ON THE INFLUENCE OF CLIMATE AND SOCIO-ECONOMIC CONDITION TO THE DENGUE INCIDENCES: A SEMIPARAMETRIC PANEL REGRESSION APPROACH

Dengue is one of the most dangerous diseases in the worlds. In particularly in East Java province Indonesia, dengue has been identified as one of the major causes of death. Hence, it is important to investigate the factors that induce the number of dengue incidences in this region. This study examines climate and socio-economic conditions, which are assumed to influence the number of dengue in the examined region. The semiparametric panel regression approach has been applied and the results are compared with the standard panel regression. In this case, the socio-economic condition is treated parametrically while climate effect is modeled nonparametrically. The analysis showed that the number of dengue incidences is significantly influenced by the income per-capita and the number of inhabitant below 15 years. Furthermore, the dengue incidence is optimum under rainfall of 1500 to 3670 mm, temperature of 22 to 27 degree and humidity of 82 to 87%. The elasticity allows us to identify the most responsive and most irresponsive district towards the changes of climate variable. The study shows that Surabaya is the most responsive district with respect to the change of climate variables.


INTRODUCTION
Dengue is one of the most dangerous diseases in the world. The spread of the dengue is infected by a vector called as Aides Aegepty. Dengue frequently arises as an extraordinary case that causes relatively high degree of mortality rate leading to economic loses. Indonesia becomes one of the countries with the highest rate of dengue incidence among ASEAN countries. Statistic data shows that there are about 1317 dengue cases in 2010. One of the provinces in Indonesia with high rate of dengue case is East Java. The number of dengue in East Java gradually increases from 8287 cases in 2004 and reaches the peak in 2007 with 25950 cases (DH, 2009). Therefore, study about the dengue incidence in East Java is an important issue.
Researches on identifying the causes of dengue in a region have been intensively studied. Khormi and Kumar (2011) investigate the influence of socioeconomic condition to the number of dengue in Saudi Arabia. Ma et al. (2008) studied the similar case in Singapura. The study argues that the number of dengue incidence does not depend only on the socio-economic condition, however it is affected also by the climate condition. Some researches that study the impact of climate to dengue are Zhou et al. (2004); Hales et al. (2002) and Maslukha (2010). These researches assume that either the socio-economic or climate condition affects the dengue incidence separately. In fact, climate and socio-economic condition may affect the dengue incidence simultaneously. Mondzozo et al. (2011) is the Science Publications AJES only study that models the dengue incidence by taking into account both climate and socio-economic conditions at the same time, using semiparametric panel regression approach. The study was conducted in Africa and it shows that the model can provide more information about the dengue incidence. In term of the goodness of the model, semiparametric panel regression approach outperforms the standard panel regression.
This study applies the similar approach as in Mondzozo et al. (2011) to model the dengue incidence in East Java Indonesia. The responsiveness degree of the examined districts towards dengue incidence will be performed. Moreover, elasticity representing the degree of vulnerability of each district towards dengue with refers to the climate condition will also be discussed.

Panel Regression
Panel regression is a regression composed from cross-section and time series data. The panel regression is given by the following model: where k is number of coefficient β (Baltagi, 1995).

Semiparametric Panel Regression
Semiparametric model combines parametric and nonparametric model. Semiparametric regression model can be written as follow Tseng et al. (2009);Li and Racine (2007) and Racine (2008): Parameters estimated in this regression are f,β,a i , where f is an unknown function while parametric variables are assumed to influence the spread of the dengue linearly Estimation method is done by two steps procedure following Robinson (1988).
First step: Form a conditional expection of response variable (y) on climate variables (X it ) such that: where, the nonlinear variables X it is assumed to be uncorrelated with regional individual effect and error. By substracting the first from second equation above we obtain: is the kernel estimator. The semirapaetric model is then transformed into linear equation: where, the parameter β ⌢ is estimated by similar procedure as standard panel regression.

