Introduction

Since its identification in 1981, Acquired Immunodeficiency Syndrome (AIDS) has spread over the world and has become responsible for the deaths of 40.1 million individuals (Sharp and Hahn, 2011). By the year 2022, there were 39 million individuals living with Human Immunodeficiency Virus (HIV) and 630,000 people who had died as a result of AIDS-related illnesses. In Eastern and Southern Africa, which are the most affected regions, there are 20.8 million people living with HIV. In comparison, there are 6.5 million people living with HIV in the Asia Pacific, 4.8 million in West and Central Africa, 2.3 million in Western Europe, Central Europe, and North America, 2.2 million in Latin America, and 190,000 in the Middle East and North Africa.Footnote 1There is substantial evidence that demonstrates that poverty is an HIV risk factor. Antiretroviral treatment coverage is lower in countries that have a lower gross national income (GNI) per capita. This contributes to a higher HIV prevalence rate in such nations (Levi et al., 2018; Rajaraman et al., 2006). The pandemics, economic, and humanitarian crises that have occurred in recent years have had a detrimental influence on the global HIV response, with the biggest damage occurring in indebted low- and middle-income countries (Beltran et al., 2022; Thu et al., 2021). It was stated that private medical care can be an alternative, particularly in nations that have high incomes. It may be difficult for people in low-income nations to afford certain medical care if they do not have a sufficient household income (Booysen and Visser, 2010). Sexual behaviors in low-income households can move toward various undesirable consequences, particularly following an incident that causes economic degradation (Dinkelman et al., 2008). HIV epidemics have a devastating effect on national incomes in Sub-Saharan Africa (Lovász and Schipp, 2009). Having access to antiretroviral medication is very important in lowering the rate of HIV transmission as well as the incidence of HIVD. Ghana serves as an example of how, despite the fact that antiretroviral drugs are free, access to other healthcare may be limited by insurance restrictions or out-of-pocket costs (Poku et al., 2020). Higher levels of education have a positive effect on people living with HIV by enhancing their sex-related knowledge, behaviors, and attitudes, social networks, and economic status (Jukes et al., 2008). There is mounting evidence that sex education in schools enhances HIV awareness, stimulates HIV testing, and decreases risky behaviors such as unsafe sex (Fonner et al., 2014; Kim et al., 2019). Children whose parents or themselves are HIV positive cannot attend school, which has a detrimental influence on the educational results (Zinyemba et al., 2020). Because of its influence on school attendance, HIV contributes to a lower level of education, which, in turn, contributes to an increased HIV prevalence via a variety of different mechanisms of action.

It has been emphasized for years that safer sex is important in HIV prevention; however, sexual preferences, number of partners, sexual behaviors adopted by partners, methods they use for protection and how accurately they practice these methods, and the serological status of partners are all important in the context of safer sex (Stevens and Hall, 2001; Graham et al., 2005; Crosby, 2013; Quirk et al., 1998; Díaz et al., 2019). Drug and alcohol use has been linked to sexual risk-taking because it reduces the desire to take precautions (Zhao et al., 2006; Rehm et al., 2012). Another danger associated with IV drug use is the spread of disease via shared needles (Norton et al., 2008). Drug and alcohol users are more likely to engage in risky sexual behavior, have more sexual partners, and be less likely to be tested for HIV (Zhao et al., 2006). Disease prevalence is influenced by a number of factors, including income, which is linked to environmental and housing conditions, demographic characteristics, and education level, which is linked to knowledge about HIV and HIV testing (Silva et al., 2022; Bossonario et al., 2022; Dounebaine and Winskell, 2021; George et al., 2022).

