Before we will move to our logistic regression analysis, we have to understand that we have one categorical depended variable called Projects_2 and nine independent variables called Elections, Regime, Governors, Resources, Innovations, RE_potential, El_prices, GSP and Population.

Table 1 displays the likelihood (probability) of the capacity based renewable energy support scheme projects (Projects_2) to be launched in Russian regions when there are different types of factors including socio-economic, political and electricity market ones. Here you can see 6 conducted logistic regression models with different combinations of factors. First, we start the analysis with the three control variables, then we add the El_prices, Innovations and Resources factors one at a time, and finally, we add all the political factors. All the models are statistically significant (see P-value). We can also see a trend that the more IVs are used in a model, the greater the DV is explained (see Pseudo R² ).

The first model where the simplest sketch of likelihoods between the DV and the control IVs such as RE_potential, GSP, and Population is made suggests an interesting result: first of all, we can clearly observe that only the RE potential in regions significantly influences the likelihood that the CRESS projects are implemented in regions. In other words, we can prove our sixth hypothesis, at the mean significant level, that the more RE sources are available in Russian regions, the higher the possibility is that the RE mechanism will be implemented. The power coefficient of such influence of the RE potential IV is close to 1 so that we can say that its power is sufficient enough to claim that the significance has the influence on the DV. Second of all, it is a pity, the other two control variables about regions' population growth and gross regional product are not statistically significant.

The second and third models do not bring any appreciable differences to the analysis after we added the electricity prices and innovations IVs. Both of them do not show any statistical significance, so, we have removed them not to spoil the further models. As for the RE potential IV, it maintains its positions with regard to the mean significance level and the power coefficient.

The fourth model shows tangible progress after we added the next IV - Resources. Speaking of which, it shows the just positive level of significance which theoretically tells us that the more fossil resources are available in Russian regions, the higher the likelihood is that the CRESS will be implemented. If we relied on the p-value indicator of statistical significance exclusively, we would have a ground to say that this IV's result disproves our second hypothesis. Nevertheless, we do not rely on p-values exclusively, we also consider the power coefficient of the influence which Resources IV has on the least level. It means that the likelihood that the more energy-endowed regions are more likely to implement the CRESS is unrealistic. We will take a look at this during a discussion of the analysis.

Two more points to consider in the fourth model. First, it is that regions' gross regional product IV becomes the just significant after we have added Resources to the analysis. It says that the Russian regions' public material capacities negatively influence the likelihood that the CRESS will be implemented in these regions. However, this IV comes across the same problem as Resources - its coefficient value is too low to consider this influence as realistic. Second, the level of significance of regions' RE potential decreases to the just significant when we have added the fossil resources abundance of Regions to the analysis.

The fifth model does not bring any new statistical results. We have added two political factors about elections and ENP and both of them are statistically insignificant. The same goes for the sixth model. Regions' fossil resources abundance, GSP, and renewable energy potential save their significance and power values.

The last sixth model differs from the fifth one only by the new-added IV - Governors. The level of connection of regions with the federal center has the least significant level with the negative value. In other words, it assumes us that the less level of regions' connection is with the federal center, the higher the chances are that the CRESS will be implemented in these regions. It is worth mentioning that the power indicator is yet extremely low to claim that the likelihood of this IV is realistic. However, on the theoretical level, we can say that this result disproves our sixth hypothesis. What is more, the GSP's significant level slightly decreases from the just significance to the least significant.

As we can see from the analysis, our concerns about the low dispersed data are affirmed. Despite the fact that all models are statistically significant, they show not as low indicator of the predictability capacities of the models (LogLike), we can barely see any independent variables which can go through the selection of the p-value and the coefficient indicators. The main reason of such results - the low frequency of our binary dependent variable where the positive results of the CRESS implementation in regions equals to 8% whereas their absence equals to 92%. That is why, now, we are confident enough to confirm that we have to apply the second type of regression analysis - the Poisson regression which specifically provides for the low frequency data.

