patient+english+reason_om+country") #+mig_phase langjob_inv_type2_fe2<-glm(formula = langjob_full_coninv_type2_fe, family = binomial(link = "probit"), data = gi_inv[gi_inv$type2!="No answer",]) langjob_inv_type2_fe2_het <- print(coeftest(langjob_inv_type2_fe2, vcovCL), digits=6) nobs(langjob_inv_type2_fe2) # Marginal effects (Average marginal effects) langjob_inv_type2_fe2_ame <- probitmfx(formula = langjob_full_coninv_type2_fe, data=gi_inv[gi_inv$type2!="No answer",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) # average marginal effects with heteroscedasticity-robust White standard errors print(langjob_inv_type2_fe2_ame) # Percent correctly predicted values langjob_inv_type2_fe2_predict<- (table(true = gi_inv[gi_inv$type2!="No answer",]$langjob_d, pred = round(fitted(langjob_inv_type2_fe2)))[1,1]+table(true = gi_inv[gi_inv$type2!="No answer",]$langjob_d, pred = round(fitted(langjob_inv_type2_fe2)))[2,2] ) /nobs(langjob_inv_type2_fe2) *100 # 71,4 % correctly predicted values # McFadden's Pseudo R-squared langjob_inv_type2_fe2_baseline <- update(langjob_inv_type2_fe2, formula= langjob_d ~ 1) langjob_inv_type2_fe2_McFadden <- 1 - as.vector(logLik(langjob_inv_type2_fe2)/logLik(langjob_inv_type2_fe2_baseline)) langjob_inv_type2_fe2_McFadden #### 2 b3b langjob con #### gi_con<-gi[gi$type==FALSE,] langjob_full_coninv_type2_fe2<-as.formula("langjob_d ~ type2+factor(age)+female+children_d_na+part+occupationac+degree+risk+patient+english+reason_om+country") #+mig_phase langjob_con_type2_fe2<-glm(formula = langjob_full_coninv_type2_fe, family = binomial(link = "probit"), data = gi_con[gi_con$type2!="No answer",]) langjob_con_type2_fe2_het <- print(coeftest(langjob_con_type2_fe2, vcovCL), digits=6) nobs(langjob_con_type2_fe2) # Marginal effects (Average marginal effects) langjob_con_type2_fe2_ame <- probitmfx(formula = langjob_full_coninv_type2_fe, data=gi_con[gi_con$type2!="No answer",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) # average marginal effects with heteroscedasticity-robust White standard errors print(langjob_con_type2_fe2_ame) # Percent correctly predicted values langjob_con_type2_fe2_predict<- (table(true = gi_con[gi_con$type2!="No answer",]$langjob_d, pred = round(fitted(langjob_con_type2_fe2)))[1,1]+table(true = gi_con[gi_con$type2!="No answer",]$langjob_d, pred = round(fitted(langjob_con_type2_fe2)))[2,2] ) /nobs(langjob_con_type2_fe2) *100 # 71,4 % correctly predicted values # McFadden's Pseudo R-squared langjob_con_type2_fe2_baseline <- update(langjob_con_type2_fe2, formula= langjob_d ~ 1) langjob_con_type2_fe2_McFadden <- 1 - as.vector(logLik(langjob_con_type2_fe2)/logLik(langjob_con_type2_fe2_baseline)) langjob_con_type2_fe2_McFadden #### 2 b3c langjob inv #### gi_inv<-gi[gi$type==TRUE,] langjob_full_coninv_type_fe<-as.formula("langjob_d ~ factor(age)+female+children_d_na+part+occupationac+degree+risk+patient+english+reason_om+country") #+mig_phase langjob_inv_type_fe2<-glm(formula = langjob_full_coninv_type_fe, family = binomial(link = "probit"), data = gi_inv) langjob_inv_type_fe2_het <- print(coeftest(langjob_inv_type_fe2, vcovCL), digits=6) nobs(langjob_inv_type_fe2) # Marginal effects (Average marginal effects) langjob_inv_type_fe2_ame <- probitmfx(formula = langjob_full_coninv_type_fe, data=gi_inv, atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) # average marginal effects with heteroscedasticity-robust White standard errors print(langjob_inv_type_fe2_ame) # Percent correctly predicted values langjob_inv_type_fe2_predict<- (table(true = gi_inv$langjob_d, pred = round(fitted(langjob_inv_type_fe2)))[1,1]+table(true = gi_inv$langjob_d, pred = round(fitted(langjob_inv_type_fe2)))[2,2] ) /nobs(langjob_inv_type_fe2) *100 # 71,4 % correctly predicted values # McFadden's Pseudo R-squared langjob_inv_type_fe2_baseline <- update(langjob_inv_type_fe2, formula= langjob_d ~ 1) langjob_inv_type_fe2_McFadden <- 1 - as.vector(logLik(langjob_inv_type_fe2)/logLik(langjob_inv_type_fe2_baseline)) langjob_inv_type_fe2_McFadden #### 2 b3d langjob con #### gi_con<-gi[gi$type==FALSE,] langjob_full_coninv_type_fe<-as.formula("langjob_d ~ factor(age)+female+children_d_na+part+occupationac+degree+risk+patient+english+reason_om+country") #+mig_phase langjob_con_type_fe2<-glm(formula = langjob_full_coninv_type_fe, family = binomial(link = "probit"), data = gi_con) langjob_con_type_fe2_het <- print(coeftest(langjob_con_type_fe2, vcovCL), digits=6) nobs(langjob_con_type_fe2) # Marginal effects (Average marginal effects) langjob_con_type_fe2_ame <- probitmfx(formula = langjob_full_coninv_type_fe, data=gi_con, atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) # average marginal effects with heteroscedasticity-robust White standard errors print(langjob_con_type_fe2_ame) # Percent correctly predicted values langjob_con_type_fe2_predict<- (table(true = gi_con$langjob_d, pred = round(fitted(langjob_con_type_fe2)))[1,1]+table(true = gi_con$langjob_d, pred = round(fitted(langjob_con_type_fe2)))[2,2] ) /nobs(langjob_con_type_fe2) *100 # 71,4 % correctly predicted values # McFadden's Pseudo R-squared langjob_con_type_fe2_baseline <- update(langjob_con_type_fe2, formula= langjob_d ~ 1) langjob_con_type_fe2_McFadden <- 1 - as.vector(logLik(langjob_con_type_fe2)/logLik(langjob_con_type_fe2_baseline)) langjob_con_type_fe2_McFadden #### Output ------ stargazer( langjob_con_type_fe_ame$fit, langjob_inv_type_fe_ame$fit, langjob_con_type_fe2_ame$fit, langjob_inv_type_fe2_ame$fit, coef = list(langjob_con_type_fe_ame$mfxest[,1], langjob_inv_type_fe_ame$mfxest[,1], langjob_con_type_fe2_ame$mfxest[,1], langjob_inv_type_fe2_ame$mfxest[,1]), se = list(langjob_con_type_fe_ame$mfxest[,2], langjob_inv_type_fe_ame$mfxest[,2], langjob_con_type_fe2_ame$mfxest[,2], langjob_inv_type_fe2_ame$mfxest[,2]), dep.var.labels = "Professional use of German", column.labels = c("consumpt. good", "hum. cap. invest.", "consumpt. good", "hum. cap. invest."), covariate.labels= c( "Age: under 35 years", "Gender: female", "Gender: n/a", "Children", "Partner (native German)", "Partner (other native)", "Occ.: high appl./low comm. skills", "Occ.: high appl./high comm. skills", "Occ.: in education", "Occ.: other occ./no answer", "University degree", "Risk attitude", "Patience", "English speaker", "Other investm. reason"), header=TRUE, p.auto=FALSE, style="default", nobs=TRUE, align=TRUE, no.space=TRUE, single.row=FALSE, model.names = FALSE, omit = c("Constant", "country", "empstat4"), font.size = "small", add.lines = list(c("Country FE", "Yes", "Yes", "Yes", "Yes" ), c("McFadden Pseudo R2", round(langjob_inv_type_fe_McFadden , digits=2) , round(langjob_con_type_fe_McFadden , digits=2) , round(langjob_inv_type_fe2_McFadden , digits=2) , round(langjob_con_type_fe2_McFadden , digits=2) ), c("% correctly predicted", round(langjob_inv_type_fe_predict , digits=1) , round(langjob_con_type_fe_predict , digits=1), round(langjob_inv_type_fe2_predict , digits=1) , round(langjob_con_type_fe2_predict , digits=1))), omit.stat = c("aic", "ll"), omit.table.layout = "n") #### 2 c Country-wise #### #### 2 c1 Countries #### TOT<-probitmfx(formula = langjob_full_type2_fe, data=gi, atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) BIH<-probitmfx(formula = langjob_full_type2, data=gi[gi$country=="BIH",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) CZE<-probitmfx(formula = langjob_full_type2, data=gi[gi$country=="CZE",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) ESP<-probitmfx(formula = langjob_full_type2, data=gi[gi$country=="ESP",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) GBR<-probitmfx(formula = langjob_full_type2, data=gi[gi$country=="GBR",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) ITA<-probitmfx(formula = langjob_full_type2, data=gi[gi$country=="ITA",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) JPN<-probitmfx(formula = langjob_full_type2, data=gi[gi$country=="JPN",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) KOR<-probitmfx(formula = langjob_full_type2, data=gi[gi$country=="KOR",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) NLD<-probitmfx(formula = langjob_full_type2, data=gi[gi$country=="NLD",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) POL<-probitmfx(formula = langjob_full_type2, data=gi[gi$country=="POL",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) ROU<-probitmfx(formula = langjob_full_type2, data=gi[gi$country=="ROU",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) IND<-probitmfx(formula = langjob_full_type2, data=gi[gi$country=="IND",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) IDN<-probitmfx(formula = langjob_full_type2, data=gi[gi$country=="IDN",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) UKR<-probitmfx(formula = langjob_full_type2, data=gi[gi$country=="UKR",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) MEX<-probitmfx(formula = langjob_full_type2, data=gi[gi$country=="MEX",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) stargazer(TOT$fit, BIH$fit, CZE$fit, ESP$fit, GBR$fit, ITA$fit, JPN$fit, KOR$fit, NLD$fit, POL$fit, ROU$fit, IND$fit, IDN$fit, UKR$fit, MEX$fit, coef = list(TOT$mfxest[,1], BIH$mfxest[,1], CZE$mfxest[,1], ESP$mfxest[,1], GBR$mfxest[,1], ITA$mfxest[,1], JPN$mfxest[,1], KOR$mfxest[,1], NLD$mfxest[,1], POL$mfxest[,1], ROU$mfxest[,1], IND$mfxest[,1], IDN$mfxest[,1], UKR$mfxest[,1], MEX$mfxest[,1]), se = list(TOT$mfxest[,2], BIH$mfxest[,2], CZE$mfxest[,2], ESP$mfxest[,2], GBR$mfxest[,2], ITA$mfxest[,2], JPN$mfxest[,2], KOR$mfxest[,2], NLD$mfxest[,2], POL$mfxest[,2], ROU$mfxest[,2], IND$mfxest[,2], IDN$mfxest[,2], UKR$mfxest[,2], MEX$mfxest[,2]), column.labels = c("TOT", "BIH", "CZE", "ESP", "GBR", "ITA", "JPN", "KOR", "NLD", "POL", "ROU", "IND", "IDN", "UKR", "MEX"), header=FALSE, style="default", dep.var.labels = "Professional use of German", covariate.labels= c( "Domestic labour market", "Educational and labour migration", "Personal reasons", "Cultural interest", "Age: under 35 years", "Gender: female", "Gender: n/a", "Children", "Partner (native German)", "Partner (other native)", "Occ.: in education", "Other occ./no answer", "University degree", "Risk attitude", "Patience", "Intern. applic. of education" ), nobs=TRUE, align=TRUE, no.space=TRUE, single.row=TRUE, model.names = FALSE, omit = c("Constant", "country", "empstat4"), font.size = "small", add.lines = list(c("Country FE", "Yes", "No", "No","No","No","No","No","No","No","No","No","No","No","No","No")), notes= "Average marginal effects. Heteroscedasticity robust White standard errors in parentheses.", omit.stat = c("aic", "ll")) TOT_type<-tidy(TOT) %>% filter(term != "(Intercept)") %>% mutate(model = "Total") BIH_type<-tidy(BIH) %>% filter(term != "(Intercept)") %>% mutate(model = "BIH (n=126)") CZE_type<-tidy(CZE) %>% filter(term != "(Intercept)") %>% mutate(model = "CZE (n=311)") ESP_type<-tidy(ESP) %>% filter(term != "(Intercept)") %>% mutate(model = "ESP (n=467)") GBR_type<-tidy(GBR) %>% filter(term != "(Intercept)") %>% mutate(model = "GBR (n=339)") ITA_type<-tidy(ITA) %>% filter(term != "(Intercept)") %>% mutate(model = "ITA (n=218)") JPN_type<-tidy(JPN) %>% filter(term != "(Intercept)") %>% mutate(model = "JPN (n=196)") KOR_type<-tidy(KOR) %>% filter(term != "(Intercept)") %>% mutate(model = "KOR (n=215)") NLD_type<-tidy(NLD) %>% filter(term != "(Intercept)") %>% mutate(model = "NLD (n=68)") POL_type<-tidy(POL) %>% filter(term != "(Intercept)") %>% mutate(model = "POL (n=138)") ROU_type<-tidy(ROU) %>% filter(term != "(Intercept)") %>% mutate(model = "ROU (n=223)") IND_type<-tidy(IND) %>% filter(term != "(Intercept)") %>% mutate(model = "IND (n=406)") IDN_type<-tidy(IDN) %>% filter(term != "(Intercept)") %>% mutate(model = "IDN (n=539)") UKR_type<-tidy(UKR) %>% filter(term != "(Intercept)") %>% mutate(model = "UKR (n=512)") MEX_type<-tidy(MEX) %>% filter(term != "(Intercept)") %>% mutate(model = "MEX (n=316)") gi %>% count(country) %>% arrange(-n) country_l_type<-rbind(TOT_type, ESP_type, GBR_type, CZE_type, ITA_type, POL_type, ROU_type, NLD_type, BIH_type, UKR_type, MEX_type, IND_type, IDN_type, KOR_type, JPN_type) country_l_type<-country_l_type %>% mutate(sign=ifelse(model=="Total" & p.value<0.05, TRUE, FALSE)) %>% group_by(term) %>% mutate(sign_5=max(sign), sign=ifelse(sign_5==0, "P>0.05 (aggr. model)", "P<0.05 (aggr. model)"), sign=as.factor(sign)) %>% filter(model!="Total") library(dotwhisker) library(ggsci) #### 2 c2a2 Country groups #### gi<-gi[gi$type2!="No answer",] eu_sw<-probitmfx(formula = langjob_full_type2_fe, data=gi[gi$country=="ESP" | gi$country=="ITA"| gi$country=="GBR" | gi$country=="NLD" ,], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) eu_e<-probitmfx(formula = langjob_full_type2_fe, data=gi[gi$country=="CZE" | gi$country=="POL" | gi$country=="ROU",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) eur<- probitmfx(formula = langjob_full_type2_fe, data=gi[gi$country=="UKR" | gi$country=="BIH",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) noneu_d<-probitmfx(formula = langjob_full_type2_fe, data=gi[gi$country=="JPN" | gi$country=="KOR",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) noneu_ld<-probitmfx(formula = langjob_full_type2_fe, data=gi[gi$country=="IDN" | gi$country=="IND" | gi$country=="MEX",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) eu<-probitmfx(formula = langjob_full_type2_fe, data=gi[gi$country=="ESP" | gi$country=="ITA"|gi$country=="NLD" | gi$country=="GBR"|gi$country=="CZE" | gi$country=="POL" | gi$country=="ROU",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) noneu<- probitmfx(formula = langjob_full_type2_fe, data=gi[gi$country=="MEX" | gi$country=="IDN" | gi$country=="IND"|gi$country=="JPN" | gi$country=="KOR"|gi$country=="UKR" | gi$country=="BIH",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) stargazer(eu_sw$fit, eu_e$fit, eur$fit, noneu_d$fit, noneu_ld$fit, eu$fit, noneu$fit, coef = list(eu_sw$mfxest[,1], eu_e$mfxest[,1], eur$mfxest[,1], noneu_d$mfxest[,1], noneu_ld$mfxest[,1], eu$mfxest[,1], noneu$mfxest[,1]), se = list(eu_sw$mfxest[,2], eu_e$mfxest[,2], eur$mfxest[,2], noneu_d$mfxest[,2], noneu_ld$mfxest[,2], eu$mfxest[,2], noneu$mfxest[,2]), column.labels = c("ESP, ITA, GBR, NLD", "CZE, POL, ROU", "UKR, BIH", "JPN, KOR", "IND, IDN, MEX", "EU", "Non-EU"), header=FALSE, style="default", dep.var.labels = "Professional use of German", covariate.labels= c( "Domestic labour market", "Educ. and