Evidence suggests that HIV/AIDS is widespread across Africa (Harman, 2015; Hickel, 2012). On the other hand, increasing globalization and technology advancements led to more mobility among the population. This may make the impact of any virus, as well as the influence that viruses have on society, much greater. Therefore, an analysis of HIV and the effects it has should be conducted at a global scale. This is one of the first works to combine all of these concerns to examine the influence of various significant factors on HIVD by integrating nations all across the globe in a panel data analysis. This makes it one of the first works of its kind. In addition to the data acquired for the whole study, which covers 135 nations, the results obtained for seven geographical areas and particular countries are also provided. To conduct the study, methods such as panel data analysis, VAR analysis, and panel Granger causality tests were employed. Consequently, the purpose of this paper is to analyze alternate panel data methodologies, sample size, and empirical results that may be sensitive to differences across nations and regions in order to get more reliable findings. Research has been hindered by factors such as social stigma, fear, and a general lack of interest in the social and economic issues at play despite the fact that HIV has become a critical global outbreak. Although a significant quantity of research has been conducted in medical departments, the political and economic aspects of HIV retain a lower profile, particularly in terms of current studies in this field. This study is essential to fill this gap in knowledge. It also focuses on the economic factors that are contributing to the spread of HIV around the globe. In this scenario, the implementation of policies that have the potential to address the economic challenges that are the root cause of HIV might pave the way for the discovery of solutions to reduce HIVD. The purpose of this research is to determine whether or not GNI, mean years of schooling (MYS), drug use (DD), and unsafe sex (UD) all play a role in the development of HIVD, and if so, which of these risk factors plays a larger role in the development of that. The final topic will focus on determining the political implications of lowering HIVD. This work contributes fresh evidence to the existing body of research in the areas of alternative panel data analyses as well as the investigation of the impacts of GNI, education, and behavioral risk factors on HIVD on a global scale and across a variety of geographical areas. The assertion that the impact size of these explanatory factors in the context of HIVD is not the same across places that vary substantially in terms of wealth and education, such as nations in Sub-Saharan Africa and developed European countries. As a consequence of the empirical findings gathered from the research, it is also feasible to produce various policy suggestions for the fight against HIV that are specific to different locations. People who reside in low-income nations or who themselves have poor incomes are presumed to be more likely to participate in risky behaviors and to have fewer financial resources available to them for accessing medical care.

The following is the structure of this paper: The literature in this area from a variety of scientific domains was examined in the section “Previous research and the current work”, along with the innovative aspects of the present study, which were given out in this section. Section “Data and methodology” provided an overview of the data and techniques used. The empirical analyses were performed in the section “Empirical results.” The outcomes of the empirical investigations have been laid out in the section “Discussion.” The paper was finished in section “Conclusion”.

Previous research and the current work

Previous research

There are recent studies that studied the probable relationship between HIV/AIDS, economic variables, and other factors in nations. HIV-positive individuals in Malaysia were studied by Chow et al. (2022), who used regression analysis to examine how their sociodemographic, economic, health status and lifestyle factors affected their health-related quality of life between January 2020 and August 2020. It was discovered that smoking was the most important risk factor for having poorer scores in the physical, social, and environmental domains. It has been suggested that a poor physical condition is associated with unemployment as well as an unpleasant response to drugs. In particular, lifestyle and socioeconomic status were revealed to have a considerable impact on HIV-positive people’s health-related quality of life. The research conducted by Ekholuenetale et al. (2022) focused on the connection between socioeconomic characteristics and HIV self-testing knowledge among South African women. In the analysis of the year 2016, the multivariable logistic model was applied. It was shown via research that women from richer homes who have a strong understanding of HIV infection are more likely to have undergone HIV self-testing compared to women from households with the least amount of income who also have a good understanding of HIV infection. It was also discovered that learning to take exams was a key factor in raising test-taking rates. Reed et al. (2022) investigated the economic vulnerability, violent risk factors, and sexual risk factors for HIV among female sex workers in the city of Tijuana in Mexico. The study was conducted using questionnaires and personal interviews. During the course of the study, both crude and adjusted logistic regression models were utilized. According to the results of this research, there may be a connection between enhancing the well-being of female workers and a reduction in the incidence of violence and HIV. Employing a simulation based on an agent-based model, Demeulemeester et al. (2022) studied the economic effect of generic antiretrovirals in France for the period between 2019 and 2023. It was thought that the economic impact of HIV/AIDS might grow as antiretroviral therapy continues to improve and HIV patients can enjoy a normal lifetime. It was brought up in the discussion that monetary cost reductions may play a part in the provision of resource allocations for innovation and prevention. Research was carried out by Nosyk et al. (2020) to investigate the ideal mix of interventions with the goal of reducing the incidence of HIV/AIDS by 90% in ten years across six cities in the US. A dynamic HIV transmission model that was calibrated with the most successful tactics was implemented to deliver the largest possible health benefit in a manner that was also cost-efficient. This was done as part of an economic modeling effort. It was suggested that this approach could be effective in lowering the prevalence of HIV/AIDS. Nevertheless, accomplishing a national goal is very necessary to eradicate HIV prevalence.

