Because the World Bank doesn’t include it in the WDI, we omit Taiwan (713). We also omit East Germany (265) and South Yemen (680)
We only use the 163 countries in Suparna’s anti-NGO law data
Following Finkel, Pérez-Liñán, and Seligson (2007) (p. 414), we exclude long-term consolidated democracies from our analysis. These are classified by the World Bank as high income; they score below 3 on Freedom House’s Scale, receive no aid from USAID, and are not newly independent states:
When converting GW codes to COW codes, following Gleditsch and Ward, we treat post-2006 Serbia as 345 (a continuation of Serbia & Montenegro). And we also treat Serbia as a continuation of Yugoslavia with 345 (following V-Dem, which does that too).
In both COW and GW codes, modern Vietnam is 816, but countrycode() thinks the COW code is 817, which is old South Vietnam (see issue), so we use custom_match to force 816 to recode to 816.
Also, following V-Dem, we treat Czechoslovakia (GW/COW 315) and Czech Republic (GW/COW 316) as the same continuous country (V-Dem has both use ID 157).
Thus, here’s our actual panel skeleton:
We have 142 countries in our final data, spanning 24 possible years. Here’s a lookup table of all the countries included:
The OECD collects detailed data on all foreign aid flows (ODA) from OECD member countries (and some non-member countries), multilateral organizations, and the Bill and Melinda Gates Foundation (for some reason they’re the only nonprofit donor) to all DAC-eligible countries (and some non non-DAC-eligible countries).
The OECD tracks all this in a centralized Creditor Reporting System database and provides a nice front end for it at OECD.Stat with an open (but inscrutable) API (raw CRS data is also available). There are a set of pre-built queries with information about ODA flows by donor, recipient, and sector (purpose), but the pre-built data sources do not include all dimensions of the data. For example, Table DAC2a includes columns for donor, recipient, year, and total ODA (e.g. the US gave $X to Nigeria in 2008) , but does not indicate the purpose/sector for the ODA. Table DAC5 includes columns for the donor, sector, year, and total ODA (e.g. the US gave $X for education in 2008), but does not include recipient information.
Instead of using these pre-built queries or attempting to manipulate their parameters, it’s possible to use the OECD’s QWIDS query builder to create a custom download of data. However, it is slow and clunky and requires significant munging and filtering after exporting.
The solution to all of this is to use data from AidData, which imports raw data from the OECD, cleans it, verifies it, and makes it freely available on GitHub.
AidData offers multiple versions of the data, including a full release, a thin release, aggregated donor/recipient/year data, and aggregated donor/recipient/year/purpose data. For the purposes of this study, all we care about are ODA flows by donor, recipient, year, and purpose, which is one of the ready-made datasets.
Notably, this aggregated data shows total aid commitments, not aid disbursements. Both types of ODA information are available from the OECD and it’s possible to get them using OECD’s raw data. However, AidData notes that disbursement data is sticky and slow—projects take a long time to fulfill and actual inflows of aid in a year can be tied to commitments made years before. Because we’re interested in donor reactions to restrictions on NGOs, any reaction would be visible in the decision to commit money to aid, not in the ultimate disbursement of aid, which is most likely already legally obligated and allocated to the country regardless of restrictions.
USAID provides the complete dataset for its Foreign Aid Explorer as a giant CSV file. The data includes both economic and military aid, but it’s easy to filter out the military aid. Here we only look at obligations, not disbursements, so that the data is comparable to the OECD data from AidData. The data we downloaded provides constant amounts in 2015 dollars; we rescale that to 2011 to match all other variables.
USAID distinguishes between domestic, foreign, and international NGOs, companies, multilateral organizations, etc. implementing partners (formerly called channels):
Suparna Chaudhry expanded this data substantially (it now covers 163 countries and goes to 2013), so we use that.
Notes on year coverage
In our original paper from 2017, we used Suparna’s data and backfilled it to 1980, since going back in time is possible with the DCJW data—lots of the entries in DCJW include start dates of like 1950 or 1970. Accordingly, our analysis ranged from 1980-2013. However, not all of Suparna’s expanded countries went back in time that far, and she focused primarily on 1990+ changes. Additionally—and more importantly—the whole nature of foreign aid and civil society changed drastically after the Cold War. Civil society regulations weren’t really used as a political strategy until after 1990. We can confirm that by plotting V-Dem’s core civil society index:
Code
vdem_clean %>%filter(year >=1980) %>%select(year, v2xcs_ccsi) %>%group_by(year) %>%summarize(avg_ccsi =mean(v2xcs_ccsi)) %>%ggplot(aes(x = year, y = avg_ccsi)) +geom_line() +geom_vline(xintercept =1990, color ="red") +labs(x ="Year", y ="Average Core Civil Society Index", caption ="Source: V-Dem's v2xcs_ccsi") +theme_donors()
Something systematic happened to civil society regulations worldwide in 1990, and rather than try to model pre-Cold War regulations, which were connected to foreign aid in completely different ways than they were after the dissolution of the USSR, we limit our analysis to 1990+
We still collect as much pre-1990 data as possible for the sake of (1) lagging, so we can get lagged values from 1989 and 1988 when looking at lagged variables in 1990, and (2) robustness checks that we run using the 98 backfilled DCJW countries.