Second step:
The second step will estimate the function f. After the value of β ⌢ has been estimated, the function f is estimated by substituting β ⌢ through: , we obtain nonparametric form: And f is estimated locally by minimizing density function. Ullah and Mudra (2002) showed that f can be expressed as This specification considers a fact that climate change causes a significant jump of dengue in a certain region. In general, elasticity is defined as change (in percentage) of response variable caused by one percent changing on the predictor variable. Elasticity is calculated by the following Equation 1: where, e xy is elasticity of y variable on x. This equation is used to calculate the dengue projection based on the elasticity definition.

Data
Data used in this research is secondary panel data spanning from 2003 to 2010. Data of climate variables are obtained from the climatology station Karang Ploso Malang, while socio-economic data is collected from Center of Statistic-East Java Province. Data about the number of dengue incidences are obtained from the Ministry of Public Health-East Java.

Variables
There are three kinds of data used in this research i.e., climate variable, socio-economic as the predictors and the number of dengue/100000 population as the response variable. The Table 1 shows the variables used in the study.
All climate variables should be are measured as the average of annually data as the socio-economic data are available annually. The idea of rescaling the number of inhabitants below 15 years old and poor inhabitants is to reduce the lag of the data due to the scale of data measurement.

Analytical Method
This research uses dataset from 13 districts in East-Java region spanning from 2003-2010. The reason of using 13 districts data considering availability of climate data, in which complete temperature and humidity data Science Publications AJES are completely available only in 13 districts i.e., Blitar, Kediri, Malang, Jember, Banyuwangi, Pasuruan, Sidoarjo, Nganjuk, Magetan, Gresik, Sumenep, Surabaya and Batu. Nevertheless, these districts are representative enough for East Java province.
In summary, the steps of analysis can be described as follows: • Perform panel regression between number of dengue incidents with socio economic and climate variables • Parameters estimation on the semiparameteric panel regression • Nonparametric estimation of function f • Calculate the elasticity from smooth coefficient of the nonparametric kernel regression.

Descriptive Statistic of Variables
This section briefly describes the statistic of each variable observed over all districts during the observed periods (2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010). The mean of dengue incidence per 100000 inhabitants is 43.85 with the minimum of 3 incidents happened in Sidoarjo, while the maximum incidents happened in Batu 2010 with 167 events. The average of income per-capita is IDR 7748000 with standard deviation of 6411000. The mean of the number of inhabitants below 15 years old per 100000 inhabitants is 24257. It means that if there exists 100000 inhabitants in any district, thus 24257 of them are supposed to be below 15 years old. The average of annual rainfall is 1686.3 mm with the minimum is observed in Sumenep (2009) corresponding to 61 dengue incidents per 100000 inhabitants, while the maximum is 3670 mm in Blitar (2010) with dengue of 80 incidents per 100000.

Panel Regression
Panel regression analysis is done by regressing in panel the number of dengue per 100000 inhabitants with socio-economic variables (Z1 to Z4) and climate variables (X1 to X4). This model assumes that both groups of variable linearly affect the dengue incidences. Table 2 presents the parameters of the regression.
The table shows the coefficients (estimate), standard error and statistical test of the panel regression. The value of the coefficient of determination is 42.698% which means that the 42.69% variability of dengue incidents can be explained by socio-economic and climate variables. Based on the value of t-statistic, it is known that there are two socio-economic variables that are significantly influence the number of dengue i.e., density and income per-capita. Density influences dengue with negative coefficient, meaning that the number of dengue incidence decreases with increasing the density, assuming that other variables are constant. Another fact is observed for income per-capita, where the number of dengue incidence increases with increasing income per capita in any districts.
For the climate variables, none of them influences the dengue incidence. This insignificant influence may happen due to incapability the model to overcome nonlinearity on the relationship between both variables. Therefore, standard panel regression may fail to generate a good result.

