| (1) | (2) | |
|---|---|---|
| Estimates are median posterior log odds from ordered Beta regression models; 95% credible intervals (highest density posterior interval, or HDPI) in brackets. | ||
| Intercept | -0.22 | -0.22 |
| [-0.83, 0.38] | [-0.84, 0.43] | |
| Issue [Arts and culture] | 1.28 | 1.29 |
| [0.49, 2.10] | [0.47, 2.14] | |
| Issue [Education] | -1.12 | -1.14 |
| [-1.80, -0.47] | [-1.79, -0.45] | |
| Issue [Industry association] | 0.46 | 0.46 |
| [-0.11, 1.04] | [-0.12, 1.06] | |
| Issue [Economy and trade] | -0.17 | -0.17 |
| [-0.72, 0.39] | [-0.73, 0.45] | |
| Issue [Charity and humanitarian] | -0.27 | -0.26 |
| [-0.85, 0.33] | [-0.89, 0.28] | |
| Issue [General] | 0.08 | 0.073 |
| [-0.51, 0.70] | [-0.565, 0.704] | |
| Issue [Health] | -0.23 | -0.24 |
| [-0.89, 0.42] | [-0.89, 0.41] | |
| Issue [Environment] | 0.27 | 0.25 |
| [-0.38, 0.87] | [-0.37, 0.90] | |
| Issue [Science and technology] | 0.28 | 0.27 |
| [-0.40, 0.92] | [-0.40, 0.96] | |
| Local connections | -1.7 | -1.9 |
| [-2.1, -1.2] | [-2.7, -1.1] | |
| Years since law | 0.13 | 0.13 |
| [-0.26, 0.50] | [-0.24, 0.50] | |
| Year registered [2018] | -0.20 | -0.19 |
| [-0.60, 0.26] | [-0.62, 0.25] | |
| Year registered [2019] | -0.31 | -0.29 |
| [-1.21, 0.49] | [-1.18, 0.50] | |
| Year registered [2020] | -0.42 | -0.42 |
| [-1.54, 0.93] | [-1.64, 0.78] | |
| Year registered [2021] | -0.073 | -0.13 |
| [-1.603, 1.591] | [-1.57, 1.58] | |
| Local connections × years since law | 0.10 | |
| [-0.21, 0.40] | ||
| φ | 2.2 | 2.2 |
| [1.8, 2.6] | [1.8, 2.6] | |
| Num.Obs. | 593 | 593 |
| R2 | 0.202 | 0.203 |
Online appendix for “Enforcing Boundaries: China’s Overseas NGO Law and Operational Constraints for Global Civil Society”
Model details and results
We use Stan (Stan Development Team 2023b, v2.26.1; 2023a, v2.3.1) through R (R Core Team 2023, v4.3.1) and brms (Bürkner 2017, v2.20.5) to estimate our ordered Beta regression models (Kubinec 2022). We simulate 4 MCMC chains with 2,000 draws in each chain, 1,000 of which are used for warmup, resulting in 4,000 (1,000 × 4) draws per model parameter. We assess convergence with visual inspection, and all chains converge. Complete results from all the models, along with posterior predictive checks, goodness-of-fit measures, and model diagnostics—as well as our code and data—are available ANONYMIZED_URL. We include the formal definition and priors for our model below.
