The Socio-Economic Factors of the Covid-19 Pandemic in Turkey: A Spatial Perspective

Abstract

This study investigates the role of various socioeconomic determinants and vaccination rates in the spread of Covid-19 in a spatial setting in Turkey. For this aim, we employ the 41 sub-indicators of Life Index in Provinces data provided by the Turkish Statistical Institute which is obtained based on the Organization for Economic Cooperation and Development (OECD) Better Life Index approach. Our results indicate no global interactions in the transmission process of the disease among Turkish provinces. This means that the infection burden in the neighboring province does not significantly affect the infection burden of a given state. Yet, we show that vaccination rates and the median age of a neighboring province significantly affect the number of total cases in a given province. We find that as the vaccination rates of a neighboring province rise, the number of total cases in a given province also increases. This finding can be attributed to the “neighbor–reliant immunity” concept. It seems that people with vaccine hesitancy toward Covid-19 feel safer without a vaccine when their neighbors are mostly vaccinated. Last, people with a higher satisfaction rate with their health status are more likely to catch the disease due to underestimation of negative consequences.

1. Introduction

Coronavirus Disease 19 (Covid-19) marked the years 2020 and 2021 with its very fast diffusion rates and severity. With the quick development of vaccines against the disease, the pandemic right now seems to come to an end. Yet, living the last 2 years with a contagious disease has left some serious questions: What is the role of socio–economic determinants in the transmission of an airborne contagious disease like Covid–19? What factors are most influential and make countries more vulnerable to such diseases? What is the role of spatiality in the spread? In this study, we aim to investigate the answers to these questions for Turkey. More specifically, we try to point out the most influential socio–economic factors in the spread of Covid-19 in Turkey in a spatial setting.

The first Covid-19 case is confirmed in Turkey on 11th March 2020 in İstanbul. It spread quickly all over the country. To limit its transmission among the Turkish provinces, similar strategies to other countries, such as travel restrictions and partial curfews, were applied in the initial days. Yet, in time, it has become clear that every country has its own dynamics that limit the effectiveness of precautions against the Covid-19. For example, [1] find that the extreme poverty level is an important determinant in the national performance of low– and middle–income countries, since it determines the ability of social distancing. They also note that the disadvantaged share of the population in terms of socio–economic status is more vulnerable to contagious diseases. Therefore, each country must be assessed individually to understand its needs and to be prepared for future diseases. Analyzing the spread of the Covid-19 and the socio–economic determinants behind is important to be ready for any country as well as Turkey.

The ties between the socio–economic status in the spread of Covid-19 were discussed previously in the literature. These studies mainly focus on mainland China [2, 3] and the USA [4, 5]. Some of them compare the national performances of many countries based on the socio–economic variables, (e.g., [6, 7, 8]). Yet, as [4] clearly state, “In a quickly changing pandemic landscape…county-level data and analysis is crucial to understanding needs and supporting planning efforts.” We, therefore, turn our attention to Turkey, which is one of the most affected countries in the world. Jain and Singh [9] indicate that 60% of the cases in Asia clustered in Turkey alongside mainland China and Iran. Yet, the number of studies examining the impacts of these variables on the spread of Covid-19 is still limited (among these studies, one can note the study by [10]. Our paper aims to fill this gap while considering the effects of being close to the places where the Covid-19 cases are dense.

Ref. [11] emphasize the role of spatiality in the analysis of contagious diseases by stating that “when people move, they take contagious diseases with them.”. Much before the Covid-19 pandemic, [12] indicates that infectious diseases are the main concerns of medical geography which defines the “place” as a vital dimension of the transmission process besides the other risk factors. In the SARS epidemic example, [13] notes the importance of detecting spatial linkages which shows the potential spreading ways and spatial clusters. Similarly, [14] argues that the diffusion of infectious diseases is directly related to the location. As a result, to understand the diffusion process of such diseases, spatial analysis is a requirement.

