The Effect of Nyepi Day
Alemarie Ceria
11/6/2020
Introduction
Nyepi Day, also known as “Day of Silence,” is a Balinense Hindu holiday that takes place in Bali, Indonesia every Çaka New Year. It occurs annually, typically in March, at 6:00 a.m. on Nyepi Day to 6:00 a.m. the following day, which is known as Ngembak Geni Day. Instead of fireworks and celebration on December 31st, like in the United States, the Balinese people use Nyepi Day to “pray for the purification of humanity, earth, an(d) the universe” (Why Do We Celebrate Nyepi Day?, 2018). Their time is spent remaining silent, fasting, meditating, and observing Catur Brata Penyepian, which consists of the four things that are restricted during Nyepi Day. The prohibitions, according to Bali Plus Magazine (2018), include no fire or light (Amati Geni), no working or activity (Amati Karya), no venturing outside (Amati Lelungan), and no entertainment (Amati Lelanguan). Though this is a part of Hindu culture, it is a public holiday that is observed throughout the island and by non-Hindu residents and tourists.
The restrictions of Nyepi Day, which is strictly enforced by traditional security men called “Pecalang,” causes activity to dramatically decrease (Wikipedia Contributors, 2020). In my research, I am interested in measuring the effect of Nyepi Day on the amount of light seen from space satellites, specifically around the Java, Bali, and Lombok Islands and then relative to Indonesia as a whole. Nyepi Day provides researchers the rare opportunity to observe and measure the impact of a population by comparing and contrasting the polar opposites of little/no activity of one night to the normal night of a developing country the next. Artificial light pollution affects the environment, wildlife, our health, and ability to study space due to an increase in skyglow and decrease in dark nights. As populations rapidly grow and countries continue to develop, wasteful energy consumption will soar without proper policies and standards in place. Therefore, additional analysis is needed to keep track of and find more efficient solutions that will combat this environmental issue.
Satellite Pictures Java, Bali, and Lombok
Based on these satellite images, we can clearly see a difference in light count when comparing the images of the day before and after to Nyepi Day in all years from 2017 to 2020. It is as if they are preparing for Nyepi Day one day before because it is not as bright. On Nyepi Day it is barely seen, meaning almost blacked out. The day after, we can assume that people are celebrating Ngembak Geni Day or the new year, as mentioned previously, because there are more lights seen. This is especially true in in 2018 and 2020.
Summary Statistics Tables
Indicator Variables
# Nyepi Day Indicator:
javabalilombok$nyepi <- as.numeric((javabalilombok$date == "2017-03-28") | (javabalilombok$date == "2018-03-17") | (javabalilombok$date == "2019-03-07") | (javabalilombok$date == "2020-03-25"))
indonesia$nyepi <- as.numeric((indonesia$date == "2017-03-28") | (indonesia$date == "2018-03-17") | (indonesia$date == "2019-03-07") | (indonesia$date == "2020-03-25"))
# Bali Island Indicator:
javabalilombok$bali <- as.numeric((javabalilombok$provincename == "Bali"))
indonesia$bali <- as.numeric((indonesia$provincename == "Bali"))Java, Bali, and Lombok
| Total (N=216) |
|
|---|---|
| Nyepi Day | |
| Yes | 24 (11.1%) |
| No | 192 (88.9%) |
| Bali Island | |
| Yes | 36 (16.7%) |
| No | 180 (83.3%) |
| Logged Light Count | |
| Mean (SD) | 17.2 (1.30) |
| Median [Min, Max] | 17.2 [14.4, 19.1] |
There is a total of 216 observations in the javabalilombok data set. Nyepi Day accounts for 24 or 11.1% of the observations, while Bali Island has 36 or 16.7% of the observations. The mean of logged light count is 17.2 with a minimum of 14.4 and a maximum of 19.1. Logged light count has 216 observations.