Table 1. Standard errors in parentheses: ⁺ p < 0.1, ^* p < 0.05, ^** p < 0.01, ^*** p < 0.001

Logit models. Table 1: DV - categorical Projects_2

Source: author's calculations.

3.4 Poisson regression results

Table 2 displays the Poisson regression models with the same independent variables divided into the three groups: socio-economic, political and electricity market ones. All of the IVs are used as the possible predictors of the number of the CRESS projects implemented by Russian regions since 2013 till 2018. An increase (delta) in a number of RE projects in a region in contrast to a previous year is our dependent variable which IVs try to predict. As well as in the logistic regression, we will analyze a likelihood that an IV has an influence on the increase in the number of the CRESS projects.

First, we start the analysis with the three control variables such as RE_potential, GSP, and Population, then we add the El_prices, Innovations and Resources factors one at a time, and finally, we add all the political factors. All the six models are statistically significant (see P-value). We can also see a trend that the more IVs are used in a model, the greater the DV is explained (see Pseudo R² ).

The last aspect which needs to be mentioned before we move to the analysis is about the incidence rate ratio (IRR) which we are going to use to show each independent variable's rate (risk, possibility, likelihood). This indicator which we purposely calculate for each the IV shows, in other words, a “time at risk” that the CRESS projects will be implemented if there is one of the IVs in a region. The IRR is calculated as a number of events divided by the event of interest-year (the CRESS project) at time. "Incidence Rate Ratios And Incidence Rate Differences". 2020. Sphweb.Bumc.Bu.Edu. http://sphweb.bumc.bu.edu/otlt/MPH-Modules/PH717-QuantCore/PH717-Module3-Frequency-Association/PH717-Module3-Frequency-Association11.html. We list of the IRRs in Table 3.

The first model where the control variables are tested shows us an interesting result: first of all, we can see that regions' RE potential is a significant predictor of the CRESS projects. At the very significant level, the sixth hypothesis is proved that the more RE potential is available in Russian regions, the higher the likelihood is that the CRESS will be implemented. The power of the RE potential predictability is extremely high, so we can conclude that this likelihood is realistic. What is more, Table 3 tells us that the IRR of the RE potential indicates that for every one unit increase on the renewable energy potential, the possibility of the number of CRESS projects in a region increased by factor of 1.479 - more than 100%.

Secondly, the first model illustrates that the population growth of Russian regions has the negative influence on the increase of the CRESS projects in region at the least significant level. It contradicts to our ninth hypothesis where the theory suggests that the greater the growth of a population in a region, the higher the chances that the CRESS will be implemented. The power coefficient is not as low, so we can conclude that the influence on the likelihood of the CRESS implementation of Population is more or less realistic. Table 3 also tells us that the IRR of the population growth indicates that for every one unit decrease on the population growth IV, the possibility of the increase of CRESS projects in a region increased by factor of 1.203 - more than 100%. The GSP IV does not show any statistical significance neither in the first model, nor in the second and third.

The second and third models do not reveal any changes after we added electricity prices and innovations IVs. The significance level and the power coefficient of the RE potential and population remain the same.

However, when we added the Resources IV in the fourth model, we notice a change. The RE potential saves its significance but lowers its predictability's power smoothly whereas the population growth is no longer significant at all. Nevertheless, now we have the least significant Innovations and El_prices, and the very significant GSP which we do not consider as important results due to their low power coefficients.

The situation with Resources is not that easy as with the rest of low predictable IVs. We can see that the fossil resources abundance of regions has the positive and very significant influence on the probability of the CRESS projects' increase in a region. It again disproves our hypothesis number two where on the theoretical level the less resources are in the regions, the better the chances are of the CRESS to be implemented. Nevertheless, the power of its predictability is extremely low which logically should give us a signal that there is no point in interpreting this IV due to the possibility which it has is unrealistic. However, we will still want to interpret it because the problem here is not in the absence of influence as the indicator shows, it is in the measurement system of this variable. We will keep this explanation to the discussion part. Table 3 also tells us that the IRR of the fossil resources abundance indicates that for every one unit increase on the amount of fossil resources, the possibility of the increase of CRESS projects in a region increased by factor of 1.0 - 100%.