Although the importance of location in the transmission process of such diseases besides the other risk factors is mentioned heavily in the literature, studies considering geography in the Covid-19 incidence rates are scarce and they mostly make a choice between the spatial autoregressive model (SAR) and spatial error model (SEM). Ehlert [15], for example, attempts to determine the socio-economic and region-specific in the Covid-19 transmission in German counties with a choice between SAR and SEM specifications. Andersen et al. [16] examine the local transmission of Covid-19 cases in the USA. Again, they made a selection between SAR and SEM based on the Lagrange Multiplier (LM) tests. Sun et al. [17] employ SAR, SEM, and SAC models to detect the Covid-19 period prevalence in the US counties. Baum and Henry [4] consider several demographic factors and income as well as air pollution and health-related variables in order to explain the spread of Covid-19 in the US states. They also employ a SAR model. Guliyev [18] use the number of confirmed new cases in mainland China as the dependent variable where the recovered cases and the rate of deaths are the explanatory variables in a spatial panel setting. He compares SEM and SAR models, but cannot show spatiality in the explanation of the rate of new cases. He concludes that the spatial lag of X (SLX) model fits the nature of local spillovers in this association for China.

The situation for the scarce studies that consider the spread of Covid-19 in Turkey from a spatial perspective is parallel to the world literature. Tuzcu [8] provides an exploratory spatial analysis with different weight matrices for Turkish Covid-19 cases and deaths in which high spatial autocorrelation is detected particularly for major Turkish provinces. Similarly, [19] use Moran I and Local Indicator Spatial Association (LISA) statistics to determine the hot and cold spots among Turkish provinces. Dinç and Erilli [20] examine the effects of a group of socio–economic determinants as well as climate–based variables on the number of Covid-19 cases with SEM and SAR specifications. Göktaş [21] looks at the relationship between centrality in terms of trade, transportation, and health and the number of cases in a Turkish province while considering other socioeconomic factors as control variables. For this aim, he employs SAR and SAC models. Aral and Bakır [22] use the impact of population density, elderly dependency ratio, Gross Domestic Product (GDP) per capita, literacy rate, and health capacity variables to explain the diffusion of Covid-19 in Turkey with a SAR model. They find global spillovers and significant coefficients for population density and elderly dependency ratio while explaining the increase in the Covid-19 cases.

With this study, we also contribute to the scarce literature on Covid-19 studies in Turkey with a spatial perspective. One of the novelties of this paper comes from the spatial model it adopts. Unlike the previous spatial studies on Covid-19 diffusion, we argue that a spatial Durbin model (SDM) must be the first model to adopt for the analysis. The SDM approach is well known for containing both the global and local spillovers at the same time, which is a feature of the Covid-19 pandemic. In fact, when the best describing model is unknown, [23] suggests using SDM as a starting point as well. As a result, we start our analysis with an SDM setting to detect the local and global spillovers in the diffusion of Covid-19 cases across 81 Turkish provinces. Different from the existing studies, we use the vaccination rates and sub–indicators of Life Index in Provinces by the Turkish Statistical Institute (TSI) as the explanatory variables. Life Index in Provinces report includes 41 sub–indicators about income, work life, safety, housing, environment, social life, access to infrastructure services, education, life satisfaction, and civic engagement. By using these sub–indicators, we believe that every aspect of socioeconomic status in Turkish provinces, from per km2 green area to health capacity, can be taken into account. Hence, an exhaustive list of variables that have the potential to impact the spread of Covid-19 is considered. Controlling the vaccination rates also allows us to detect its role among other variables and its impact on the spread of the disease. By doing so, we are able to contribute to the very limited literature on Covid-19 vaccine hesitancy.

To the extent of our knowledge, a similar study to our setting that examines the spread of Covid-19 in Turkey belongs to [24]. He employs 11 leading indicators of the Life Index in Provinces report, not the sub–indicators as well as other socioeconomic and environmental variables such as GDP, household size, age, air quality, humidity, and average temperature. Although this study also mentions the spatial distribution of Covid-19 cases in Turkey and benefits from some spatial maps, the main analysis method is Ordinary Least–Squares (OLS), not spatial models. By using spatial analysis methods with an exhaustive set of socioeconomic indicators, we believe that our study closes an important gap in the literature.