Indonesia
| Total (N=1224) |
|
|---|---|
| Nyepi Day | |
| Yes | 136 (11.1%) |
| No | 1088 (88.9%) |
| Bali Island | |
| Yes | 36 (2.9%) |
| No | 1188 (97.1%) |
| Logged Light Count | |
| Mean (SD) | 12.4 (1.14) |
| Median [Min, Max] | 12.7 [8.81, 14.9] |
In the indonesia data set there are a total of 1,224 observations. Of those observations, 136 or 11.1% are Nyepi Day. 36 or 2.9% of the observations are Bali Island. Moreover, logged light count has 1,224 observations with a mean of 12.4, minimum of 8.81 and maximum of 14.9.
Regressions and Results
Java, Bali, and Lombok
one.lm <- lm(log(lightcount+1) ~ nyepi, javabalilombok)
two.lm <- lm(log(lightcount+1) ~ nyepi*bali, javabalilombok)
three.lm <- lm(log(lightcount+1) ~ nyepi*bali + factor(provincename), javabalilombok)| Dependent variable: | |||
| Logged Light Count | |||
| Model 1 | Model 2 | Model 3 | |
| (1) | (2) | (3) | |
| Nyepi Day | 0.028 | 0.346 | 0.346*** |
| (0.281) | (0.685) | (0.082) | |
| Bali Island | 0.833 | -0.122 | |
| (0.707) | (0.088) | ||
| Nyepi Day on Bali Island | -0.383 | -0.383*** | |
| (0.750) | (0.090) | ||
| Fixed Effects? | No | No | Yes |
| Observations | 216 | 216 | 216 |
| R2 | 0.00004 | 0.021 | 0.986 |
| Adjusted R2 | -0.005 | 0.008 | 0.986 |
| F Statistic | 0.010 (df = 1; 214) | 1.545 (df = 3; 212) | 2,107.371*** (df = 7; 208) |
| Note: | p<0.1; p<0.05; p<0.01 | ||
The results become statistically significant with a 99.9% confidence level for both Nyepi Day and Nyepi Day on Bali Island. This finding means that there is in fact an affect of Nyepi Day on Bali Island on logged light count, specifically, a 38.3% reduction.
Indonesia
four.lm <- lm(log(lightcount+1) ~ nyepi, indonesia)
five.lm <- lm(log(lightcount+1) ~ nyepi*bali, indonesia)
six.lm <- lm(log(lightcount+1) ~ nyepi*bali + factor(provincename), indonesia)| Dependent variable: | |||
| Logged Light Count | |||
| Model 4 | Model 5 | Model 6 | |
| (1) | (2) | (3) | |
| Nyepi Day | -0.042 | 0.326 | 0.326*** |
| (0.103) | (0.583) | (0.114) | |
| Bali Island | 2.022*** | 2.459*** | |
| (0.558) | (0.114) | ||
| Nyepi Day on Bali Island | -0.380 | -0.380*** | |
| (0.592) | (0.115) | ||
| Fixed Effects? | No | No | Yes |
| Observations | 1,224 | 1,224 | 1,224 |
| R2 | 0.0001 | 0.063 | 0.965 |
| Adjusted R2 | -0.001 | 0.061 | 0.964 |
| F Statistic | 0.168 (df = 1; 1222) | 27.525*** (df = 3; 1220) | 945.587*** (df = 35; 1188) |
| Note: | p<0.1; p<0.05; p<0.01 | ||
Bali Island in model 5 as well as Nyepi Day, Bali Island, and Nyepi Day on Bali Island in model 6 are statistically significant with a 99.9% confidence level. When including fixed effects, we see that there is still a 38% logged light count reduction from Nyepi Day on Bali Island relative to the whole of Indonesia.
Scatterplots
Just by observing the scatterplots, we can see that there is an obvious reduction in logged light count in both cases when comparing the results of Bali Island to all other provinces. Furthermore, there is a bigger gap between the logged light count of Bali Island and other provinces when comparing the surrounding areas of Java, Bali, and Lombok to the whole of Indonesia.
