The Causal Effect of Mountain Partnership on the Environment

METHODOLOGY

In our study, we try to assess the causal effect of the MPGA on greenhouse gas emissions by using a matching analysis technique proposed by17, which consists of a non-parametric generalisation of the difference-in-differences estimator. The details in form of shorter vignette of the method can be find in [18 ]. We designated from an appropriate tailored dataset accession to the MPGA during our period of analysis to be the treatment group. The environmental intervention referring to greenhouse gas footprint is indicated by accession to the MPGA. The signed application at governmental level signals accession to the MPGA. The collective action, we presumed, would somehow affect all of the members of the organisation due to the requirement of strict adherence to the MPGA rules about environmental policy. These rules include conservation, maintaining the health, vitality and stewardship of mountain ecosystems, as well as taking care of economic sustainability, green development, etc. All these were navigated and enforced by domestic laws, in function of better environmental protection attitudes; with verification and balancing, and supervising critical points by various monitoring institutions. These assumed measures are exogenous because they are instituted and imposed by the global strategy of MPGA, in order to provide long-run environmental sustainability among the members. Because these countries’ accessions to the MPGA were not randomly assigned, we generated a more accurate causal estimate through a matching technique across comparable country-level factors. Matching involves a systematic selection of similar cases (mapping similar countries) to approximate a counterfactual for each treated case. In other words, our analysis compares countries that have joined the MPGA that had potentially carried out some kind of environmental quality policy (treatment) to similar countries that did not institute the policy (the control group). Since panel data varies over time, matching allows control units to include both the time leading up to the “treatment” for a given country of interest as well as the period for similar observations of countries that were not treated. We estimate the causal effects of MPGA country accession, which generates contingent policies included in the dataset using the PanelMatch package in R written by18. These authors proposed that innovation builds upon the synthetic control method (SCM) of19 and the generalised SCM conceived by20, allowing a simple understanding of how counterfactual outcomes are assessed; a particular advantage of this method that stands out is it does not require data on many pre-treatment time periods and is more flexible than the generalised SCM. Namely, the non-experimental data in social sciences often have a limited time span; our data are rather of this kind. The set of environmental quality policy effects (the particular details of which is unknown, but we assume some kind of policy intervention has been undertaken after joining the MPGA) had to be quantified through a nonparametric generalisation of difference-in-differences, with T−4 serving as the baseline. A country was considered to be a treated case if the environment policy was adopted at a specified time as a consequence of the country’s accession to the MPGA; otherwise, it was considered to be an untreated case. Panel matching estimation would reduce bias compared to the standard two-way fixed effects (TWFE) regression commonly utilised in panel analysis, which was otherwise the second-choice methodology option. The aim was to compare the outcomes of interest for a sample of countries under the new policy regime, in fact the trend in GHG emissions after transition. The policy conditioned some kind of restrictions on carbon footprint. It is a by-product of assumed MPGA accession at a specific time, and should negatively affect GHG, as predicted. However, using the matching technique, we searched for the differences between the outcomes under the old and new policy regime (in a country with a similar covariate history) but it does not have to have occurred at the same time. For the country-year covariates, we did not select a time period as they do not vary across years, while for our time-varying covariates, we specify the selection of the period of analysis (T−4:T+10 ). We then calculate and compare different refinement strategies to assess the best among the few, which would minimise covariate imbalance during the pre-analysis period (T−4:T0). As the covariates on which refinement occurs, we specified a Kyoto Protocol dummy variable as well as an economic power variable (GDP per capita), tourism demand (international arrivals per capita) and demographic (population) control variables. Namely, we have a strong presumption that those variables had significant, relevant impact, considering their weight in the overall paper design, when it comes to air pollution or environmental degradation, as well as in processing our robust matching model. We assume that the Kyoto Protocol as a global voluntary instrument that regulates global environmental policy indicates a nation’s positive attitude and commitment to mitigating global-warming. It is an efficient indicator for identifying which countries strategically wish to curb GHG emissions; as a covariate it would formally mitigate the confounding of our basic treatment effect estimates, whereas other included covariates would interfere, as such, in the opposite direction. After the refined matched sets of treated countries are obtained, we calculate the counterfactual outcome (with non-treated countries) using the weighted average of control countries. We made a weight for each control unit in M_(i,t) ,e.g. in the refined matched set, where a higher weight had to be assigned to a more similar unit. In the next step, for each treated country, we estimated the counterfactual outcome in regard to GHG emissions using the weighted average of the control units in the refined matched set. For 𝐾 treated countries and 𝑇 observation years, we followed the estimator proposed by the authors (17): average treatment effect (ATT) on the treated countries in changing GHG emissions expressed as follows:

where outcome variable Y_(i,t), is the change in GHG emission per capita in country 𝑖 in year 𝑡, and is an indicator for a treatment Xit set of countries with non-empty matching that includes countries which joined the MPGA in a certain year interval prior to time point 0. To compute the standard error of estimator 𝐴𝑇𝑇(𝐹, 𝐿), and to derive significance out of the transiting direction in ATT, we used a block-bootstrap procedure designed for matching with time series cross sectional (TSCS) data on the weight21. The leads (F) and lags (L) next to ATT in parentheses at the left side of the equation denotes identical country history regarding cofounders from time t –L to t –1 by which algorithm process outcome is measured at time t + F. The outcome of interest would be the change in new, reported GHG emission per capita from year t to year t + F, that has occurred in the interval from year 0 to 10. The number of bootstrap iterations is set to 10 000. With respect to timing, we consider that a country will have adjusted to a new environmental policy at least one year after accession to the MPGA. How a country’s regulatory policy in regard to GHG emissions actually impacts a treatment group, that is selected countries in the MPGA, explained above in the presented thesis, will be discussed in the subsequent sections.