Our results indicate no significant global impacts in the spread of Covid-19 cases across Turkey, but significant local interactions. We show that vaccination in a given province decreases the total number of cases per hundred thousand people in the same province, but increases the Covid-19 cases in the neighboring province. This seemingly puzzling finding is a result of vaccine hesitancy toward Covid-19 vaccines. The “neighbor–reliant immunity” argument by [1] explains that people with vaccine hesitancy feel safer when more people around are vaccinated, so they can act more freely. This situation significantly and negatively affects the total number of cases. We also find that people that are more satisfied with their health status act more carelessly, and the number of total cases increases significantly with higher levels of this variable. The median age of neighbors and the satisfaction rate with a social life are variables that are inversely related to the number of total cases. As the median age of neighbors increases, the social interactions and traveling between provinces decreases to avoid the negative consequences of Covid-19. On the contrary, the rate of membership to political parties in a given province is positively related to the total number of cases in the same province. This finding can be attributed to more social interactions and less social distancing with increased civic engagement.

Based on the findings of this study, we can suggest that the usage of clear communication channels with society has vital importance in fighting against infectious diseases. In this way, it is possible to correct the misperceptions both about the nature of the disease and the vaccinations. Overconfidence about the health status and vaccine hesitancy might increase the overall number of cases, so the burden on the health care system.

The rest of the study continues with an explanation of the data and methodology utilized. The next section presents our findings. The last section concludes with the policy suggestions to the Turkish authorities for the next pandemics.

2. Data and methodology

2.1 Dependent variable and the selection of socio-economic variables

This study uses the total number of confirmed COVID-19 cases per 100,000 population from 81 Turkish provinces as the dependent variable. This data is publicly available and reported as weekly averages by the Turkish Ministry of Health.

To be able to determine which explanatory variables might be important in the spread of Covid-19, we examine thoroughly the previous literature that applies both a spatial and non-spatial analysis. Bassino and Ladmiral [25] argue economic variables like wealth or income are the main drivers of the person-to-person spread of infectious diseases such as COVID-19. Bassino and Ladmiral [26] demonstrate that low literacy has been influential in the spread of the disease. Sun et al. [17, 27] find that age is effective on the spread of COVID-19 cases. Population and population density were also noted as significant variables by [15, 25, 28, 29]. The number of doctors and the number of hospital beds are considered important factors by [3, 25, 26] because their availability has the potential to draw more COVID-19 patients to the area. Living in an urban vs. rural area might be another determinant in the spread of cases as noted by [25, 29]. Ehlert [15] also considers household size as a factor. Social life indicators and average space available per household are used by [25].

We proxy the socioeconomic and health status of each province that is noted in the literature by using the Life Index in Provinces provided by the Turkish Statistical Institute in 2016. This index is produced based on the approach of the OECD Better Life Index. The aim of the Life Index in Provinces is to compare the well–being and living quality of Turkish provinces as well as their economic status. To do so, 11 leading indicators and 41 sub–indicators that include both objective and subjective aspects of life are created. These indicators include income, work life, safety, housing, environment, social life, access to infrastructure services, education, life satisfaction, and civic engagement dimensions. Based on the previous literature, we select explanatory variables among the 41 sub–indicators. The dimensions that can affect the spread of Covid-19 but are not captured by the Life Index in Provinces, such as median age, percentage of individuals 65 years old and above, or population density are also added to the analysis.

Besides the socio–economic factors, the vaccine uptake decision of societies is a crucial weapon against the spread of Covid-19. Therefore, we use the vaccination rates for individuals 18 years old and above for each province as a control variable in the models.

A summary of explanatory variables that are employed in this analysis and the data sources are reported in Table 1.