labour migration", "Personal reasons", "Cultural interest", "Age: under 35 years", "Gender: female", "Gender: n/a", "Children", "Partner (native German)", "Partner (other native)", "Occ.: in education", "Other occ./no answer", "University degree", "Risk attitude", "Patience", "Intern. applic. of education" ), nobs=TRUE, align=FALSE, no.space=TRUE, single.row=FALSE, model.names = FALSE, omit = c("Constant", "country", "empstat4"), font.size = "small", add.lines = list(c("Country FE", "Yes", "Yes","Yes","Yes","Yes","Yes","Yes")), omit.stat = c("aic", "ll"), omit.table.layout = "n") eu_sw_type<-tidy(eu_sw) %>% filter(term != "(Intercept)") %>% mutate(model = "ESP, ITA, GBR, NLD (n=1092)") nrow(gi[gi$country=="ESP" | gi$country=="ITA"| gi$country=="GBR" | gi$country=="NLD" ,]) eu_e_type<-tidy(eu_e) %>% filter(term != "(Intercept)") %>% mutate(model = "CZE, POL, ROU (n=682)") nrow(gi[gi$country=="CZE" | gi$country=="POL" | gi$country=="ROU",]) eur_type<-tidy(eur) %>% filter(term != "(Intercept)") %>% mutate(model = "UKR, BIH (n=658)") nrow(gi[gi$country=="UKR" | gi$country=="BIH",]) noneu_d_type<-tidy(noneu_d) %>% filter(term != "(Intercept)") %>% mutate(model = "JPN, KOR (n=455)") nrow(gi[gi$country=="JPN" | gi$country=="KOR",]) noneu_ld_type<-tidy(noneu_ld) %>% filter(term != "(Intercept)") %>% mutate(model = "IND, IDN, MEX (n=857)") nrow(gi[gi$country=="IDN" | gi$country=="IN" | gi$country=="MEX",]) eu_type<-tidy(eu) %>% filter(term != "(Intercept)") %>% mutate(model = "EU (n=1774)") nrow(gi[gi$country=="ESP" | gi$country=="ITA"|gi$country=="NLD" | gi$country=="GBR"|gi$country=="CZE" | gi$country=="POL" | gi$country=="ROU",]) noneu_type<-tidy(noneu) %>% filter(term != "(Intercept)") %>% mutate(model = "Non-EU (n=2376)") nrow(gi[gi$country=="MEX" | gi$country=="IDN" | gi$country=="IND"|gi$country=="JPN" | gi$country=="KOR"|gi$country=="UKR" | gi$country=="BIH",]) countrygr_l_type<-rbind(TOT_type, eu_sw_type, eu_e_type, eur_type, noneu_d_type, noneu_ld_type, eu_type, noneu_type) countrygr_l_type<-countrygr_l_type %>% mutate(sign=ifelse(model=="Total" & p.value<0.05, TRUE, FALSE)) %>% group_by(term) %>% mutate(sign_5=max(sign), sign=ifelse(sign_5==0, "P>0.05 (aggr. model)", "P<0.05 (aggr. model)"), sign=as.factor(sign)) %>% filter(model!="Total") table(countrygr_l_type$term) #### 2 c2a2 Country characteristics #### gi<-gi[gi$type2!="No answer",] tot<-probitmfx(formula = langjob_full_type_fe, data=gi, atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) germ<-probitmfx(formula = langjob_full_type_fe, data=gi[gi$country=="NLD" | gi$country=="GBR" | gi$country=="IND",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) ngerm<-probitmfx(formula = langjob_full_type_fe, data=gi[gi$country=="ESP" | gi$country=="ITA" | gi$country=="CZE" |gi$country=="ROU" | gi$country=="BIH" | gi$country=="MEX"| gi$country=="IND"| gi$country=="UKR" | gi$country=="JPN" | gi$country=="KOR" | gi$country=="POL", ], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) hi<-probitmfx(formula = langjob_full_type_fe, data=gi[gi$country=="ESP" | gi$country=="ITA"| gi$country=="GBR" | gi$country=="NLD" | gi$country=="CZE" | gi$country=="POL" | gi$country=="JPN"| gi$country=="KOR" ,], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) umi<-probitmfx(formula = langjob_full_type_fe, data=gi[gi$country=="ROU" | gi$country=="BIH" | gi$country=="MEX",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) lmi<-probitmfx(formula = langjob_full_type_fe, data=gi[gi$country=="IDN" | gi$country=="IND"| gi$country=="UKR" ,], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) eur<- probitmfx(formula = langjob_full_type_fe, data=gi[gi$country=="UKR" | gi$country=="BIH",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) eu<-probitmfx(formula = langjob_full_type_fe, data=gi[gi$country=="ESP" | gi$country=="ITA"|gi$country=="NLD" | gi$country=="GBR"|gi$country=="CZE" | gi$country=="POL" | gi$country=="ROU",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) noneur<- probitmfx(formula = langjob_full_type_fe, data=gi[gi$country=="MEX" | gi$country=="IDN" | gi$country=="IND"|gi$country=="JPN" | gi$country=="KOR",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) cultclose<- probitmfx(formula = langjob_full_type_fe, data=gi[gi$country=="JPN" | gi$country=="BIH" | gi$country=="CZE"|gi$country=="IDN" | gi$country=="UKR"|gi$country=="KOR" | gi$country=="ITA",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) cultdist<- probitmfx(formula = langjob_full_type_fe, data=gi[gi$country=="MEX" | gi$country=="IND" | gi$country=="ESP"|gi$country=="NLD" | gi$country=="ROU"|gi$country=="POL" | gi$country=="GBR",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) stargazer(germ$fit, ngerm$fit, umi$fit, lmi$fit, hi$fit, eu$fit, eur$fit, noneur$fit, cultclose$fit, cultdist$fit, coef = list(germ$mfxest[,1], ngerm$mfxest[,1], lmi$mfxest[,1], umi$mfxest[,1], hi$mfxest[,1], eu$mfxest[,1], eur$mfxest[,1], noneur$mfxest[,1], cultclose$mfxest[,1], cultdist$mfxest[,1] ), se = list(germ$mfxest[,2], ngerm$mfxest[,2], lmi$mfxest[,2], umi$mfxest[,2], hi$mfxest[,2], eu$mfxest[,2], eur$mfxest[,2], noneur$mfxest[,2], cultclose$mfxest[,2], cultdist$mfxest[,2] ), column.labels = c("Germanic", "Non-Germanic", "Lower-middle income", "Upper-middle income", "High income", "EU", "European (non-EU)", "Non-European", "Culturally close", "Culturally far" ), header=TRUE, p.auto=FALSE, style="default", nobs=TRUE, align=TRUE, no.space=TRUE, single.row=FALSE, model.names = FALSE, dep.var.labels = "Professional use of German", covariate.labels= c( "Investment", "Age: under 35 years", "Gender: female", "Gender: n/a", "Children", "Partner (native German)", "Partner (other native)", "Occ.: high appl./low comm. skills", "Occ.: high appl./high comm. skills", "Occ.: in education", "Occ.: other occ./no answer", "University degree", "Risk attitude", "Patience", "English speaker"), omit = c("Constant", "country"), font.size = "small", add.lines = list(c("Country FE", "Yes", "Yes","Yes","Yes","Yes","Yes","Yes","Yes", "Yes", "Yes")), omit.stat = c("aic", "ll"), omit.table.layout = "n") germ_type<-tidy(germ) %>% filter(term != "(Intercept)") %>% mutate(model = "Germanic") ngerm_type<-tidy(ngerm) %>% filter(term != "(Intercept)") %>% mutate(model = "Non-Germanic") umi_type<-tidy(umi) %>% filter(term != "(Intercept)") %>% mutate(model = "Upper-middle income") hi_type<-tidy(hi) %>% filter(term != "(Intercept)") %>% mutate(model = "High income") lmi_type<-tidy(lmi) %>% filter(term != "(Intercept)") %>% mutate(model = "Lower-middle income") eu_type<-tidy(eu) %>% filter(term != "(Intercept)") %>% mutate(model = "EU") eur_type<-tidy(eur) %>% filter(term != "(Intercept)") %>% mutate(model = "Non-EU (European)") noneur_type<-tidy(noneur) %>% filter(term != "(Intercept)") %>% mutate(model = "Non-EU (non-European)") cultclose_type<-tidy(cultclose) %>% filter(term != "(Intercept)") %>% mutate(model = "Cuturally close") cultdist_type<-tidy(cultdist) %>% filter(term != "(Intercept)") %>% mutate(model = "Culturally dist.") TOT_type<-tidy(tot) %>% filter(term != "(Intercept)") %>% mutate(model = "Total") countrygr_l_type<-rbind(TOT_type, ngerm_type, germ_type, lmi_type, umi_type, hi_type, eu_type, eur_type, noneur_type, cultclose_type, cultdist_type) countrygr_l_type<-countrygr_l_type %>% mutate(sign=ifelse(model=="Total" & p.value<0.05, TRUE, FALSE)) %>% group_by(term) %>% mutate(sign_5=max(sign), sign=ifelse(sign_5==0, "P>0.05 (aggr. model)", "P<0.05 (aggr. model)"), sign=as.factor(sign)) %>% filter(model!