Load data

data <- read_rds('../../temp/data_reg.rds')

Regression data prep

Sorting etc

# Just to be sure
data %<>%
  distinct(AID, year, .keep_all = TRUE) %>%
  relocate(AID, year, type, transition, transited_t, seniority)

Lagged variables

# Decide what to lag
data %<>%
  arrange(AID, year) %>%
  group_by(AID) %>%
  mutate(
    across(c(concepts,
             DL,
             paper_n, 
             citation_n, 
             author_mean,
             oa_mean, 
             novelty_mean, 
             deg_cen, 
             deg_cen_comp, 
             novelty_fw, 
             citation_fw,
             aff_citation_fw,
             aff_novelty_fw
             ), ~ lag(.x, 1))
  ) %>%
  ungroup()

# Some last preprocessing
data %<>%
  filter(year >= 2000,
         year <= 2021) %>%
  mutate(co_author_industry = deg_cen_comp > 0) %>% #,year_min = ifelse(year_min < 2000, 2000, year_min))
    group_by(AID) %>%
  mutate(t = 1:n(),
         t_max = n()) %>%
  ungroup()

# Check missing data
data  %>% naniar::gg_miss_var()

# Replace again missing
data %<>%   
  group_by(AID) %>%
    arrange(AID, year) %>%
    fill(DL, concepts, 
         aff_id, aff_type, aff_novelty_fw, aff_citation_fw, 
         novelty_fw, citation_fw, paper_n, oa_mean, novelty_mean, deg_cen_comp, deg_cen, # IS not really correct, but can be done. Consider
         co_author_industry, citation_n, author_mean, # IS not really correct, but can be done. Consider
         .direction = "downup") %>%
  ungroup()

# Replace NA. All others are by construction
data %<>%
  replace_na(list(DL = 0, concepts = 'unknown'))

data %>% count(concepts, sort = TRUE) %>% 
  mutate(rank_pct = percent_rank(n))

Survival analysis

data_surv <- data %>% 
  # Filter for last observation (=right cernsoring or transition)
  filter(is.na(transited_t) | transited_t <= 1) %>%
  group_by(AID) %>%
  slice_max(order_by = t, n= 1, with_ties = FALSE) %>%
  ungroup() %>%
  # Add some variables
  mutate(auth_star = citation_fw >= 0.9,
         aff_star = aff_citation_fw >= 0.9,
         )


# Previous filtering approach:
# filter(  (type == 'education' & year == year_max) | (type == 'switcher' & transition == 1)) %>%

data_surv %>% count(type)

# Surf object
surv_object <- Surv(time = data_surv$t, event = data_surv$transition)

table(surv_object)

surv_object
    1     2     3     4    4+     5    5+     6    6+     7    7+     8    8+     9    9+    10   10+    11   11+    12   12+    13   13+    14   14+    15 
    2    71    97   117 17504   110 13583   128 10859    95  8588    63  7404    51  6341    33  5430    19  5214    27  4649    16  4068    13  3409     3 
  15+    16   16+    17   17+    18   18+    19   19+    20   20+    21   21+   22+ 
 3012     7  2689     3  2276     8  1917     4  1739     5  1610     9  1547  2576

Kaplan-Meier Method and Log Rank Test

fit_km0 <- survfit(surv_object ~ 1, 
                   data = data_surv)
fit_km1 <- survfit(surv_object ~ DL, 
                   data = data_surv)
fit_km2 <- survfit(surv_object ~ auth_star, 
                   data = data_surv)
fit_km3 <- survfit(surv_object ~ aff_star,
                   data = data_surv)

p1 <- fit_km0 %>% autoplot() + labs(title= 'KM: baseline')
p2 <- fit_km1 %>% autoplot() + labs(title= 'KM: DL researcher')
p3 <- fit_km2 %>% autoplot() + labs(title= 'KM: Star researcher')
p4 <- fit_km3 %>% autoplot() + labs(title= 'KM: Star institution')

(p1 + p2) / (p3 + p4) +  plot_layout(guides = "collect") & theme(legend.position = 'bottom') & labs(x = 'Time', y = 'Survival Probability')

rm(fit_km0, fit_km1, fit_km2, fit_km3,
   p1, p2, p3, p4)

Cox Proportional Hazard Model

### Fit Cox proportional hazard model
set.seed(1337)

#Baseline
fit_cox0 <- coxph(surv_object ~ seniority + concepts + year,
                  data = data_surv)

Warning: Loglik converged before variable  13,31,38,40,43,48,51,53,54,56,58,60,61,62,64,67,70,72,76,78,79,85,87,89,90,94,95,96,98,99,101 ; coefficient may be infinite.

fit_cox1 <- coxph(surv_object ~ seniority + concepts + year 
                  + DL,
                  data = data_surv)