Index creation
We create several indexes for each of the categories of regulation, following Christensen and Weinstein’s classification:
entry (Q2b, Q2c, Q2d; 3 points maximum, actual max = 3 points maximum): barriers to entry
Q2c is reversed, so not being allowed to appeal registration status earns 1 point.
Q2a is omitted because it’s benign
funding (Q3b, Q3c, Q3d, Q3e, Q3f; 5 points maximum, actual max = 4.5): barriers to funding
Q3a is omitted because it’s benign
Scores that range between 0–2 are rescaled to 0–1 (so 1 becomes 0.5)
advocacy (Q4a, Q4c; 2 points maximum, actual max = 2): barriers to advocacy
Q4b is omitted because it’s not a law
Scores that range between 0–2 are rescaled to 0–1 (so 1 becomes 0.5)
barriers_total (10 points maximum, actual max = 8.5): sum of all three indexes
These indexes are also standardized by dividing by the maximum, yielding the following variables:
entry_std: 1 point maximum, actual max = 1
funding_std: 1 point maximum, actual max = 1
advocacy_std: 1 point maximum, actual max = 1
barriers_total_std: 3 points maximum, actual max = 2.5
Chaudhry 2020 data
The most recent version of Suparna’s data is already in nice clean panel form, so it’s super easy to get cleaned up.
For fun and robustness checks, we use DCJW’s non-panel data to generate a panel starting in 1980, since they have entries where laws start in the 1960s and 70s and other pre-1980 years.
Suparna made updates to existing the DCJW countries too, like Honduras (gwcode 91), which has more correct values for q4a, for instance, which DCJW marks as 0, but is actually 1. So even though we can go back in time to 1980 with DCJW, it’s not comparable with Suparna’s expanded and more recent data.
So we live with just 1990+, even for the sake of lagging 🤷.
Except, we’re not quite done yet!
In Suparna’s clean data, due to post-Cold War chaos, Russia (365) is missing for 1990-1991 and Serbia/Serbia and Montenegro/Yugoslavia (345) is missing every thing pre-2006. DCJW don’t include any data for Serbia, so we’re out of luck there—we’re limited to Serbia itself and not past versions of it. DCJW do include data for Russia, though, so we use that in our clean final NGO laws data. Fortunately this is easy, since Russia’s values are all 0 for those two years:
An alternative way of measuring civil society restrictions is to look at the overall civil society regulatory environment rather than specific laws, since de jure restrictions do not always map clearly into de facto restrictions (especially in dictatorships where the implementation of laws is more discretionary).
Andrew Heiss develops a new civil society regulatory environment index (CSRE) in his dissertation, which combines two civil society indexes from the Varieties of Democracy project (V-Dem): (1) civil society repression (v2csreprss) and (2) civil society entry and exit regulations (v2cseeorgs). The CSRE ranges from roughly −6 to 6 (though typically only from −4 to 4ish), and shows more variation over time since it ostensibly captures changes in the implementation of the regulatory environment rather than the presence or absence of legislation.
Additionally, since Andrew’s dissertation, the V-Dem project has created its own core civil society index (v2xcs_ccsi) (entry/exit (v2cseeorgs) + repression (v2csreprss) + participatory environment (v2csprtcpt)). We use that instead, since V-Dem rescales it to a 0-1 scale so it’s not weirdly distributed from like my additive two-factor CSRE index.
While the main focus of this paper is donor response to new legislation, we also look at donor response to changes in the overall civil society index as a robustness check. This also allows us to potentially include data from 1980–2018 (since Suparana’s law data ranges from 1990–2014).
Other variables
We also use a bunch of other V-Dem variables as confounders:
Civil society things
CSO entry and exit: v2cseeorgs
CSO repression: v2csreprss
CSO consultation: v2cscnsult
CSO participatory environment: v2csprtcpt
CSO women’s participation: v2csgender
CSO anti-system movements: v2csantimv
Core civil society index (entry/exit, repression, participatory env): v2xcs_ccsi
Human rights and politics
Polity scores: e_polity2 (only for reference with polyarchy; we don’t use these in models)
Electoral democracy index: v2x_polyarchy
Regimes of the world scores: v2x_regime_amb (only for determining average autocracy; we don’t use these in models)
Political corruption index: v2x_corr (less to more, 0-1) (public sector + executive + legislative + judicial corruption)
Rule of law index: v2x_rule
Civil liberties index: v2x_civlib
Physical violence index: v2x_clphy
Private civil liberties index: v2x_clpriv
Political civil liberties index: v2x_clpol (but not this because it includes v2cseeorgs and v2csreprss)
We’re also interested in how these civil society dynamics work in autocracies in particular. We generate a crude list of autocracies based on average “Regimes of the World” scores (for ambiguous cases) from V-Dem. In that measure, 4 is the upper bound and 5 is the lower bound for electoral autocracy, so we use 4 as the cutoff. There are bound to be better ways, but this works for now.
The reason we don’t just use WDI data for GDP and % of GDP from trade is that the WDI data is incomplete, especially pre-1990. To get around that, we create our own GDP and trade measures using data directly from the UN (at UNData). They don’t have a neat API like the World Bank, so you have to go to their website and export the data manually.