="Total") p_countrygroup<-dwplot(subset(countrygr_l_type, ( countrygr_l_type$term=="typeTRUE" | countrygr_l_type$term=="factor(age)TRUE" | countrygr_l_type$term=="female1" | countrygr_l_type$term=="children_d_naTRUE" | countrygr_l_type$term=="part1" | countrygr_l_type$term=="part2" | countrygr_l_type$term=="occupationac1" | countrygr_l_type$term=="occupationac2" | countrygr_l_type$term=="occupationac3" | countrygr_l_type$term=="degreeTRUE" | countrygr_l_type$term=="risk" | countrygr_l_type$term=="patient" | countrygr_l_type$term=="englishTRUE" )), vline = geom_vline(xintercept = 0, colour = "grey60", linetype = 2), dodge_size= 0.7, dot_args = list(aes(shape =as.factor(sign)), size=4)) + scale_x_continuous(limits = c(-0.2, 0.4))+ scale_y_discrete( labels=c("typeTRUE" = "Investment", "factor(age)TRUE" = "Age: under 35 years", female1 = "Gender: female", children_d_naTRUE = "Children", "part1" = "Partner (native German)", "part2" = "Partner (other native)", degreeTRUE= "University degree", "occupationac1" = "Occ.: high appl./low comm. skills", "occupationac2" = "Occ.: high appl./high comm. skills", "occupationac3" = "Occ.: in education", risk = "Risk attitude", patient = "Patience", "englishTRUE" = "English speaker" ))+ xlab("AME")+ theme(legend.title = element_blank(), axis.text=element_text(size=10))+ scale_color_grey(guide=guide_legend(reverse = TRUE)) pdf("../graphics/langjob_countrychar.pdf", paper="a4", width = 12, height = 14) p_countrygroup dev.off() #### 2 c3 leave one country out #### TOT<-probitmfx(formula = langjob_full_type2_fe, data=gi, atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) BIH<-probitmfx(formula = langjob_full_type2_fe, data=gi[gi$country!="BIH",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) CZE<-probitmfx(formula = langjob_full_type2_fe, data=gi[gi$country!="CZE",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) ESP<-probitmfx(formula = langjob_full_type2_fe, data=gi[gi$country!="ESP",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) GBR<-probitmfx(formula = langjob_full_type2_fe, data=gi[gi$country!="GBR",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) ITA<-probitmfx(formula = langjob_full_type2_fe, data=gi[gi$country!="ITA",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) JPN<-probitmfx(formula = langjob_full_type2_fe, data=gi[gi$country!="JPN",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) KOR<-probitmfx(formula = langjob_full_type2_fe, data=gi[gi$country!="KOR",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) NLD<-probitmfx(formula = langjob_full_type2_fe, data=gi[gi$country!="NLD",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) POL<-probitmfx(formula = langjob_full_type2_fe, data=gi[gi$country!="POL",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) ROU<-probitmfx(formula = langjob_full_type2_fe, data=gi[gi$country!="ROU",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) IND<-probitmfx(formula = langjob_full_type2_fe, data=gi[gi$country!="IND",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) IDN<-probitmfx(formula = langjob_full_type2_fe, data=gi[gi$country!="IDN",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) UKR<-probitmfx(formula = langjob_full_type2_fe, data=gi[gi$country!