Variable	Proxy	Source
COVID-19 variables	Total number of cases per 100,000 population by Turkish provinces	Turkish Ministry of Health (weekly)
	Percentage of 18+ population vaccinated against COVID-19 at least once by Turkish provinces	Turkish Ministry of Health (daily)
Housing Conditions	Number of rooms per person	Life Index in Provinces by Turkish Statistical Institute (2016)
	The household size in Turkish provinces	Turkish Statistical Institute (2020)
Work Life	Employment Rate	Life Index in Provinces by Turkish Statistical Institute (2016)
	Unemployment Rate	Life Index in Provinces by Turkish Statistical Institute (2016)
	Average Daily Earnings	Life Index in Provinces by Turkish Statistical Institute (2016)
Income and Wealth	Percentage of Households in middle and higher Income Groups	Life Index in Provinces by Turkish Statistical Institute (2016)
	Percentage of Households declaring to fail on meeting basic needs	Life Index in Provinces by Turkish Statistical Institute (2016)
	GDP per capita by Turkish provinces	Turkish Statistical Institute (2013)
Health	Infant Mortality Rate	Life Index in Provinces by Turkish Statistical Institute (2016)
	Life Expectancy at Birth	Life Index in Provinces by Turkish Statistical Institute (2016)
	Satisfaction Rate with Health Status	Life Index in Provinces by Turkish Statistical Institute (2016)
	Health Capacity Index	Turkish Ministry of Health, Health Statistics (2018)
Education	Percentage of higher education graduates	Life Index in Provinces by Turkish Statistical Institute (2016)
Safety	Murder Rate	Life Index in Provinces by Turkish Statistical Institute (2016)
	Percentage of people feeling safe when walking alone at night	Life Index in Provinces by Turkish Statistical Institute (2016)
Civic engagement	Voter turnout at local administrations	Life Index in Provinces by Turkish Statistical Institute (2016)
	Rate of membership to political parties	Life Index in Provinces by Turkish Statistical Institute (2016)
	Percentage of persons interested in union/association activities	Life Index in Provinces by Turkish Statistical Institute (2016)
Access to Infrastructure Services	Number of internet subscriptions (per hundred persons)	Life Index in Provinces by Turkish Statistical Institute (2016)
Social life	Number of cinema and theater audiences (per hundred persons)	Life Index in Provinces by Turkish Statistical Institute (2016)
	Shopping mall area per thousand people (m2)	Life Index in Provinces by Turkish Statistical Institute (2016)
	Satisfaction rate with social relations	Life Index in Provinces by Turkish Statistical Institute (2016)
Median Age	Median of individuals’ age in Turkish provinces	Turkish Statistical Institute (2020)
Age 65+	Percentage of population over 65 + by Turkish provinces	Turkish Statistical Institute (2020)
Population	Population density of Turkish provinces	Turkish Statistical Institute (2019)

Table 1.

Socioeconomic and Covid–19 related variables and the data sources.

The data period is determined by the announcement periods of the Turkish Ministry of Health. Vaccination rates started to be announced at the province level on 04.07.2021 on a daily basis. The total number of cases per 100,000 population is announced weekly. Therefore, we consider the average total number of cases and vaccination rates for July 2021 in this analysis.

2.2 Methodology and model selection process

A standard OLS model is often estimated as a reference for the following spatial models. This study employs the same starting point. To understand the effect of location on the Covid-19 cases, many studies employ SAR and SEM specifications. You et al. [2] note that the SAR model will show how the infection burden in a location is affected by the infection burden in the neighboring locations. SEM is used to understand whether the OLS residuals are correlated to residuals of the neighboring locations. In the lines of [2, 3] also consider a SAC model. They argue that since the SAC model contains a spatial lag and a spatial error term, it can be seen as a combination of these two.

In fact, the spatial model family has a large set of approaches¹, and model selection is a crucial part of its applications. Baum and Henry [4] argues that this selection must be based on the spillover type that the economic theory points out. Unlike [2]‘s suggestion, [4] stresses that the SAC model is not the linear combination of SAR and SEM approaches. Not considering the spillover types in the selection of appropriate spatial models leads us to the identification problem noted in [5].

Jamison et al. [6] state that the locations that are closer to the center of the pandemic are affected more quickly than the distant ones. However, besides geographical proximity, Covid-19 can spread easily when the locations are connected on a network, such as traveling. It means that both global and local spillovers exist in the diffusion of infectious diseases. We argue in this paper that this nature of Covid-19 can be best captured with an SDM approach. Aydin and Yurdakul [7] also recommends using SDM as a departure point, when the true data generating process, as in the case of Covid-19, is unknown. SDM will also give the linear combination of SAR and SEM specifications [4], as intended by [2, 3].

The OLS model that is used as a benchmark is presented in Eq. (1).

$$y_i={\beta }_0+{\mathit{\beta X}}_i+{\epsilon }_i$$E1

where y_i is the total number of Covid-19 per 100,000 people in a given Turkish province. ${\beta }_0$ reflects the intercept term and β is the vector of coefficients for the explanatory variables. X_i is the socioeconomic variables that are shown in Table 1 and ε_i is the error term with iid. We check the OLS assumptions. No multicollinearity problem is detected. The insignificant variables (p < 0.10) are excluded from the model in order to refine.