Dealing with the GDP data is a little trickier because it’s in 2015 dollars, while AidData is in 2011 dollars. To fix this, we create a GDP deflator and rebase the 2015 values to 2011 values.
The UCDP/PRIO Armed Conflict Dataset tracks a ton of conflict-releated data, including reasons for the conflict, parties in the conflict, intensity of the conflict, and deaths in the conflict. We’re only interested in whether a conflict happened in a given year (or in the past 5 years), so here we simply create an indicator variable for whether there was internal conflict in a country-year (conflict type = 3).
Natural disaster data comes from the International Disaster Database (EM-DAT). The data includes the number of deaths, injuries, homeless displacements, and monetary losses (in 2000 dollars) for a huge number of natural and technological disasters (see EM-DAT’s full classification).
Natural disasters could matter for aid too, since donor countries might increase their aid to countries suffering more.
EM-DAT does not provide a single link to download their data. Instead, you have to create a query using their advanced search form. We downloaded data using the following query:
Select all countries from 1950–2021
Select all three disaster classification groups (natural, technological, complex)
Download Excel file and save in Data/raw_data/Disasters/
With both donor- and country-level data, we have lots of different options for analysis. Since our hypotheses deal with questions of donor responses, the data we use to model donor responses uses donor-years as the unit of observation. Not all donors give money to the same countries, so this final data is not a complete panel (i.e. it does not include every combination of donors and years), which will pose some interesting methodological issues when modeling if we use donor-level data.
Inflation
USAID’s conversion to constant 2015 dollars doesn’t seem to take country differences into account—the deflator for each country in 2011 is essentially 96.65. When there are differences, it’s because of floating point issues (like, if there are tiny grants of $3, there aren’t enough decimal points to get the fraction to 96.65). So we just take the median value of the deflator for all countries and all grants and use that.
Missing data
The donor data is complete with no missing variables(!).
Code
gg_miss_var(donor_level_data, show_pct =TRUE) +theme_donors()## Warning: The `guide` argument in `scale_*()` cannot be `FALSE`. This was deprecated in ggplot2 3.3.4.## ℹ Please use "none" instead.## ℹ The deprecated feature was likely used in the naniar package.## Please report the issue at <]8;;https://github.com/njtierney/naniar/issueshttps://github.com/njtierney/naniar/issues]8;;>.
The country-level panel data is relatively complete, with only a few variables suffering from missing data, mostly from the World Bank and V-Dem. There are a lot of NGO-related missing variables, but that’s because we don’t have data from 1980–1989 and 2015+
We remove everything from Yugoslavia/Serbia and Montenegro (345) prior to 2006
Infant mortality e_peinfmor is missing from Kosovo (2008–2014), and the World Bank doesn’t have data for it, but Eurostat does in their demo_minfind indicator. Their data, however, is missing a couple years
To fix this, we use linear interpolation to fill in 2013 and 2014.
v2x_corr is only missing data from Bahrain, which oddly has no data from 1980–2004. Because corruption levels do not really change after 2005, we impute the average corruption for the country in all previous years.
v2x_polyarchy is only missing in Mozambique from 1980–1993. To address this, we calculate the average value of V-Dem’s polyarchy index (v2x_polyarchy) for each level of Polity (−8, −7, and −6 in the case of Mozambique), and then use that corresponding average polyarchy
We also create an imputed column for those rows in Bahrain and Mozambique to see if imputation does anything weird in the models
Kosovo is missing pre-existence infant mortality, which is fine becuase it didn’t exist yet.
Russia is missing GDP, GDP per capita, and percent of GDP from trade from 1980–1989. There’s no easy way around this. V-Dem has GDP per capita data from the long-running Maddison Project Database, and it includes 1980s Soviet Russia, but the values aren’t really comparable to the stuff we calculated using UN GDP data. At first glance it seems that this is a difference in real years, since the Maddison Project uses 2011 dollars and the UN uses 2015 dollars, and there’s not an easy way to shift the Maddison Project’s values up to 2015 (i.e. there’s no deflator). But even if they were in the same dollar-years, the values from the Maddison Project seem really really low compared to what we made with the UN GDP data, so they don’t seem to be comparable.
Czechoslovakia is missing percent of GDP from trade from 1980–1989. This is because it is missing imports data in the UN GDP data. It has exports data and overall GDP data, but for whatever reason, imports are missing. Boo.
# A tibble: 20 × 7
gwcode country year un_trade_pct_gdp un_gdp gdpcap gdpcap_log
<dbl> <chr> <dbl> <dbl> <dbl> <dbl> <dbl>
1 316 Czechia 1980 NA 183919110258. 17772. 9.79
2 316 Czechia 1981 NA 183743647861. 17726. 9.78
3 316 Czechia 1982 NA 183984035455. 17750. 9.78
4 316 Czechia 1983 NA 188240745011. 18180. 9.81
5 316 Czechia 1984 NA 194289061254. 18789. 9.84
6 316 Czechia 1985 NA 199875093661. 19347. 9.87
7 316 Czechia 1986 NA 205065586592. 19857. 9.90
8 316 Czechia 1987 NA 209182118795. 20255. 9.92
9 316 Czechia 1988 NA 214531745219. 20766. 9.94
10 316 Czechia 1989 NA 215359391664. 20836. 9.94
11 365 Russia 1980 NA NA NA NA
12 365 Russia 1981 NA NA NA NA
13 365 Russia 1982 NA NA NA NA
14 365 Russia 1983 NA NA NA NA
15 365 Russia 1984 NA NA NA NA
16 365 Russia 1985 NA NA NA NA
17 365 Russia 1986 NA NA NA NA
18 365 Russia 1987 NA NA NA NA
19 365 Russia 1988 NA NA NA NA
20 365 Russia 1989 NA NA NA NA
Since those issues are all pre-1990, our data is perfect post-1990 in cases with Suparna’s law coverage:
We do get some missing values when lagging, though, but that only happens in countries that emerged post-1990 (like Kosovo, South Sudan, etc.), since there are no 80s-era values to look at for lagging
Final data
Now that we know all the data is clean and pretty much nothing is missing, we can do a few final windowed operations that will add missing values (e.g. lagging). We also add an indicator marking if a disaster happened in the past 5 years.