The fifth and sixth models do change the general picture after we added the Elections and Governors IVs. We can proceed to the analysis from the top down. First, The RE potential in the fifth model saves its significance and the power coefficient. The same cannot be said about the sixth model where it slightly lowers its significance level to the mean significant one as well as it lowers the power coefficient. Second, the IVs such as GSP, El_prices and Innovations do not start showing results which are not only significant, but also have some effect on the DV. That is why the decision not to consider them as important ones remains the same. Third, the situation with Resources is also identical, it maintains its significance and power level in both fifth and sixth models.

Fourth, the new interested result which is worth mentioning is the very and the mean negative significance level of Elections in the fifth and sixth models correspondingly. It generally means that our first hypothesis is proved because it says that if there is an upcoming legislature elections or governor elections in a region, the possibility that the CRESS will be implemented is lower. The data shows us the same. Moreover, the power of its predictability is surprisingly high which means the effect of this IV on our DV is realistic. Table 3 also tells us that the IRR of the upcoming elections indicates that for every one upcoming election, the possibility of the decrease of CRESS projects in a region increased by factor of 0.392 - or 39%.

The last IV which occurs only in the sixth model tends to be significant too. It has the very negative significant level and not as high level of its predictability's power. Due to the fact that almost the same result this IV shows in the logistic regression, we think that it might be worth interpreting. Its result disproves our third hypotheses saying that the less the level of regions' connection is with the federal center, the higher the likelihood is that the CRESS will be implemented. What is more, Table 3 also tells us that the IRR of the upcoming elections indicates that for every unite increase on the level of regions' connection with the federal center, the possibility of the decrease of CRESS projects in a region increased by factor of 0.965 - or 96%.

As we were expecting, the use of the Poisson regression dramatically improves our results brining into the statistical significance area twice as more variables. With the help of this method, we have managed to overcome the low data frequency problem which we came across when conducting the logistic regression analysis. The next logical step of ours will be the results' interpretation. We will mostly rely of the Poisson regression's results just because of the availability of conclusions we can possibly have with it. However, we will try to compare both analyses during this discussion stages.

Table 2. Standard errors in parentheses: ⁺ p < 0.1, ^* p < 0.05, ^** p < 0.01, ^*** p < 0.001

Poisson models. Table 2: DV - countable Num_Project

Source: author's calculations.

Table 3. Standard errors in parentheses:*** p<0.01, ** p<0.05, * p<0.1

Poisson regression's IRR.

Source: author's calculations.

3.5 Interpretation of results

Having conducted both regression analyses to cover all of the collected IVs, we have tried to prove our hypotheses with respect to the regressions results. Each hypothesis presents one of the socio-economic, electricity market and political factors which we have derived from our theoretical literature review. If a hypothesis is proved, it means that a particular factor has a certain contribution to the likelihood that the CRESS is (not) implemented in a region. It is important to remember that we are trying to identify political determinants of regions' decision (not) to implement the CRESS. The priority in the analysis is given to them (political factors).

The most interesting result with regard to political determinants of the CRESS implementation by Russian regions is about the upcoming legislature and governors elections. The results of the Poisson regression show us that this IV has the very significant and influential predictability of the CRESS implementation. As we have already discussed in the theoretical chapter, political business cycles push politicians at the end of their political term to make a stake on low-risk populistic policies with no long-term perspectives. Such short-term low risks policies allow them to try to gain electorate support as quick as possible to possibly win an upcoming election. Such an effect reduces so called `windows of opportunities' for unpopular policies. Nordhaus, William D. 1975. "The Political Business Cycle". The Review Of Economic Studies 42 (2): 169.; Akhmedov A.M. Chelovecheskiy Kapital i Politicheskiye Biznes-tsikly // Akhmedov, A. 2006. "Chelovecheskiy Kapital I Politicheskiye Biznes-Tsikly". Konsortsium Ekonomicheskikh Issledovanii€ I Obrazovaniya, Rossiya I SNG, no. 6: 7. In Russian case, the upcoming change in a political business cycle in a region influences regional political elites making them implement policies which are less risky than the CRESS policy. The CRESS might have a possible negative effect on electricity prices which can destabilize citizens' comfort zone which entails the negative reaction from the federal center which requires the citizens' social satisfaction to be stable. It may lead to a dismissal or reappointment of regions' authorities for the next elections.