The SDM specification that is used in this paper is shown in Eq. (2).

$$y_i={\gamma }_0+\mathrm{\rho W}{y}_i+{\mathit{\gamma X}}_i+\mathrm{W}{X}_i\mathrm{\theta }+u$$E2

In Eq. (2), the dependent and the explanatory variables are the same as the OLS model defined in Eq. (1). However, here, we scale both the dependent and explanatory variables with a spatial weight matrix (W). The coefficient $\mathrm{\rho }$ reflects the global interactions in the spread of Covid-19 in Turkey, while $\mathrm{\theta }$ demonstrates the local interactions. u is the error term.

Our model selection process follows [7] and we also compare our results with SAR and SLX specifications. The SAR and SLX models are shown in Eq. (3) and Eq. (4) respectively.

$$y_i={\alpha }_0+\mathrm{\rho W}{y}_i+{\mathit{\alpha X}}_i+\lambda$$E3

$$y_i={\delta }_0+{\mathit{\delta X}}_i+\mathrm{W}{X}_i\mathrm{\theta }+\tau$$E4

The spatial weight matrix used throughout all these models is the same. The elements of W take the value of 1 if two Turkish provinces are neighbors, and zero otherwise.

3. Findings

3.1 Spatial map of Total number of cases in Turkey

First, we examine the spatial variation of the total number of cases in Turkey. The spread of Covid-19 cases across Turkish provinces is shown in Figure 1.

Figure 1.

The variation of Covid-19 cases across provinces in Turkey in July 2021.

The map in Figure 1 demonstrates that there are regional variations in the diffusion of Covid-19 cases. The total number of confirmed cases increases from west to east of Turkey. We also consider the four main regions of Turkey and statistically compare the average cases in these regions. These regions are defined as follows: i. Marmara, Aegean and Mediterranean Regions, ii. Black Sea Region, iii. Central Anatolia Region, and iv. Eastern and Southeastern Anatolia. Since the normal distribution assumption of ANOVA cannot be satisfied, we compare the means of these regions with the aid of Kruskal–Wallis test. The results reject the equality of means of the Covid-19 cases across regions at a 5% level (χ² = 8.757, p–value = 0.0327). Pairwise comparisons revealed that Eastern and Southeastern Anatolia have statistically higher rates than all other regions in Turkey. This result also supports the visual findings in Figure 1.

3.2 Results from spatial modeling

We begin our analysis with the classical OLS model. By excluding the insignificant variables at the 10% level, we refine the model and obtain the ultimate model. We check the OLS assumptions. We find heteroskedasticity in our main model which might be a result of spatial dependency.

As explained before, since the association between the total number of cases and the various socioeconomic variables is not well discussed in the previous literature, we start with an SDM specification to avoid the omitted variable problem [23], which is also the linear combination of SAR and SEM specifications [30]. However, the SDM specification does not show significant results. The LR tests comparing SDM vs. OLS and SAR vs. OLS cannot reject the null hypothesis of no significant global interactions. The lack of significant global spillovers indicates that the burden of the disease at one location is not affected by the burden of the disease in the neighboring locations. Yet, the LR test for the coefficients of local interactions in the SDM specification is significant at the 1% level (LR test is 52.5983, and the p–value is 0.0015). That is to say, although no global impacts can be detected in the transmission process of Covid-19 cases in Turkey, geography still matters in the form of local interactions. The socioeconomic features of neighboring provinces are influential on the spread of Covid-19 in a given province. This finding is in line with the study by [18] in which an SLX model is found appropriate to model the new cases in mainland China. Therefore, following [23], we continue our analysis with an SLX model. The final SLX model and the OLS model as a benchmark are shown in Table 2.

	(1)	(2)
	Final OLS Model	Final SLX Model
Vaccine Rate	−1.5737***	−1.3693**
	(0.5605)	(0.5926)
Membership to Political Parties	2.0558	3.2805**
	(1.5622)	(1.5043)
satisfaction rate with social relations	−2.1859	−2.2772***
	(1.6857)	(0.7206)
satisfaction rate with health status	3.4398	3.3891**
	(2.6869)	(1.6631)
Constant	−21.0895	−26.1843
	(117.8685)	(105.4470)
W*Vaccine Rate		4.9654***
		(1.7096)
W* Median Age		−10.2514***
		(3.4318)
AIC	870.386	874.219
Adjusted R^2	0.2341
Number of Observations	81	81

Table 2.