Semiparametric Panel Regression
The semiparametric panel regression that will be examined is: Where: y it = The dengue incidence per 100000 inhabitants in ith distric on the period t X it = A vector of climate variables Z it = A vecor of socio-economic variables  6.41×10 −1 7.04×10 −1 0.9107 R 2 = 42,698% *; Significant at the level of α = 0.05, ** Significant at α = 0.01, ***; Significant at α = 0.001, Significant at α = 0.1 *; Significant at the level of α = 0.05, **; Significant at α = 0.01, ***; Significant at α = 0.001, Significant at α = 0.1  The semiparametric panel regression model yields on two models i.e., nonparametric model with unknown function and parametric model with coefficients of β. Hausman test is used to identify whether the effect will be fixed or random. As it is significant, thus we conclude that panel regression model with fixed effect is more appropriate for the case.

AJES
Based on the t statistic, it suggests that there are two variables that significantly influence the dengue i.e., income per-capita and number of inhabitants below 15 years old. For the income per capita, it shows a positive coefficient which means that region or district with higher income per-capita tends to have larger number of dengue incidents. In contrast to this, negative coefficient is obtained for number of inhabitants below 15 years old. In other words, we may say that dengue attacks adults more than children.
The fixed effect model leads to different values of a i for every district as shown in the Table 4.
This a coefficient shows intercept of the panel model with fixed effect and hence this value can be used to show the vulnerability of the district if all districts have same socio-economic condition. The more negative the value, the more vulnerable the district towards dengue incidence. Regions that are least vulnerable are adalah Malang, Pasuruan dan Sidoarjo. While districts that are most vulnerable are Blitar, Nganjuk, Magetan, Sumenep dan Kota Surabaya. The remaining districs are moderate. The socio economic variables solely can explain the variability of dengue with only 34.47%. Although it is lower than parametric panel regression, the goodness of the semiparametric panel regression model should be combination of R 2 from both parametric and nonparametric component.
Having known the coefficients of the socio-economic variables, we proceed to the nonparametric model for the climate variables. The nonparametric component consists of three variables i.e., rainfall, temperature and humidity, by which the kernel function will be estimated and result on the plots shown in Fig. 1. The nonparametric model can explain the variability of the dengue incidence with value of 80.54%, shown by the R 2 value. Figure 1 show plots resulted from nonparametric kernel regression using 95% confidence level. The upper panel of Fig. 1 shows that dengue tends to be stable at the rainfall 500 to 1500 mm in a year, while if the rainfall reaches 1500 mm to 3500, it will increase the number of dengue. The middle panel of the figure shows that the relationship between dengue and temperature is nonlinear. It is shown by the unsystematic pattern of the plot. In general, dengue incidences will increase on the temperature of 22-27°C, while for temperature exceeds 27°C, the dengue incidence will decrease. The last plot shows that humidity influences dengue with nonlinear pattern. With refers to the line plot, the dengue incidents will decrease with humidity of 70 to 75% and it will increase with humidity of 82 to 87%. The dengue will also increase under humidity of below 70%. Semiparametric panel regression yields on more flexible model as the effect of each region can be explained individually.

Elasticity of Dengue Incidence to the Climate Variables
The elasticity is obtained from the smooth coefficient of the nonparametric regression and the value represents the vulnerability of each district towards the climate condition. The Table 5 below listed the elasticity.
The elasticity on the table shows the fluctuation of the dengue incidences due to the change of rainfall, temperature and humidity. The dengue in each district is responsive to the change of the temperature and humidity. Meanwhile, it is less responsive to the change of the rainfall intensity. Surabaya is the most responsive district towards the change of temperature and humidity, while Sumenep is the least responsive towards the change of both variables. Responsive in this case is defined as the vulnerability of the district to the dengue incidents if there is a change on the climate variables.

CONCLUSION
This study investigates the influence of socioeconomic and climate variables to the dengue incidence using semiparametric panel regression approach. The analysis shows that the proposed model outperforms the standard panel regression. With this regression, the vulnerability and elasticity of each region in East Java province can be explained.

ACKNOWLEDGEMENT
The reachers gratefully acknowledge the financial support from the Institut Teknologi Sepuluh Nopember through the research grant Unggulandecentralisation program.