Haakenstad et al. (2019) made an effort to research the costs associated with HIV/AIDS and looked at the constraints that governments have when it comes to providing further funding for the epidemic. Between the years 2000 and 2016, the spatiotemporal Gaussian process regression analysis was performed on 137 low- and middle-income nations worldwide. In addition, a stochastic frontier analysis was used to determine the possibility of higher expenditures by the government in the area of HIV/AIDS. It was claimed that some nations could put more domestic funding into the fight against HIV/AIDS. Despite this, there will still be a gap on the global scale when it comes to fighting the virus. Dauda (2018) examined the connection between HIV/AIDS and economic development in West Africa by taking into consideration 11 different nations. For the time period between 1990 and 2016, a GMM technique was used, and variables such as incidence, prevalence, the number of persons living with HIV/AIDS, and mortality related to AIDS were included in the analysis. It was discovered that a rise in the frequency and incidence of the virus, as well as the number of individuals living with the virus and fatalities caused by the virus, considerably hindered economic development. It was suggested that the decline in life expectancy brought on by HIV has a negative impact on economic development. Poudel et al. (2017) conducted a survey using quantitative research methodology in order to evaluate the financial impact that HIV/AIDS has had on people and families in Nepal. The direct costs of therapy, the productivity costs, and other relevant variables behind costs were the primary focuses of this research. Income in the household, occupation, current health state, and geographic district were the factors that influenced the direct expenses. The costs of productivity vary depending on factors such as a person’s health, race, sexual orientation, and district. It was reported that HIV/AIDS imposed a huge economic hardship on those who were infected with HIV as well as their families. Maynard and Ong (2016) made use of dependency theory, which asserts that the dependence of developing nations on debt, trade, and foreign investments has a detrimental effect on public health. In the research, a two-way fixed-effects OLS regression analysis was conducted, and a total of 80 nations were taken into consideration during the time period between 1989 and 2012. It was anticipated that an increase in HIV prevalence would be caused by an increase in total debt, short-term debt, foreign debt, and GDP.

In addition, Bidzha et al. (2024) employed both linear and non-linear econometric approaches in order to explore the impact of antiretroviral therapy on the connection between HIV/AIDS and real GDP per capita in South Africa from 1995 to 2019. The study revealed that antiretroviral therapy effectively reduced the impact of HIV/AIDS on real GDP per capita by increasing worker productivity. According to the findings, the impact of HIV/AIDS on real GDP per capita was shown to be greater among men. The results of this study indicate that expanding HIV testing and treatment might be effective in improving productivity. Stevens et al. (2024) applied Bayesian mixed-effects spatial regression to analyze data from 2010 to 2023 for the nations in sub-Saharan Africa. It was pointed out that critical populations in sub-Saharan Africa have a high HIV prevalence but receive less antiretroviral therapy. This indicates the need for targeted preventive and treatment programs for HIV. Dadzie et al. (2024) conducted a decomposition analysis using health surveys from 2015 to 2022 to investigate the socioeconomic disparities in the adoption of HIV testing during antenatal care in sub-Saharan Africa. HIV testing was found to be more prevalent among women in the highest income quintile, according to estimates. It was proposed that financial resources should be allocated to women who come from disadvantaged socioeconomic backgrounds. Implementation of health programs is necessary to mitigate female illiteracy and poverty. In other words, there should be a greater degree of awareness about HIV. In their study, Mcinziba et al. (2023) examined the connection between household income and adherence to antiretroviral medication among individuals living with HIV in a low-income population in South Africa. The analysis focused on the qualitative data obtained throughout the HIV prevention experiment conducted from 2016 to 2018. It was stated that the adverse effects of income variations should be taken into account for antiretroviral medication adherence assistance programs. Reshadat-Hajiabad et al. (2023) performed a cross-sectional cost of illness research to provide a detailed population-based analysis of the financial expenditures and economic burdens associated with HIV/AIDS in Iran. It was emphasized that prioritizing diagnostic testing and care is crucial for reducing expenses for patients with HIV/AIDS. This is because early diagnosis and prompt beginning of antiretroviral therapy may effectively restrict the progression of the disease. The research conducted by Zegeye et al. (2022) looked at women’s ability to make decisions and their level of knowledge of HIV/AIDS in 23 different countries in Sub-Saharan Africa. For the years 2010–2020, investigations using bivariate logistic regression as well as multivariate multilevel logistic regression were carried out. It was discovered that there is a low level of comprehensive knowledge about HIV/AIDS across the area; however, this varies greatly between nations. It is possible that the ability to make decisions will play a key part in the knowledge. Factors such as education, location of residence, media, HIV testing status, socioeconomic position, religion, and distance to health clinics may all play a significant part in an individual degree of awareness about HIV/AIDS. Virdasi et al. (2022) conducted research in Indonesia on women between the ages of 15 and 49 to determine the socioeconomic and demographic characteristics that influence women’s knowledge and attitudes towards HIV/AIDS. In the analysis of the year 2017, multiple logistic regression was applied. It was projected that factors such as age, education, money, location and region, access to information, possessing mobile phones, and autonomy might all have a substantial influence on HIV/AIDS knowledge and attitudes.