In H3 we hypothesize that more aid will be allocated to international or US-based NGOs than domestic NGOs in response to harsher anti-NGO restrictions. While AidData unfortunately does not categorize aid by channel (i.e. aid given to international vs. US vs. domestic NGOs), USAID does. For this hypothesis, then we only look at aid given by USAID, not the rest of the OECD. As with the proportion of contentious aid, we create similar variables to measure the proportion of aid given to international NGOs, US-based NGOs, and both international and US-based NGOs.
Finkel, Steven E., Anı́bal Pérez-Liñán, and Mitchell A. Seligson. 2007. “The Effects of U.S. Foreign Assistance on Democracy Building, 1990–2003.”World Politics 59 (3): 404–38. https://doi.org/10.1017/s0043887100020876.
Source Code
---title: "Process and merge data"author: "Suparna Chaudhry and Andrew Heiss"date: "Last run: `r format(Sys.time(), '%F')`"link-citations: true---```{r setup, include=FALSE}knitr::opts_chunk$set(fig.align ="center", fig.retina =3,fig.width =6, fig.height = (6*0.618),out.width ="85%", collapse =TRUE,dev ="png", dev.args =list(type ="cairo"))options(digits =3, width =120,dplyr.summarise.inform =FALSE,knitr.kable.NA ="")``````{r load-libraries, warning=FALSE, message=FALSE}library(tidyverse)library(targets)library(DT)library(kableExtra)library(naniar)tar_config_set(store = here::here('_targets'),script = here::here('_targets.R'))# Generated via random.orgset.seed(376)# Load targets stuffinvisible(list2env(tar_read(graphic_functions), .GlobalEnv))invisible(list2env(tar_read(misc_funs), .GlobalEnv))# Background stufftar_load(c(democracies, skeleton))# Aid stufftar_load(c(aid_donors, aid_recipients, aid_purposes, aiddata_final, usaid_clean, donor_level_data))# NGO laws stufftar_load(c(dcjw_clean, chaudhry_raw, chaudhry_clean))# V-Dem stufftar_load(c(vdem_clean, autocracies))# Other datatar_load(c(wdi_clean, un_pop, un_gdp, ucdp_prio_clean, disasters_summarized))# Final datatar_load(c(country_aid, country_aid_complete, country_aid_final, country_aid_no_lags))```# Create country-year skeletonWe use [Gleditsch-Ward country codes](http://ksgleditsch.com/data-4.html) to identify each country across the different datasets we merge. We omit a bunch of things though:- We omit microstates- Because the World Bank doesn't include it in the WDI, we omit Taiwan (713). We also omit East Germany (265) and South Yemen (680)- We only use the `r chaudhry_raw %>% distinct(gwcode) %>% nrow()` countries in Suparna's anti-NGO law data- Following @FinkelPerez-LinanSeligson:2007 (p. 414), we exclude long-term consolidated democracies from our analysis. These are classified by the World Bank as high income; they score below 3 on Freedom House's Scale, receive no aid from USAID, and are not newly independent states:```{r show-democracies, indent=" "}democracies %>%pull(country_name) %>%matrix_from_vector(ncol =5) %>%kbl() %>%kable_styling(bootstrap_options =c("condensed", "striped"), full_width =FALSE)```To get consistency in country codes, we do this:- When converting GW codes to COW codes, [following Gleditsch and Ward](https://www.andybeger.com/states/articles/differences-gw-cow.html), we treat post-2006 Serbia as 345 (a continuation of Serbia & Montenegro). And we also treat Serbia as a continuation of Yugoslavia with 345 (following V-Dem, which does that too).- In both COW and GW codes, modern Vietnam is 816, but `countrycode()` thinks the COW code is 817, which is old South Vietnam (see [issue](https://github.com/vincentarelbundock/countrycode/issues/16)), so we use `custom_match` to force 816 to recode to 816.- Also, following V-Dem, we treat Czechoslovakia (GW/COW 315) and Czech Republic (GW/COW 316) as the same continuous country (V-Dem has both use ID 157).Thus, here's our actual panel skeleton:We have `r country_aid_final %>% distinct(gwcode) %>% nrow()` countries in our final data, spanning `r country_aid_final %>% distinct(year) %>% nrow()` possible years. Here's a lookup table of all the countries included:::: {.dt-small}```{r show-final-countries}skeleton$skeleton_lookup %>%select(-years_included) %>%datatable(class ="compact")```:::# Foreign aid## OECD and AidDataThe OECD collects detailed data on all foreign aid flows (ODA) from OECD member countries (and some non-member countries), multilateral organizations, and the Bill and Melinda Gates Foundation (for some reason they're the only nonprofit donor) to all DAC-eligible countries (and some non non-DAC-eligible countries). The OECD tracks all this in a centralized Creditor Reporting System database and provides a nice front end for it at [OECD.Stat](http://stats.oecd.org/) with an open (but inscrutable) API ([raw CRS data](http://stats.oecd.org/DownloadFiles.aspx?HideTopMenu=yes&DatasetCode=CRS1) is also available). There are a set of pre-built queries with information about ODA flows by donor, recipient, and sector (purpose), but the pre-built data sources do not include all dimensions of the data. For example, [Table DAC2a](http://stats.oecd.org/Index.aspx?DataSetCode=TABLE2A) includes columns for donor, recipient, year, and total ODA (e.g. the US gave \$X to Nigeria in 2008) , but does not