EMPIRICAL ANALYSIS

First, we drew the plot that shows the number of matched control units that share the same treatment history as a treated observation which originates from accession to the MPGA, assigned as a treated observation in the case of four year lag. The plot given in Figure 1 shows the number of matched control units that share the same treatment history as a treated observation. We see that there are 14 treated observations for administered policy that have about 90 control units with the same treatment history when the number of lags is four. So, we have enough matched control units to test the MPGA transmitted policy along the GHG trajectories: most treated observations have more than enough matched control units.

In a next step, in order to refine the matched set further, and having in mind that one of the weighting techniques had to be chosen as the best (among Mahalanobis distance matching, propensity score matching, and propensity score weighting), we proceed. So, we compare the performance of each refinement method after it had generated a new matched set. Following a result assessment, we keep the propensity score weighting refinement results as it shows to be the optimal matched set. In this auxiliary part of our panel matching research, while searching for the optimal set, all of the previously mentioned time-varying covariates were implemented e.g. indicators for economic, tourism and demographic forces, as well as a binary type of variable for the Kyoto Protocol label. The PSW refinement in our exercises essentially eliminates almost all imbalance, the standardised mean difference distance in all confounders of continuing type (yet the Kyoto covariate which is a binary type variable is somehow problematic) are minimised, e.g. has been kept to around 0. Although some degree of imbalance remains for this type of improved matching, the standardised mean difference for the lagged outcomes (CO2, CH4 and N2O, all per capita) stay rather constant over time, this is a substantial difference as before, the entire pre-treatment period. This suggests that the assumption of a parallel trend for the proposed difference-in-difference estimator is valid. This encourages us to deliberately proceed with the analysis. We now present the estimated ATTs based on the matching methods. Subsequent figures show the matching estimates of the effects of administering environmental policy which induce the act of accession to the MPGA, each for various GHG per capita types for the period of ten years after the transition point of accession to the MPGA, i.e., F = 0, 1, ..., 10. Regarding the matching set obtained by the propensity score weighting (PSW) refinement method, and when the dependent variable of interest is set up to be CO2_pc, we assessed that the point estimates of the effects for accession to the MPGA are mostly close to zero in the year of accession (see Figure2). While moving forward on a year-by-year basis, the CO2_pc variable, as an outcome, transit more and more in the intuitively expected direction. We show that in the 2nd year and future years the assumed treatment dynamics negatively impact on CO2 per capita emission. The ATT coefficient in the CO2_pc case becomes more significant and more prominently negative with the lapsing time. We report the same estimated treatment effects and confidence intervals in Table 3 (in the Appendix). It shows the estimated average treatment effect on the treated countries (ATT) when both the post-treatment (F) and pre treatment period L are set to be 10 and 4 years, respectively, along with standard error (SE) and 97,5 % confidence interval. These results suggest that the effect on CO2_pc is significant; namely, all point estimates certainly, with bootstrapped estimates, uncover negative values, after MPGA’s policy intervention. Starting from the first year, the constraints driven by policy directed to the MPGA members made substantial benefit in terms of CO2 _pc decreasing. After the turning point, which occurred in the 7th year of supposed treatment, when the average treatment effect estimate measured by CO2 emission dropped the most –by as much as 1624 kilotonnes per capita – in future years this volume in the atmosphere above treated countries changed slightly yet its volume remained always in the negative zone of values, denoting a general fall, as we had predicted. Note related to all figures 2,3 and 4: The average treatment effects are represented by point estimates over time (in years), with lines corresponding to standard errors obtained with 1000 bootstrap iterations. Here 2,5 % and 97,5 % are quantiles of the bootstrapped estimates reported. Propensity score weighting was used to create the matched sets. The Figure 3 shows the level of CH4_pc movement in the MPGA group after acceptance of mandatory environmental policy. The methane emissions per capita (CH4_pc), at first paradoxically increased following the implementation of environment-led protection policy covered by appropriate laws, but the difference was insignificant. The results indicate an expected decrease of methane per capita, as expected after the adoption of environmental policy, in the years after the second year. Yet, the bootstrapped estimates yield both positive and negative values in the entire interval, suggesting little evidence of a non-zero effect. Summarising the case of the CH4_pc trajectory path on the change due to MPGA association, we show that the effect sizes had a strange trajectory path, zigzagging negatively but were insignificant. Furthermore, in the 5th year, beside others, we underline that the causal impact became borderline significant. The fall in the emission of CH4_pc in that year, we assess to be about 0,177 kilotonnes. According to Figure 4 we cannot find sufficient evidence for our hypothesis that MPGA induced environmental policy would reduce N2O_pc emission in the treatment group of countries. Some of the treatment effect estimates are positive, more precisely in the years: four, five and seven) indicating that an assumed environmental law regulation coming into force causally increases emission of N2O_pc. The weak magnitude in the second year brought in effect the significant impacts. We also noticed positive impacts paradoxically unevenly scattered which are not otherwise negligible in the other observed years. In some of them their confidence intervals were zero.