The current work

This article incorporates a variety of novel ideas, setting it apart from other publications previously published. The majority of research is restricted to using more standard methodologies and regression analysis based on surveys. This work contributes fresh evidence to the existing body of research in the areas of alternative panel data analyses and the investigation of the impacts of GNI, education, and behavioral risk factors on HIV-related fatalities on a global scale across a variety of geographic regions and on a number of specific countries. The assertion that the impact size of these explanatory factors in the context of HIV-related mortality is not the same across places that vary substantially in terms of wealth and education. As a consequence of the empirical findings gathered from the research, it is also feasible to produce various policy suggestions for the fight against HIV that are specific to different locations. According to our best knowledge, this is the first time in HIV research where not only panel regression analysis but also forecast error variance decomposition and panel Granger causality tests have been employed in conjunction with each other. In terms of empirical investigations, this study fills the gap that previously existed in the literature. In addition to the overall findings spanning a vast number of nations, the results acquired for seven geographical areas, as well as the results for specific chosen countries, are also provided.

This research tries to address not only the issue of whether GNI, MYS, DD, and UD have an influence on HIV-related deaths but also the question of which of these explanatory factors has the stronger influence on HIV-related deaths. In addition, based on the findings of the causality tests, an investigation was carried out to determine which factors are likely to be the most accurate predictors of mortality caused by HIV. The empirical findings of the research presented in this work, in particular those based on the VAR model and panel Granger causality tests, have the potential to provide additional evidence for the literature. It is presumed that this study may serve as a model for more research in the social sciences and medical sciences and that it can give different ways to explore the influence that various factors have on HIV/AIDS.

Data and methodology

In this research, the impacts of risk variables and economic factors on HIVD were evaluated using panel data models that included annual data between the years 1990 and 2019. Using the procedure of panel data analysis, the findings were assessed. The HIVD served as the dependent variable in this study. The rates of HIV infection that are caused by DD and UD are key behavioral risk factors that might lead to HIVD. The Global Burden of Disease Study database served as the source for the collection of these three variables. Both the MYS as an educational indicator and the GNI as an economic indicator were generated by UNDP. These are crucial risk factors for contracting HIV.

All variables are described by the following notations:

\({Y}^{{HIVD}}\) : HIV-related deaths (per 100,000 population).

\({X}^{{DD}}\) : HIV-related deaths due to drug use (per 100,000 population).

\({X}^{{UD}}\): HIV-related deaths due to unsafe sex (per 100,000 population).

\({X}^{{MYS}}\): Mean years of schooling

\({X}^{{GNI}}\): Gross national income per capita (2017 PPP based on US dollars)

Panel data is a dataset that integrates cross-sectional and time-series data. This results in increased efficacy, more degrees of freedom, less collinearity among the variables, more variability, and more informative data. The method generally provides researchers with a substantial amount of data, which enhances the efficacy of econometric estimates by reducing collinearity among explanatory variables and increasing the degrees of freedom. More importantly, longitudinal data enable researchers to examine a variety of critical economic inquiries that are not amenable to cross-sectional or time-series data sets.

The use of variance decomposition and Granger causality tests, together with the inclusion of empirical findings, helped to provide conclusions that were more reliable. Furthermore, empirical findings are given according to certain sub-regions and specific nations. After reviewing the study’s methodology and data, the empirical findings based on the panel data analysis are presented. The structure of the model is employed to define and estimate the heterogeneity of units and time-dependent heterogeneity in regression analyses using the panel dataset. This not only prevents a serious specification error but also ensures that the estimates are more reliable. The panel regression model in the double logarithmic form was defined as follows:

$$\begin{array}{l}{{LnY}}_{{it}}^{{HIVD}}={\alpha }_{i}+{\beta }_{1}{{LnX}}_{{it}}^{{DD}}+{\beta }_{2}{{LnX}}_{{it}}^{{UD}}{+\beta }_{3}{{LnX}}_{{it}}^{{MYS}}\\\qquad\qquad\qquad\quad+{\beta }_{4}{{LnX}}_{{it}}^{{GNI}}+{\beta }_{5}{{LnY}}_{i,t-1}^{{HIVD}}+{u}_{{it}}\end{array}$$

The error term \({u}_{{it}}\) is independently, identically distributed over \(i\) and \(t\), with a mean of zero and a variance of \({\sigma }_{u}^{2}\), \(N\) is the number of countries, and \(T\) is the number of observations for each period.