Warning: Loglik converged before variable  13,31,38,40,43,48,51,53,54,56,58,60,61,62,64,67,70,72,76,78,79,85,87,89,90,94,95,96,98,99,101 ; coefficient may be infinite.

fit_cox2 <- coxph(surv_object ~ seniority + concepts + year 
                  + deg_cen + deg_cen_comp,
                  data = data_surv)

Warning: Loglik converged before variable  55 ; coefficient may be infinite.

fit_cox3 <- coxph(surv_object ~ seniority + concepts + year 
                  + citation_fw + novelty_fw,
                  data = data_surv)

Warning: Loglik converged before variable  13,31,38,40,48,51,53,54,56,58,60,61,62,67,70,72,76,78,79,85,89,90,94,95,96,98,99,101 ; coefficient may be infinite.

fit_cox4 <- coxph(surv_object ~ seniority + concepts + year 
                  + aff_citation_fw,
                  data = data_surv)

Warning: Loglik converged before variable  13,31,38,40,43,48,51,53,54,56,58,60,61,62,64,67,70,72,76,78,79,85,87,89,90,94,95,96,98,99,101 ; coefficient may be infinite.

fit_cox5 <- coxph(surv_object ~ seniority + concepts + year 
                  + DL + deg_cen + deg_cen_comp + citation_fw + novelty_fw + aff_citation_fw,
                  data = data_surv)

#stargazer(fit_cox0, fit_cox1, fit_cox2, fit_cox3, fit_cox4, fit_cox5, type = 'latex', 
# omit = c( 'concepts', 'year', 'Constant'), out ='../../output/surv_res1.tex')

#stargazer(fit_cox0, fit_cox1, fit_cox2, fit_cox3, fit_cox4, fit_cox5, type = 'text',  omit = c( 'concepts', 'year', 'Constant'))

stargazer(fit_cox0, fit_cox1, fit_cox2, fit_cox3, fit_cox4, fit_cox5, type = 'html', omit = c( 'concepts', 'year', 'Constant'))

Warning: length of NULL cannot be changedWarning: length of NULL cannot be changedWarning: length of NULL cannot be changedWarning: length of NULL cannot be changedWarning: length of NULL cannot be changedWarning: number of rows of result is not a multiple of vector length (arg 2)Warning: number of rows of result is not a multiple of vector length (arg 2)Warning: number of rows of result is not a multiple of vector length (arg 2)Warning: number of rows of result is not a multiple of vector length (arg 2)Warning: number of rows of result is not a multiple of vector length (arg 2)


	Dependent variable:

	surv_object
	(1)	(2)	(3)	(4)	(5)	(6)

seniority	-0.232^***	-0.226^***	-0.239^***	-0.213^***	-0.220^***	-0.209^***
	(0.011)	(0.011)	(0.011)	(0.011)	(0.011)	(0.011)

DL		1.349^***				0.692^***
		(0.076)				(0.081)

deg_cen			0.190^***			-0.308^***
			(0.028)			(0.047)

deg_cen_comp			0.129^***			0.127^***
			(0.007)			(0.007)

citation_fw				4.261^***		4.181^***
				(0.130)		(0.156)

novelty_fw				-0.876^***		-0.880^***
				(0.157)		(0.159)

aff_citation_fw					5.379^***	0.292
					(0.275)	(0.326)


Observations	105,296	105,296	105,296	105,296	105,296	105,296
R²	0.017	0.020	0.020	0.030	0.020	0.033
Max. Possible R²	0.169	0.169	0.169	0.169	0.169	0.169
Log Likelihood	-8,867.794	-8,709.206	-8,684.612	-8,173.723	-8,697.137	-8,017.109
Wald Test	262.910^*** (df = 99)	597.870^*** (df = 100)	2,188.910^*** (df = 70)	1,383.660^*** (df = 98)	642.810^*** (df = 100)	3,284.940^*** (df = 74)
LR Test	1,781.754^*** (df = 99)	2,098.929^*** (df = 100)	2,148.117^*** (df = 70)	3,169.896^*** (df = 98)	2,123.067^*** (df = 100)	3,483.124^*** (df = 74)
Score (Logrank) Test	1,560.421^*** (df = 99)	1,926.216^*** (df = 100)	7,626.015^*** (df = 70)	3,211.857^*** (df = 98)	1,954.019^*** (df = 100)	8,859.206^*** (df = 74)

Note:	p<0.1; p<0.05; p<0.01

#fit_cox5 %>% ggforest(data = data_surv)

#save models
list(fit_cox0, fit_cox1, fit_cox2, fit_cox3, fit_cox4, fit_cox5) %>% write_rds('../../temp/reg_surv1.rds')

#rm(fit_cox0, fit_cox1, fit_cox2, fit_cox3, fit_cox4, fit_cox5)