="UKR",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) MEX<-probitmfx(formula = langjob_full_type2_fe, data=gi[gi$country!="MEX",], atmean = FALSE, robust = TRUE, clustervar1 = NULL, clustervar2 = NULL, start = NULL, control = list()) stargazer(TOT$fit, BIH$fit, CZE$fit, ESP$fit, GBR$fit, ITA$fit, JPN$fit, KOR$fit, NLD$fit, POL$fit, ROU$fit, IND$fit, IDN$fit, UKR$fit, MEX$fit, coef = list(TOT$mfxest[,1], BIH$mfxest[,1], CZE$mfxest[,1], ESP$mfxest[,1], GBR$mfxest[,1], ITA$mfxest[,1], JPN$mfxest[,1], KOR$mfxest[,1], NLD$mfxest[,1], POL$mfxest[,1], ROU$mfxest[,1], IND$mfxest[,1], IDN$mfxest[,1], UKR$mfxest[,1], MEX$mfxest[,1]), se = list(TOT$mfxest[,2], BIH$mfxest[,2], CZE$mfxest[,2], ESP$mfxest[,2], GBR$mfxest[,2], ITA$mfxest[,2], JPN$mfxest[,2], KOR$mfxest[,2], NLD$mfxest[,2], POL$mfxest[,2], ROU$mfxest[,2], IND$mfxest[,2], IDN$mfxest[,2], UKR$mfxest[,2], MEX$mfxest[,2]), column.labels = c("TOT", "BIH", "CZE", "ESP", "GBR", "ITA", "JPN", "KOR", "NLD", "POL", "ROU", "IND", "IDN", "UKR", "MEX"), header=FALSE, style="default", dep.var.labels = "Professional use of German", covariate.labels= c( "Domestic labour market", "Educational and labour migration", "Personal reasons", "Cultural interest", "Age: under 35 years", "Gender: female", "Gender: n/a", "Children", "Partner (native German)", "Partner (other native)", "Occ.: in education", "Occ.: not in lab. market", "University degree", "Risk attitude", "Patience", "Intern. applic. of education" ), nobs=TRUE, align=TRUE, no.space=TRUE, single.row=TRUE, model.names = FALSE, omit = c("Constant", "country", "empstat4"), font.size = "small", add.lines = list(c("Country FE", "Yes", "Yes", "Yes","Yes","Yes","Yes","Yes","Yes","Yes","Yes","Yes","Yes","Yes","Yes","Yes")), notes= "Average marginal effects. Heteroscedasticity robust White standard errors in parentheses.", omit.stat = c("aic", "ll")) TOT_type<-tidy(TOT) %>% filter(term != "(Intercept)") %>% mutate(model = "Total") BIH_type<-tidy(BIH) %>% filter(term != "(Intercept)") %>% mutate(model = "BIH (n=110)") CZE_type<-tidy(CZE) %>% filter(term != "(Intercept)") %>% mutate(model = "CZE (n=280)") ESP_type<-tidy(ESP) %>% filter(term != "(Intercept)") %>% mutate(model = "ESP (n=395)") GBR_type<-tidy(GBR) %>% filter(term != "(Intercept)") %>% mutate(model = "GBR (n=274)") ITA_type<-tidy(ITA) %>% filter(term != "(Intercept)") %>% mutate(model = "ITA (n=186)") JPN_type<-tidy(JPN) %>% filter(term != "(Intercept)") %>% mutate(model = "JPN (n=183)") KOR_type<-tidy(KOR) %>% filter(term != "(Intercept)") %>% mutate(model = "KOR (n=189)") NLD_type<-tidy(NLD) %>% filter(term != "(Intercept)") %>% mutate(model = "NLD (n=60)") POL_type<-tidy(POL) %>% filter(term != "(Intercept)") %>% mutate(model = "POL (n=121)") ROU_type<-tidy(ROU) %>% filter(term != "(Intercept)") %>% mutate(model = "ROU (n=197)") IND_type<-tidy(IND) %>% filter(term != "(Intercept)") %>% mutate(model = "IND (n=373)") IDN_type<-tidy(IDN) %>% filter(term != "(Intercept)") %>% mutate(model = "IDN (n=485)") UKR_type<-tidy(UKR) %>% filter(term != "(Intercept)") %>% mutate(model = "UKR (n=458)") MEX_type<-tidy(MEX) %>% filter(term != "(Intercept)") %>% mutate(model = "MEX (n=281)") gi %>% count(country) %>% arrange(-n) country_l_type<-rbind(TOT_type, ESP_type, GBR_type, CZE_type, ITA_type, POL_type, ROU_type, NLD_type, BIH_type, UKR_type, MEX_type, IND_type, IDN_type, KOR_type, JPN_type) country_l_type<-country_l_type %>% mutate(sign=ifelse(model=="Total" & p.value<0.05, TRUE, FALSE)) %>% group_by(term) %>% mutate(sign_5=max(sign), sign=ifelse(sign_5==0, "P>0.05 (aggr. model)", "P<0.05 (aggr. model)"), sign=as.factor(sign))