indicate the purpose/sector for the ODA. [Table DAC5](http://stats.oecd.org/Index.aspx?DataSetCode=TABLE5) includes columns for the donor, sector, year, and total ODA (e.g. the US gave \$X for education in 2008), but does not include recipient information. Instead of using these pre-built queries or attempting to manipulate their parameters, it's possible to use the [OECD's QWIDS query builder](https://stats.oecd.org/qwids/) to create a custom download of data. However, it is slow and clunky and requires significant munging and filtering after exporting. The solution to all of this is to use [data from AidData](http://aiddata.org/aiddata-research-releases), which imports raw data from the OECD, cleans it, verifies it, and makes it freely available on GitHub.AidData offers multiple versions of the data, including a full release, a thin release, aggregated donor/recipient/year data, and aggregated donor/recipient/year/purpose data. For the purposes of this study, all we care about are ODA flows by donor, recipient, year, and purpose, which is one of the ready-made datasets. Notably, this aggregated data shows total aid *commitments*, not aid *disbursements*. Both types of ODA information are available from the OECD and it's possible to get them using OECD's raw data. However, [AidData notes](http://aiddata.org/faqs-about-our-data) that disbursement data is sticky and slow—projects take a long time to fulfill and actual inflows of aid in a year can be tied to commitments made years before. Because we're interested in donor reactions to restrictions on NGOs, any reaction would be visible in the decision to commit money to aid, not in the ultimate disbursement of aid, which is most likely already legally obligated and allocated to the country regardless of restrictions.So, we look at ODA *commitments*.::: {.panel-tabset}### Summary of clean data```{r aiddata-summary}#| code-fold: showaiddata_final %>%glimpse()```### List of donors::: {.dt-small}```{r show-donors}aid_donors %>%datatable(class =c("compact"))```:::### List of recipients::: {.dt-small}```{r show-recipients}#| class-output: dt-smallselect(aid_recipients, recipient) %>%datatable(class =c("compact"))```:::### List of purposes::: {.dt-small}```{r show-purposes}arrange(aid_purposes, desc(n)) %>%datatable(class =c("compact"))```::::::## USAIDUSAID provides the complete dataset for its [Foreign Aid Explorer](https://explorer.usaid.gov/aid-dashboard.html) as a [giant CSV file](https://explorer.usaid.gov/data.html). The data includes both economic and military aid, but it's easy to filter out the military aid. Here we only look at obligations, not disbursements, so that the data is comparable to the OECD data from AidData. The data we downloaded provides constant amounts in 2015 dollars; we rescale that to 2011 to match all other variables. ::: {.panel-tabset}### Summary of clean data```{r usaid-summary}#| code-fold: showusaid_clean %>%glimpse()```### Implementing/managing agenciesHere are the US government agencies giving out money:::: {.dt-small}```{r show-agencies}implementing_agencies <- usaid_clean %>%count(managing_agency_name, managing_sub_agency_or_bureau_name) %>%arrange(desc(n), managing_agency_name)implementing_agencies %>%datatable()```:::### ActivitiesThe activities listed don't follow any standard coding guidelines. There are tens of thousands of them. Here are the first 100, just for reference:::: {.dt-small}```{r show-activities}activities <- usaid_clean %>%count(activity_name) %>%slice(1:100)activities %>%datatable()```:::### Channels / implementing partnersUSAID distinguishes between domestic, foreign, and international NGOs, companies, multilateral organizations, etc. implementing partners (formerly called channels):::: {.dt-small}```{r show-channels}channels <- usaid_clean %>%count(implementing_partner_category_name, implementing_partner_sub_category_name) %>%filter(!is.na(implementing_partner_category_name))channels %>%datatable(options =list(pageLength =20))```::::::# NGO regulations## Chaudhry lawsIn 2013, Darin Christensen and Jeremy Weinstein collected [detailed data](https://darinchristensen.com/publication/defunding-dissent/) on NGO regulations for [their *Journal of Democracy* article](http://www.journalofdemocracy.org/article/defunding-dissent-restrictions-aid-ngos), covering `r dcjw_clean %>% distinct(gwcode) %>% nrow()` countries.Suparna Chaudhry expanded this data substantially (it now covers `r chaudhry_raw %>% distinct(gwcode) %>% nrow()` countries and goes to 2013), so we use that.