Propensity Score matching

library(MatchIt)

data_match <- data %>%
  filter(type == 'education' | type == 'switcher' & transition == 1) %>%
  left_join(read_rds('../../temp/author_concepts_main.rds') %>% select(AID, concepts) %>% rename(concept_main = concepts), by = 'AID') %>%
  select(-transition, -transited_t, -year_transit) %>%
  mutate(type = (type %>% factor() %>% as.numeric()) -1,
         aff_id = aff_id %>% factor(),
         concepts = concepts %>% factor(),
         concept_main = concept_main %>% factor()) %>%
  drop_na()

# Do the PSM
set.seed(1337)
match_out <-  data_match %>% 
  # Matching
  matchit(type ~ paper_n + author_mean + oa_mean + DL + t + t_max + co_author_industry + aff_citation_fw + aff_citation_fw, exact = c('year', 'seniority', 'concept_main'),
          data = ., verbose = FALSE, method = "nearest", ratio = 1) # OR method  "genetic" ... takes forever.....

# GEt the matched pairs
matched_pairs  <- bind_cols(data_match[row.names(match_out$match.matrix),"AID"] , data_match[match_out$match.matrix,"AID"] ) %>%
  rename(orig_AID = `AID...1`,
         match_AID = `AID...2`)

New names:

# construct the matched dataset
data_did <- matched_pairs %>% 
  mutate(pair_id = 1:n())

data_did <- data_did %>%
  select(orig_AID, pair_id) %>%
  rename(AID = orig_AID) %>%
  bind_rows(data_did %>%
  select(match_AID, pair_id) %>%
  rename(AID = match_AID) ) %>%
  arrange(pair_id) %>%
  left_join(data, by = 'AID')

# compute transition point for matched partner
data_did %<>%
  group_by(pair_id) %>%
  mutate(switcher = type == 'switcher',
         year_transit = max(year_transit, na.rm = TRUE),
         transition = year == year_transit,
         transited = year >= year_transit,
         transited_t = year - year_transit + 1) %>%
  mutate(across(c(year, concepts, transited), ~ factor(.x))) %>%
  mutate(transited_t = ifelse(transited_t > 0, transited_t, 0)) %>%
  drop_na()

data_did %>% saveRDS('../temp_reg/reg_data_did.rds')

Diff-in-Diff

# Delete all objects
rm(list = ls(all.names = TRUE)) #will clear all objects includes hidden objects.
gc() #free up memrory and report the memory usage.

           used  (Mb) gc trigger    (Mb) limit (Mb)   max used    (Mb)
Ncells  3874576 207.0   12010248   641.5         NA   12010248   641.5
Vcells 23545267 179.7 5445164727 41543.4      65536 8507759093 64909.1

data_did <- read_rds('../temp_reg/reg_data_did.rds')

# For the interaction plot code logical to numeric
# data_did %<>%  mutate(switcher = switcher %>% as.numeric(), transited = transited %>% as.numeric())

data_did %>% 
  count(concepts, sort = TRUE) %>% 
  mutate(rank_pct = percent_rank(n))

Models

# fit_did_0 <- lm(citation_rank ~ switcher + field_of_study_name, data = data_did)

fit_did_1 <- lm(citation_fw ~ (type + transited)^2 + year + seniority + year, # + concepts, 
                data = data_did)

fit_did_2 <- lm(citation_fw ~ (type + transited)^2 + (type + transited_t)^2 + year  + seniority, # + concepts, 
                data = data_did)

fit_did_3 <- lm(novelty_fw ~ (type + transited)^2 + year + seniority + year, # + concepts, 
                data = data_did)

fit_did_4 <- lm(novelty_fw ~ (type + transited)^2 + (type + transited_t)^2 + year + seniority, # + concepts , 
                data = data_did)

# + paper_n ?

#stargazer(fit_did_1, fit_did_2, fit_did_3, fit_did_4, type = 'latex', omit = c( 'concepts', 'year', 'Constant'), out ='../../output/did_res1.tex')

stargazer(fit_did_1, fit_did_2, fit_did_3, fit_did_4, type = 'text', omit = c('year', 'Constant'))

Warning: length of NULL cannot be changedWarning: length of NULL cannot be changedWarning: length of NULL cannot be changedWarning: length of NULL cannot be changedWarning: length of NULL cannot be changedWarning: number of rows of result is not a multiple of vector length (arg 2)Warning: number of rows of result is not a multiple of vector length (arg 2)