\[ \begin{aligned} &\ \textbf{Registered provinces for INGO } i \\ \text{Count of provinces}\ \sim&\ \operatorname{Ordered\,Beta}(\mu_i, \phi_y, k_{1_y}, k_{2_y}) \\[8pt] &\ \textbf{Model of outcome average} \\ \mu_i =&\ \mathrlap{\begin{aligned}[t] & \beta_0 + \beta_1\ \text{Issue[Arts and culture]} + \beta_2\ \text{Issue[Education]}\ +\\ & \beta_3\ \text{Issue[Industry association]} + \beta_4\ \text{Issue[Economy and trade]}\ + \\ & \beta_5\ \text{Issue[Charity and humanitarian]} + \beta_6\ \text{Issue[General]}\ + \\ & \beta_7\ \text{Issue[Health]} + \beta_8\ \text{Issue[Environment]}\ + \\ & \beta_9\ \text{Issue[Science and technology]} + \beta_{10}\ \text{Local connections}\ + \\ & \beta_{11}\ \text{Time since January 2017} + \beta_{12}\ \text{Year registered} \end{aligned}}\\[8pt] &\ \textbf{Priors} \\ \beta_0\ \sim&\ \operatorname{Student\,t}(\nu = 3, \mu = 0, \sigma = 2.5) && \text{Intercept} \\ \beta_{1..12}\ \sim&\ \mathcal{N}(0, 5) && \text{Coefficients} \\ \phi_y\ \sim&\ \operatorname{Exponential}(1 / 100) && \text{Variability in province count} \\ k_{1_y}, k_{2_y}\ \sim&\ \operatorname{Induced\,Dirichlet}(1, 1, 1), && \text{0–continuous and continuous–1 cutpoints} \\ &\ \quad \text{or } \bigl[P(\alpha_1), P(\alpha_1 + \alpha_2)\bigr] && \quad\text{(boundaries between 3 Dirichlet columns)} \end{aligned} \]
Coding rules for issue ares
Author 1 assigned each organization to one of the issue areas defined in Article 3 of the ONGO Law, and Author 2 cross-checked the coding for ambiguous cases. Organizations with multiple ROs are coded based on the overall issue area of the INGO, not the specific RO, so to ensure the issue area coding is consistent across ROs.
We followed these rules for categorizing INGOs’ stated missions into issue areas:
- Education vs. Charity and humanitarian: If charitable activities are all education related (e.g., building schools, donating education facilities and books), code as Education. If activities include other community support (e.g., building bridges, roads, medical facilities), code as Charity and humanitarian.
- Economy and trade vs. Industry association: If the trade promoted is focused on a particular industry (e.g., poultry, grains, etc.), code as Industry association.
- Arts and culture: Include sports organizations.
- Industry association vs. Health & Science and technology & Arts and culture: There are professional associations for health providers, scientific associations, and artistic industries. If activities go beyond only serving members only, code as Health or Science and technology or Arts and culture; if activities are limited to member-only service, code as Industry association.
- Health vs. Charity and humanitarian: If activities only provide medical assistance to underprivileged communities, code as Health.
- Charity and humanitarian vs. General: If there are overlapping issue areas (e.g., poverty alleviation, education, health), code as General, then code the apparent primary issue area as a second work area.
Comparison of the cases of TNC and Greenpeace
Table 2 summarizes the cases of TNC and Greenpeace. In addition to confirming the correlation between issue area and operational space found in the statistical model, the cases highlight how government preferences exert influence on both the degree of contentious programming INGOs are allowed to undertake and the severity of legal restrictions INGOs can face.
| The Nature Conservancy | Greenpeace | |
|---|---|---|
| Issue area | Environmental | Environmental |
| Presence in China | Office in Beijing since 1998 | Office in Beijing since 2001 |
| Home country | United States | The Netherlands |
| Legal status after ONGO Law | Became RO on November 17, 2017; approved to work in 27 provinces | Not registered as RO; has filed 67 temporary activities since June 2017 |
| Work approach | Technical, non-confrontational, science-based focus | Advocacy and actions, peaceful protests, and creative confrontation |
| Relationship with government | Partnership, technical consultancy | Independent, no permanent allies or enemies |
| Example program | Provided policy makers with technical support to develop the China National Biodiversity Conservation Strategy and Action Plan (2011–2030) | Monitored and released independent investigation reports on the damages of Sinar Mas Group's projects to natural forests in southern Yunnan |
References
Bürkner, Paul-Christian. 2017. “brms : An R Package for Bayesian Multilevel Models Using Stan.” Journal of Statistical Software 80 (1). https://doi.org/10.18637/jss.v080.i01.
Kubinec, Robert. 2022. “Ordered Beta Regression: A Parsimonious, Well-Fitting Model for Continuous Data with Lower and Upper Bounds.” Political Analysis, 1–18. https://doi.org/10.31235/osf.io/2sx6y.
R Core Team. 2023. “R: A Language and Environment for Statistical Computing.” Vienna, Austria: R Foundation for Statistical Computing. https://www.r-project.org/.
Stan Development Team. 2023a. “CmdStan: The Shell Interface to Stan.” https://mc-stan.org/users/interfaces/cmdstan.
———. 2023b. “Stan Modeling Language.” https://mc-stan.org.