The level of regional elites' connection with the federal center (Governors) is one of the most interesting political hypotheses which shows the appreciable contribution to the likelihood that the CRESS is not implemented by a region. As we were describing in the first chapter of our paper, the federal structure of the Russian Federation is about the hierarchy of the center over the regions. To gain an opportunity to act more or less independently or to act not with the strict supervision from the federal authorities, sub-units have to submiss several necessary demands to prove their loyalty and a region's stability: (1) permanent favorable electoral results and (2) well-functioned anti-protests structures, and (3) social patronage (citizens should be satisfied with the gas, electricity prices and housing and communal services). Gel'man, Vladimir, and Sergei Ryzhenkov. 2011. Op. Cit.; Sharafutdinova, Gulnaz, and Rostislav Turovsky. 2016. Op. Cit.; Busygina, Irina. 2016. Op Cit. 113.; Gel'man, Vladimir, and Sergei Ryzhenkov. 2011. Op. Cit. That is why our hypothesis proclaims that the better and stronger regional elites' connection with the federal center, the better they manage to answer the demands and to negotiate with the federal center, the bigger the space for independent and innovative actions like the CRESS implementation.

Nevertheless, both regressions show us that at the just and very significance levels the less regions are connected with the federal authorities, the more the possibility is that the RE support scheme is used. Obviously, it disproves our hypothesis. One possible explanation for that can be a possibility that the CRESS initiative is controlled by the center even stronger than we have expected. It might cause that the more regions are supervised, the higher a possibility the CRESS is launched.

There is, however, the big problem with the power of influence of the likelihood this IV suggests - it is extremely low, but it has the highest indicator among non-powerful IVs if we look at the average values. The reason why we decide to include it to the interpretation is that both Poisson and logistic regressions show us the same significant result towards the federal connection predictor. It makes us consider that the problem is not in the unrealistic predictability of this IV, but in the way this data is measured. There is a concern, that neither the Poisson regression, nor the logistic regression are not able to analyze the ranking data appropriately. That is why there might be such a problem with Governors predictability.

The next interesting factor statistically influencing the likelihood of the CRESS implementation by regions is Resources. As we were describing in the second chapter of our paper, if a sub-unit is locked-in with the fossil resources, the business and political institutions seek a mutual equilibrium which is also arranged around fossil fuels, a possibility that a mechanism of RE support will be implemented is low. Goldthau, Andreas, and Benjamin K. Sovacool. 2012. "The Uniqueness Of The Energy Security, Justice, And Governance Problem". Energy Policy 41: 232-240. In Russia the example of such regions is Tumenskaya Oblast' which is functioning and developing only through gas and oil revenues.

However, as the both regressions show us, the analysis disproves this hypothesis suggesting that the more fossil resources regions have, the more likely they implement the CRESS. At first, it sounds paradoxically, but we cannot explain it right now because we do not have any theoretical ground to do that. We can suggest researching it for the further development.

It is worth mentioning that this IV has some problems with its coefficient about the power of its significance. It is extremely low which says us that despite the level of significance, this IV does not have almost any contribution to the DV's likelihood. However, we think that the problem here is not in the real power of the Resources, but in its way of measurement and representation in the dataset. We assume that if we transform the measurement category from million tones to thousands, for instance, to make numbers smaller, the coefficient will be much stronger. It affects both regressions.