========================================================================================================================================
                                                                       Dependent variable:                                              
                         ---------------------------------------------------------------------------------------------------------------
                                               citation_fw                                             novelty_fw                       
                                     (1)                         (2)                         (3)                         (4)            
----------------------------------------------------------------------------------------------------------------------------------------
typeswitcher                      0.050***                    0.051***                      0.007                       0.007           
                                   (0.007)                     (0.007)                     (0.005)                     (0.005)          
                                                                                                                                        
transited                         0.123***                    0.203***                    0.101***                    0.112***          
                                   (0.009)                     (0.011)                     (0.005)                     (0.007)          
                                                                                                                                        
transited_t                                                   -0.021***                                               -0.003***         
                                                               (0.002)                                                 (0.001)          
                                                                                                                                        
seniority                         0.004***                    0.007***                    0.006***                    0.006***          
                                  (0.0004)                    (0.0004)                    (0.0002)                    (0.0002)          
                                                                                                                                        
typeswitcher:transited            0.114***                    0.172***                    -0.020***                    -0.010           
                                   (0.011)                     (0.014)                     (0.007)                     (0.009)          
                                                                                                                                        
typeswitcher:transited_t                                      -0.015***                                                -0.003*          
                                                               (0.002)                                                 (0.001)          
                                                                                                                                        
----------------------------------------------------------------------------------------------------------------------------------------
Observations                       17,151                      17,151                      17,151                      17,151           
R2                                  0.165                       0.195                       0.142                       0.143           
Adjusted R2                         0.164                       0.194                       0.140                       0.142           
Residual Std. Error          0.349 (df = 17125)          0.343 (df = 17123)          0.222 (df = 17125)          0.222 (df = 17123)     
F Statistic              135.153*** (df = 25; 17125) 153.683*** (df = 27; 17123) 113.069*** (df = 25; 17125) 106.066*** (df = 27; 17123)
========================================================================================================================================
Note:                                                                                                        *p<0.1; **p<0.05; ***p<0.01

stargazer(fit_did_1, fit_did_2, fit_did_3, fit_did_4, type = 'html', omit = c('year', 'Constant'))


	Dependent variable:

	citation_fw		novelty_fw
	(1)	(2)	(3)	(4)

typeswitcher	0.050^***	0.051^***	0.007	0.007
	(0.007)	(0.007)	(0.005)	(0.005)

transited	0.123^***	0.203^***	0.101^***	0.112^***
	(0.009)	(0.011)	(0.005)	(0.007)

transited_t		-0.021^***		-0.003^***
		(0.002)		(0.001)

seniority	0.004^***	0.007^***	0.006^***	0.006^***
	(0.0004)	(0.0004)	(0.0002)	(0.0002)

typeswitcher:transited	0.114^***	0.172^***	-0.020^***	-0.010
	(0.011)	(0.014)	(0.007)	(0.009)

typeswitcher:transited_t		-0.015^***		-0.003^*
		(0.002)		(0.001)


Observations	17,151	17,151	17,151	17,151
R²	0.165	0.195	0.142	0.143
Adjusted R²	0.164	0.194	0.140	0.142
Residual Std. Error	0.349 (df = 17125)	0.343 (df = 17123)	0.222 (df = 17125)	0.222 (df = 17123)
F Statistic	135.153^*** (df = 25; 17125)	153.683^*** (df = 27; 17123)	113.069^*** (df = 25; 17125)	106.066^*** (df = 27; 17123)

Note:	p<0.1; p<0.05; p<0.01

#library(interplot)
#fit_did_2 %>% interplot(var1 ="switcher", var2 = "transited_t", hist = TRUE)

library(interactions)
fit_did_2 %>% interact_plot(pred = transited_t, modx = switcher,
                            interval = TRUE) +
  theme(legend.position = 'bottom')

ggsave('../../output/fig_interaction_1.png', width = 10, height = 7.5, units = 'cm')

library(interactions)
fit_did_2 %>% interact_plot(pred = transited_t, modx = switcher,
                            interval = TRUE, linearity.check = TRUE)

—>

Descriptives

data %>% select(seniority, deg_cen, deg_cen_comp, co_author_industry, DL, paper_n, author_mean, citation_fw, novelty_fw, aff_citation_fw, aff_novelty_fw) %>% as.data.frame() %>%
  stargazer(summary = TRUE, type = 'html')

By university vs. switcher career

data %>% 
  select(type, seniority, deg_cen, deg_cen_comp, co_author_industry, DL, paper_n, author_mean, citation_fw, novelty_fw, aff_citation_fw, aff_novelty_fw) %>%
  as.data.frame() %>%
  split(.$type) %>% 
  walk(~ stargazer(., summary = TRUE, type = 'html'))

data %>% 
  filter(transited_t <= 0) %>%
  mutate(auth_star = citation_fw >= 0.9,
         aff_star = aff_citation_fw >= 0.9)

–>

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

Privatization of AI research: Executing regressions

Daniel S. Hain (dsh@business.aau.dk)

Updated March 28, 2023