The impact of socioeconomic variables on the total cases of Covid-19: OLS and SLX models.

Vaccination is clearly the strongest weapon in the fight against Covid-19. The findings from Table 2 also confirm this situation and reveal that the vaccination rate and the total number of cases are significantly and negatively related. Interestingly, it is found that the effect of vaccination rates in the neighboring provinces is positive and significant. That is, the increased rates of vaccination in the neighboring locations cause a growth in the total number of cases in a given province. This result seems puzzling at first, but it can be explained by the vaccine hesitancy concept. Vaccine hesitancy is defined as the “delay in acceptance or refusal of vaccination despite the availability of vaccination services” [31]. Ke and Zhou [32] state that the vaccine uptake decision of an individual can be dependent on the actions of the neighbors. They call this concept “neighbor–reliant immunity”. They argue that people that are hesitant toward the Covid-19 vaccine might feel more “immune” without uptaking the vaccine itself if the people around are already vaccinated. This situation is visible here as well. It is seen that people with Covid-19 vaccine hesitancy do not limit their actions as much as before with neighboring provinces as the vaccination rate of neighboring provinces increases. As a result, the number of confirmed cases in a given province increases.

We also show that as the satisfaction rate with health status increases, the number of total cases also rises in a particular province. This finding can be attributed to the fact that Covid-19 is mostly perceived as an older people’s disease or only dangerous for people with co–morbidities. To fight this perception, World Health Organization (WHO) made many announcements, including the one that the Chief of WHO explained that “young people are not invincible”. It seems that this perception is still valid in July 2021 in Turkey, and it might grow even stronger with the relatively less severe variants and the ongoing vaccination process.

The rate of membership to political parties is an indicator of civic engagement. As this variable has a higher rate, the social relations, and connections increase as well. This would make it difficult to keep the social distance and adapt to “stay at home” calls. Our findings in Table 2 confirm this result and demonstrate a positive effect of this variable on the total number of cases.

Median age, itself, is not a determinant of the spread of Covid-19 cases across Turkish provinces. However, the median age of the neighbors negatively impacts the number of cases in a given province. This finding is in line with [15]. He notes that the increase in the median age of neighbors reduces the social interactions with the given state and traveling, so less spread has occurred.

The satisfaction rate with social relations is a proxy for social life. Our results indicate that the higher values of this variable are related to a lower level of total cases. It seems that people who are more satisfied with their social life are most likely to keep their social distance and less engaged with many people. This finding might be explained by the existence of video–calls and other telecommunication methods. Individuals may meet their social needs via the internet and stay at home at the same time.

We cannot show any significant effect of housing conditions, work life, income and wealth, or health indicators other than health status, education, safety, and access to infrastructure services, however.

The results of this paper once more emphasize the importance of vaccinations in order to control the number of cases. In the case of such infectious diseases, governments must use clear communication channels with society to avoid misperceptions about the nature of the disease or the precautions to avoid further spread. Our findings show that over–confidence about the individual health status and vaccine hesitancy increase the number of total cases, so the burden on the health care system.

4. Conclusion

This study employs an exhaustive set of socioeconomic variables and vaccination rates to detect their roles in the spread of Covid-19 in Turkey in a spatial setting. Province-level data allows us to detect the existence of spatiality as well. We cannot detect any global interactions in the diffusion process, so the number of infected people at one location does not bring an extra infection burden to the neighboring locations. Yet, our findings show that local interactions in terms of vaccination rates and median age play an important role in the increase in the total number of cases. Increased vaccination rates in the neighboring provinces also increase the total number of cases in a given province. This result can be explained by the vaccine hesitancy toward the Covid-19 vaccine. We also find evidence that people that are more satisfied with their health status are more likely to catch the disease and increase the total number of cases. To fight the misperceptions about the nature of the disease and the vaccination procedure, the Turkish government must adopt a clear–communication policy and actively work for individuals to access reliable information.