### Notes on year coverageIn our original paper from 2017, we used Suparna's data and backfilled it to 1980, since going back in time is possible with the DCJW data—lots of the entries in DCJW include start dates of like 1950 or 1970. Accordingly, our analysis ranged from 1980-2013. However, not all of Suparna's expanded countries went back in time that far, and she focused primarily on 1990+ changes. Additionally—and more importantly—the whole nature of foreign aid and civil society changed drastically after the Cold War. Civil society regulations weren't really used as a political strategy until after 1990. We can confirm that by plotting V-Dem's core civil society index:```{r vdem-civil-society-index}vdem_clean %>%filter(year >=1980) %>%select(year, v2xcs_ccsi) %>%group_by(year) %>%summarize(avg_ccsi =mean(v2xcs_ccsi)) %>%ggplot(aes(x = year, y = avg_ccsi)) +geom_line() +geom_vline(xintercept =1990, color ="red") +labs(x ="Year", y ="Average Core Civil Society Index", caption ="Source: V-Dem's v2xcs_ccsi") +theme_donors()```Something systematic happened to civil society regulations worldwide in 1990, and rather than try to model pre-Cold War regulations, which were connected to foreign aid in completely different ways than they were after the dissolution of the USSR, we limit our analysis to 1990+We still collect as much pre-1990 data as possible for the sake of (1) lagging, so we can get lagged values from 1989 and 1988 when looking at lagged variables in 1990, and (2) robustness checks that we run using the 98 backfilled DCJW countries.### Index creationWe create several indexes for each of the categories of regulation, following Christensen and Weinstein's classification:- `entry` (Q2b, Q2c, Q2d; 3 points maximum, actual max = 3 points maximum): barriers to entry - Q2c is reversed, so not being allowed to appeal registration status earns 1 point. - Q2a is omitted because it's benign- `funding` (Q3b, Q3c, Q3d, Q3e, Q3f; 5 points maximum, actual max = 4.5): barriers to funding - Q3a is omitted because it's benign - Scores that range between 0–2 are rescaled to 0–1 (so 1 becomes 0.5)- `advocacy` (Q4a, Q4c; 2 points maximum, actual max = 2): barriers to advocacy - Q4b is omitted because it's not a law - Scores that range between 0–2 are rescaled to 0–1 (so 1 becomes 0.5)- `barriers_total` (10 points maximum, actual max = 8.5): sum of all three indexesThese indexes are also standardized by dividing by the maximum, yielding the following variables:- `entry_std`: 1 point maximum, actual max = 1- `funding_std`: 1 point maximum, actual max = 1- `advocacy_std`: 1 point maximum, actual max = 1- `barriers_total_std`: 3 points maximum, actual max = 2.5### Chaudhry 2020 dataThe most recent version of Suparna's data is already in nice clean panel form, so it's super easy to get cleaned up.```{r show-chaudhry-ngos}#| code-fold: showglimpse(chaudhry_clean)```### DCJW dataFor fun and robustness checks, we use DCJW's non-panel data to generate a panel starting in 1980, since they have entries where laws start in the 1960s and 70s and other pre-1980 years. ```{r show-dcjw-ngos}#| code-fold: showglimpse(dcjw_clean)```All clean! Except not! NEVER MIND TO ALL THAT ↑Suparna made updates to existing the DCJW countries too, like Honduras (gwcode 91), which has more correct values for `q4a`, for instance, which DCJW marks as 0, but is actually 1. So even though we can go back in time to 1980 with DCJW, it's not comparable with Suparna's expanded and more recent data. ```{r check-honduras}#| code-fold: show# Look at Honduras in 1990 in both datasets:dcjw_clean %>%filter(year ==1990, gwcode ==91)chaudhry_clean %>%filter(year ==1990, gwcode ==91)```So we live with *just* 1990+, even for the sake of lagging `r emoji::emoji("shrug")`.Except, we're not quite done yet!In Suparna's clean data, due to post-Cold War chaos, Russia (365) is missing for 1990-1991 and Serbia/Serbia and Montenegro/Yugoslavia (345) is missing every thing pre-2006. DCJW don't include any data for Serbia, so we're out of luck there—we're limited to Serbia itself and not past versions of it. DCJW *do* include data for Russia, though, so we use that in our clean final NGO laws data. Fortunately this is easy, since Russia's values are all 0 for those two years:```{r show-russia}#| code-fold: showdcjw_clean %>%filter(gwcode ==365, year %in%c(1990, 1991))```# Varieties of Democracy (V-Dem)## Regulatory environmentAn alternative way of measuring civil society restrictions is to look at the overall civil society regulatory environment rather than specific laws, since de jure restrictions do not always map clearly into de facto restrictions (especially in dictatorships where the implementation of laws is more discretionary). Andrew Heiss develops a new civil society regulatory environment index (CSRE) in his dissertation, which combines two civil society indexes from the [Varieties of Democracy project (V-Dem)](https://www.v-dem.net/en/): (1) civil society repression (`v2csreprss`) and (2) civil society entry and exit regulations (`v2cseeorgs`). The CSRE ranges from roughly −6 to 6 (though typically only from −4 to 4ish), and shows more variation over time since it ostensibly captures changes in the implementation of the regulatory environment rather than the presence or absence of legislation.Additionally, since Andrew's dissertation, the V-Dem project has created its own core civil society index (`v2xcs_ccsi`) (entry/exit (`v2cseeorgs`) + repression (`v2csreprss`) + participatory environment (`v2csprtcpt`)). We use that instead, since V-Dem rescales it to a 0-1 scale so it's not weirdly distributed from like my additive two-factor CSRE index.While the main focus of this paper is donor response to new *legislation*, we also look at donor response to changes in the overall civil society index as a robustness check. This also allows us to potentially include data from 1980–2018 (since Suparana's law data ranges from 1990–2014).