The results of analyses do not match our theoretical expectations with regard to the political regime's openness represented in the data by the ENP. It means that the political inclusiveness of a region does not have any influence on the CRESS implementation. The number of patents which regional authorities approve for the innovations does not show any realistic influence either. It can be due so due to the wrong data we have found. Probably, if we found the number of patents for the RE technologies themselves, the result would have more predictability power. Electricity prices which should by theory be low in order to implement the CRESS to leave a space for a possible increase is not significant and influential too.

Conclusion

Within the framework of the research, we conducted the logistic and Poisson regression analyses to answer the research question of our paper: what are the political factors, which made the Russian regions (not) implement the capacity-based renewable energy support scheme in between 2013 and 2018?

After reviewing the literature, we formulated 9 hypotheses where we listed 9 political and non-political factors which are theoretically able to influence the Russian regions (not) implement the capacity-based renewable energy support scheme. The results of the analyses are presented in Table 4.

It is important to say that despite the fact that some factors do not match with our theoretical expectations, we have logically interpreted them. That is why, we think that it is crucial to name them as our research's results as well because they reflect the contribution which our paper brings to the topic of renewable energy policies in Russia.

Table 4. Results.

Note: *an IV has been statistically proved in both regressions, but the coefficient of its effect on the CRESS implementation is very low. We think that this problem is not a result of the absence of its influence, it is a result of the incorrectly measured data.

Thereby, the goal of the research is reached, we established political factors, which made the Russian regions (not) implement the capacity-based renewable energy support scheme in between 2013 and 2018. They are upcoming elections in regions, the level of regions' connection with the federal center, and regions' fossil resources endowment.

There are, though, several drawbacks of our research. First of all, we think that it is appreciable generalized - we cannot say for 100% that the factors which we highlight in our conclusion are the real predictors of the CRESS implementation in Russia. In order to reduce the generalization, we think that a constructivist research is needed to locally prove which of these factors are the real predictors. Second of all, the biggest minus of our research is the assumption we have made that the Russian regions actually have the capacity to influence the CRESS policy. However, it is definitely worth conforming it. There is Boute's legal research about this point, but it was made in 2013 which makes it irrelevant nowadays. Third of all, the appreciable disadvantage of our paper is the collected data. The quality of it leaves something to be desired which also lowers our interpretation capacities.

Bibliography

Books, articles, dissertations

1. Akhmedov A.M. Chelovecheskiy Kapital i Politicheskiye Biznes-tsikly // Akhmedov, A. 2006. "Chelovecheskiy Kapital I Politicheskiye Biznes-Tsikly". Konsortsium Ekonomicheskikh Issledovanii€ I Obrazovaniya, Rossiya I SNG, no. 6: 7.

2. Bayer, Patrick, and Johannes Urpelainen. 2016. "It Is All About Political Incentives: Democracy And The Renewable Feed-In Tariff". The Journal Of Politics 78 (2): 603-619.

3. Boute, Anatole, and Alexey Zhikharev. 2019. "Vested Interests As Driver Of The Clean Energy Transition: Evidence From Russia's Solar Energy Policy". Energy Policy 133: 1-10.

4. Boute, Anatole. 2011. "A Comparative Analysis Of The European And Russian Support Schemes For Renewable Energy: Return On European Experience For Russia". The Journal Of World Energy Law & Business 4 (2): 157-180.

5. Boute, Anatole. 2013. "Renewable Energy Federalism In Russia: Regions As New Actors For The Promotion Of Clean Energy". Journal Of Environmental Law 25 (2): 261-291.

6. Brewer, Garry D. 1974. "The Policy Sciences Emerge: To Nurture And Structure A Discipline". Policy Sciences 5 (3): 239-244.

7. Busygina, Irina, and Mikhail Filippov. 2013. "Low Quality Of National Governance And Uneven Economic Competiveness Of The Russian Regions". SSRN Electronic Journal.