Classic Linear Model (Assumptions MLR.1-MLR.6 and Multicollinearity)
Linear in parameters
\[y = \beta_{0} + \beta_{1}x_{1} + \beta_{2}x_{2} + ... + \beta_{k}x_{k} + u\]
I cannot truly test whether or not the relationship between Nyepi Day and light count is linear in the population. However, based on the scatterplots above, the relationship looks linear and therefore, I believe assumption MLR.1 is satisfied.
The following model is the final model I use to estimate the population:
\[log(lightcount) = \beta_{0} + \delta_{0}nyepi + \beta_{1}bali + \delta_{1}nyepi \cdot bali + u\]
Random sampling
\[(x_{i1},x_{i2},...,y_{i}):i = 1, ...,n\]
Although Nyepi Day is held annually and typically in March, the day it lands on varies because it is determined by the lunar phases, which is what the Balinese Çaka calendar uses. Therefore, based on this, I can assume assumption MLR.2 is satisfied because Nyepi Day changes each year.
No Perfect Collinearity
Estimated model:
\[log(lightcount) = \beta_{0} + \delta_{0}nyepi + \beta_{1}bali + \delta_{1}nyepi \cdot bali + u\]
As previously mentioned, since Nyepi Day varies each year, it is not constant. Additionally, the light count on Bali varies across the different provinces and based on daily human activity and therefore, it is also not constant. Though these independent variables have a relationship, they do not have an exact perfect relationship. Thus, I conclude that assumption MLR.3 is satisfied.
Zero conditional mean
\[E(u_{i}|(x_{i1},x_{i2},...,x_{ik}) = 0\]In models 1, 2, 4, and 5 exogeneity would not hold. Models 1 and 4 do not account for unobserved factors, but is addressed in models 2 and 5 because \(bali\) accounts for the fact that Nyepi Day is a Balinese Hindu Holiday practiced primarily on Bali Island. Moreover, models 3 and 6, which are difference-in-difference models, addresses the variances stemming from varied light count across provinces by holding \(provincename\) fixed. By considering all of these factors in the latter models, I believe the explanatory variables are exogenous and conclude that assumption MLR.4 is satisfied.
Multicollinearity
\[R^2_{j} \to 1\]vif(two.lm)## nyepi bali nyepi:bali
## 6 9 14vif(five.lm)## nyepi bali nyepi:bali
## 34 9 42Under assumption MLR.3, I assume that my explanatory variables have a relationship, but one that is not perfect. The variance inflation factors (VIF) test above proves this is true because the results show values greater than 10, which indicates that my explanatory variables are highly correlated.
Although I can still run my regression with a multicollinearity problem, it makes the variance of my betas large. Still, as I have mentioned under assumption MLR.4, the utilization of difference-in-difference models fixed the problem and decreased the variances. This is shown in my regression results below comparing models 2 and 5 (with the multicollinearity problem) to models 3 and 6 (solving the multicollinearity problem), which also made the results statistically significant:
| Dependent variable: | ||||
| Logged Light Count | ||||
| Model 2 | Model 3 | Model 5 | Model 6 | |
| (1) | (2) | (3) | (4) | |
| Nyepi Day | 0.346 | 0.346*** | 0.326 | 0.326*** |
| (0.685) | (0.082) | (0.583) | (0.114) | |
| Bali Island | 0.833 | -0.122 | 2.022*** | 2.459*** |
| (0.707) | (0.088) | (0.558) | (0.114) | |
| Nyepi Day on Bali Island | -0.383 | -0.383*** | -0.380 | -0.380*** |
| (0.750) | (0.090) | (0.592) | (0.115) | |
| Fixed Effects? | No | Yes | No | Yes |
| Observations | 216 | 216 | 1,224 | 1,224 |
| R2 | 0.021 | 0.986 | 0.063 | 0.965 |
| Adjusted R2 | 0.008 | 0.986 | 0.061 | 0.964 |
| F Statistic | 1.545 (df = 3; 212) | 2,107.371*** (df = 7; 208) | 27.525*** (df = 3; 1220) | 945.587*** (df = 35; 1188) |
| Note: | p<0.1; p<0.05; p<0.01 | |||
Homoskedasticity
\[E(u_{i}|(x_{i1},x_{i2},...,x_{ik}) = \sigma^2\]bptest(one.lm)##