## Other variablesWe also use a bunch of other V-Dem variables as confounders:- **Civil society things** - CSO entry and exit: `v2cseeorgs` - CSO repression: `v2csreprss` - CSO consultation: `v2cscnsult` - CSO participatory environment: `v2csprtcpt` - CSO women's participation: `v2csgender` - CSO anti-system movements: `v2csantimv` - Core civil society index (entry/exit, repression, participatory env): `v2xcs_ccsi`- **Human rights and politics** - Polity scores: `e_polity2` (only for reference with polyarchy; we don't use these in models) - Electoral democracy index: `v2x_polyarchy` - Regimes of the world scores: `v2x_regime_amb` (only for determining average autocracy; we don't use these in models) - Political corruption index: `v2x_corr` (less to more, 0-1) (public sector + executive + legislative + judicial corruption) - Rule of law index: `v2x_rule` - Civil liberties index: `v2x_civlib` - Physical violence index: `v2x_clphy` - Private civil liberties index: `v2x_clpriv` - Political civil liberties index: `v2x_clpol` (but not this because it includes `v2cseeorgs` and `v2csreprss`)- **Economics and development** - Educational equality: `v2peedueq` - Health equality: `v2pehealth` - Infant mortality rate: `e_peinfmor````{r show-vdem}#| code-fold: showglimpse(vdem_clean)```## AutocraciesWe're also interested in how these civil society dynamics work in autocracies in particular. We generate a crude list of autocracies based on average "Regimes of the World" scores (for ambiguous cases) from V-Dem. In that measure, 4 is the upper bound and 5 is the lower bound for electoral autocracy, so we use 4 as the cutoff. There are bound to be better ways, but this works for now.```{r show-autocracies}#| code-fold: showautocracies %>%filter(autocracy) %>%glimpse()```# World Bank development indicatorsWe don't really use anything from the World Bank's data except for population data for Kosovo.```{r show-wdi}#| code-fold: showglimpse(wdi_clean)```# UNDataThe reason we don't just use WDI data for GDP and % of GDP from trade is that the WDI data is incomplete, especially pre-1990. To get around that, we create our own GDP and trade measures using data directly from the UN (at [UNData](https://data.un.org/)). They don't have a neat API like the World Bank, so you have to go to their website and export the data manually.We collect three variables: [GDP at constant 2015 prices](http://data.un.org/Data.aspx?q=gdp&d=SNAAMA&f=grID%3a102%3bcurrID%3aUSD%3bpcFlag%3a0), [GDP at current prices](http://data.un.org/Data.aspx?q=gdp&d=SNAAMA&f=grID%3a101%3bcurrID%3aUSD%3bpcFlag%3a0), and [population](https://population.un.org/wpp/Download/Standard/Population/).```{r show-un-pop}#| code-fold: showglimpse(un_pop)```Dealing with the GDP data is a little trickier because it's in 2015 dollars, while AidData is in 2011 dollars. To fix this, we create a GDP deflator and rebase the 2015 values to 2011 values.```{r show-un-gdp}#| code-fold: showglimpse(un_gdp)```# UCDP/PRIO Armed ConflictThe [UCDP/PRIO Armed Conflict Dataset](http://ucdp.uu.se/downloads/) tracks a ton of conflict-releated data, including reasons for the conflict, parties in the conflict, intensity of the conflict, and deaths in the conflict. We're only interested in whether a conflict happened in a given year (or in the past 5 years), so here we simply create an indicator variable for whether there was internal conflict in a country-year (conflict type = 3). ```{r show-ucdp-prio}#| code-fold: showglimpse(ucdp_prio_clean)```# Natural disastersNatural disaster data comes from the [International Disaster Database (EM-DAT)](http://www.emdat.be/database). The data includes the number of deaths, injuries, homeless displacements, and monetary losses (in 2000 dollars) for a huge number of natural and technological disasters (see [EM-DAT's full classification](http://www.emdat.be/classification)).Natural disasters could matter for aid too, since donor countries might increase their aid to countries suffering more. EM-DAT does not provide a single link to download their data. Instead, you have to create a query using [their advanced search form](http://www.emdat.be/advanced_search/index.html). We downloaded data using the following query:- Select all countries from 1950–2021- Select all three disaster classification groups (natural, technological, complex)- Download Excel file and save in `Data/raw_data/Disasters/````{r show-disasters}#| code-fold: showglimpse(disasters_summarized)```# Clean combined dataWith both donor- and country-level data, we have lots of different options for analysis. Since our hypotheses deal with questions of *donor* responses, the data we use to model donor responses uses donor-years as the unit of observation. Not all donors give money to the same countries, so this final data is not a complete panel (i.e. it does not include every combination of donors and years), which will pose some interesting methodological issues when modeling if we use donor-level data.