## studentized Breusch-Pagan test
##
## data: one.lm
## BP = 0.023959, df = 1, p-value = 0.877bptest(two.lm)##
## studentized Breusch-Pagan test
##
## data: two.lm
## BP = 35.892, df = 3, p-value = 7.892e-08bptest(three.lm)##
## studentized Breusch-Pagan test
##
## data: three.lm
## BP = 32.161, df = 7, p-value = 3.792e-05bptest(four.lm)##
## studentized Breusch-Pagan test
##
## data: four.lm
## BP = 0.012562, df = 1, p-value = 0.9108bptest(five.lm)##
## studentized Breusch-Pagan test
##
## data: five.lm
## BP = 9.8013, df = 3, p-value = 0.02033bptest(six.lm)##
## studentized Breusch-Pagan test
##
## data: six.lm
## BP = 157.96, df = 35, p-value < 2.2e-16Using the Breusch-Pagan test, I retain the null hypothesis of homoskedasticity for most of the models except for models 1 and 4 since the p-values are less than 0.05. Since the heteroskedasticity problem comes from preliminary models (models 1 and 4) and is addressed in models 3 and 6 by fixing for variances in logged light count across provinces, I can assume that MLR.5 is satisfied.
Normality of error terms
In the javabalilombok data set, I have 216 observations and 1,224 observations in the indonesia data set. Thus, because I have over 100 observations, I have enough data to not worry about this assumption. However, as we account for fixed effects across provinces (seen in the histograms of three.lm and six.lm’s residuals), we start to see a more normal distribution.
Conclusion
To recap, we found statistically significant results with a 99.9% confidence level that there is a 38% reduction in logged light count when observing Nyepi Day on Bali Island. This is consistently true when focusing on the surrounding islands of Java, Bali, and Lombok as well as when looking at Indonesia as a whole. These findings are expected and not only statistically significant, but also economically significant. Economic growth of countries and therefore, GDP as a measurement of it, is the priority of most countries because it is widely believed to equate to an increase in living standards and prosperity. However, as aforementioned in my introduction, an increase in development also means an increase in wasteful energy consumption and thus, leads to more pollution in societies where there is a lack of proper policies and standards to combat the environmental issue.
The economic significance of this study’s findings is relevant to the concerns many people and countries have about climate change. A follow up question we can ask is: “How does Nyepi Day effect the health of a population?” Then run a difference-in-difference regression to see the health levels of Bali residents who observe Nyepi Day in comparison to other districts who do not observe the holiday and in other words, are exposed to more pollution. In a similar manner, this may be done to also understand the effects of Nyepi Day on nature. Although this scenario may not be replicable in places like the United States because it would be unethical to force people to shut their businesses down, to stay home, and do nothing, I believe that it is helpful in understanding how different indicators may be more sustainable than GDP, which prioritizes economic growth. For example, the indicator Gross National Happiness prioritizes sustainable and equitable socio-economic development, environmental conservation, preservation and promotion of culture, and good governance. This type of study would be replicable in Bhutan if they had a similar holiday that also reduces economic activity and thus, environmental degradation.
Works Cited
Why Do We Celebrate Nyepi Day? (2018). Bali Plus Magazine. https://www.baliplus.com/2018/03/19/why-do-we-celebrate-nyepi-day/
Wikipedia Contributors. (2020, November 23). Nyepi. Wikipedia; Wikimedia Foundation. https://en.wikipedia.org/wiki/Nyepi