## InflationUSAID's conversion to constant 2015 dollars doesn't seem to take country differences into account—the deflator for each country in 2011 is essentially 96.65. When there are differences, it's because of floating point issues (like, if there are tiny grants of $3, there aren't enough decimal points to get the fraction to 96.65). So we just take the median value of the deflator for all countries and all grants and use that.## Missing dataThe donor data is complete with no missing variables(!).```{r final-data-missing-donors, fig.width=6, fig.height=3}gg_miss_var(donor_level_data, show_pct =TRUE) +theme_donors()```The country-level panel data is relatively complete, with only a few variables suffering from missing data, mostly from the World Bank and V-Dem. There are a lot of NGO-related missing variables, but that's because we don't have data from 1980–1989 and 2015+```{r data-missing-countries, fig.width=6, fig.height=7}gg_miss_var(country_aid, show_pct =TRUE)``````{r data-missing-countries-sans-laws, fig.width=6, fig.height=7}country_aid %>%select(-starts_with("funding"), -starts_with("entry"), -starts_with("advocacy"), -starts_with("barriers"),-starts_with("ngo_")) %>%gg_miss_var(., show_pct =TRUE) +theme_donors()```Here's how we address that:- We remove everything from Yugoslavia/Serbia and Montenegro (345) prior to 2006- Infant mortality `e_peinfmor` is missing from Kosovo (2008–2014), and the World Bank doesn't have data for it, but Eurostat does in [their `demo_minfind` indicator](https://ec.europa.eu/eurostat/databrowser/view/DEMO_MINFIND__custom_660664/default/table?lang=en). Their data, however, is missing a couple years```{r show-kosovo-infant-mort, indent=" ", collapse=FALSE}kosovo_infant_mort <-tibble(year =2007:2019,e_peinfmor =c(11.1, 9.7, 9.9, 8.8, 13.1, 11.4, NA, NA, 9.7, 8.5, 9.7, 10.6, 8.7))kosovo_infant_mort``` To fix this, we use linear interpolation to fill in 2013 and 2014.- `v2x_corr` is only missing data from Bahrain, which oddly has no data from 1980–2004. Because corruption levels do not really change after 2005, we impute the average corruption for the country in all previous years.- `v2x_polyarchy` is only missing in Mozambique from 1980–1993. To address this, we calculate the average value of V-Dem's polyarchy index (`v2x_polyarchy`) for each level of Polity (−8, −7, and −6 in the case of Mozambique), and then use that corresponding average polyarchy- We also create an `imputed` column for those rows in Bahrain and Mozambique to see if imputation does anything weird in the models- **Kosovo is missing pre-existence infant mortality, which is fine becuase it didn't exist yet.**- **Russia is missing GDP, GDP per capita, and percent of GDP from trade from 1980–1989.** There's no easy way around this. V-Dem has GDP per capita data from the long-running [Maddison Project Database](https://www.rug.nl/ggdc/historicaldevelopment/maddison/releases/maddison-project-database-2018), and it includes 1980s Soviet Russia, but the values aren't really comparable to the stuff we calculated using UN GDP data. At first glance it seems that this is a difference in real years, since the Maddison Project uses 2011 dollars and the UN uses 2015 dollars, and there's not an easy way to shift the Maddison Project's values up to 2015 (i.e. there's no deflator). But even if they were in the same dollar-years, the values from the Maddison Project seem really really low compared to what we made with the UN GDP data, so they don't seem to be comparable.- **Czechoslovakia is missing percent of GDP from trade from 1980–1989.** This is because it is missing imports data in the UN GDP data. It has exports data and overall GDP data, but for whatever reason, imports are missing. Boo.```{r missing-soviet-things, indent=" ", collapse=FALSE}country_aid_complete %>%select(gwcode, country, year, un_trade_pct_gdp, un_gdp, gdpcap, gdpcap_log) %>%filter(is.na(un_trade_pct_gdp))```Since those issues are all pre-1990, our data is perfect post-1990 in cases with Suparna's law coverage:```{r final-data-missing-countries, fig.width=6, fig.height=7}country_aid_no_lags %>%filter(laws) %>%gg_miss_var(., show_pct =TRUE) +theme_donors()```We do get some missing values when lagging, though, but that only happens in countries that emerged post-1990 (like Kosovo, South Sudan, etc.), since there are no 80s-era values to look at for lagging## Final dataNow that we know all the data is clean and pretty much nothing is missing, we can do a few final windowed operations that will add missing values (e.g. lagging). We also add an indicator marking if a disaster happened in the past 5 years.In H~3~ we hypothesize that more aid will be allocated to international or US-based NGOs than domestic NGOs in response to harsher anti-NGO restrictions. While AidData unfortunately does not categorize aid by channel (i.e. aid given to international vs. US vs. domestic NGOs), USAID does. For this hypothesis, then we only look at aid given by USAID, not the rest of the OECD. As with the proportion of contentious aid, we create similar variables to measure the proportion of aid given to international NGOs, US-based NGOs, and both international and US-based NGOs.```{r show-final-data}#| code-fold: showglimpse(country_aid_final)```# References