The Hausman test was implemented to decide whether to use random or fixed-effects for the panel regression model. In the panel regression model, the first-order logarithmic difference was taken. The statistical significance level was evaluated as p < 0.05.

The vector autoregressive (VAR) model is frequently employed in applied studies to investigate the dynamic relationships between variables. In general, the impulse-response functions and forecast error variance decompositions are implemented to determine the relationships between the variables, and the coefficients derived in the VAR model are not interpreted. The forecast error variance decomposition is a method employed to analyze the factors contributing to the fluctuations in a series. It is derived from the moving averages component of the VAR model. The approach displays the origins of shocks that arise in the variables themselves and in other variables as a percentage. The model demonstrates the proportion of a change in the variables that can be attributed to the variable itself against the proportion that can be attributed to other variables. The forecast error variance decomposition of HIVD based on the VAR model can allow this research to determine which of the socioeconomic and behavioral factors in the study is more effective in controlling HIVD.

The forecast error variance decomposition based on the VAR model is a tool that can be used to determine which independent variable is more effective on a dependent variable. To perform policy analysis with the help of the variance decomposition of the forecast error, a transformation that will provide “orthogonal shocks” must first be carried out in the VAR model. This transformation guarantees that the cross-correlations between the model’s error terms are zero. The orthogonality of the error terms to each other with this transformation allows them to be used for separate policy analyses. The reason for this is that for this type of analysis to be meaningful, the shocks in the other variables must be zero when shocks are given to any variable. Sims (1980) proposed Cholesky decomposition as the transformation that would allow the cross-correlations between error terms to be zero. The salient problem with this separation is that the matrix used in the transformation is not the only one. According to Sims (1980), Cholesky proposed ordering variables from the most exogenous to the most endogenous for this separation. Thus, in the AR representation of the variable at the first member of the sequence, there will be no instantaneous period of the other variables in the system. In the AR representation of the second variable in the ordering, only the instantaneous period of the first variable will be included. If the process continues this way, the AR representation of the last variable in the sequence can include the instantaneous period of all variables in the system, while it cannot act on any variable instantaneously (Darnell and Evans, 1990). The most exogenous of the variables in this study was MYS, while the most endogenous was drug use for HIVD. For these reasons, the Cholesky order should be followed from the most exogenous to the most endogenous, in the form of MYS, GNI, DD, UD, and HIVD, depending on our assumptions.

Granger non-causality testing is based on the approach provided by Toda and Yamamoto (1995), which uses the level VAR model with additional dmax lags to assess Granger causality between variables in heterogeneous mixed panels. This method was proposed by Emirmahmutolu and Kose (2011). Panels for stationary, non-stationary, cointegrated, and non-cointegrated series are all eligible for this method’s operation. It is also a strategy that may be used in situations in which some variables are stationary, and certain variables are non-stationary or when there is cointegration between two non-stationary variables for a given cross-section of individuals, but there is no cointegration in other cross-sections of individuals. The Fisher test statistic was used by Emirmahmutolu and Kose (2011) as a means of putting the Granger non-causality hypothesis to the test in the context of heterogeneous mixed panels. However, when cross-section dependency is present, the limit distribution of the Fisher test statistic can no longer be relied upon as an accurate representation of the data. Bootstrap distributions are used to generate the probability values for panel statistics. This approach takes into account the cross-section dependency. This approach works effectively in situations where there is cross-sectional dependency and heterogeneous mixed panels.

Empirical results

Cross-section dependence and unit root test

To determine whether or not panel models exhibit cross-sectional dependency, four distinct tests were applied. The findings from the test of the cross-sectional dependency are shown in Table 1.

Table 1 Test results.
Full size table

Examining the cross-sectional dependency of the models is necessary in order to prevent obtaining erroneous results. After it has been shown that the panel is subject to cross-sectional dependency, first-generation unit root tests such as the Im, Pesaran, and Shin (IPS), Augmented Dickey-Fuller (ADF-Fisher), and Phillips-Perron (PP-Fisher) analysis cannot be employed. As a result, tests of the second generation of unit roots, such as the Pesaran cross-sectionally augmented IPS (CIPS) test, were employed (Pesaran, 2007). The results of the CIPS test are given in Table 1. Panel unit root results showed that the integrated order of all variables was zero at the level. At the logarithmic level, the order of integration for \({Y}^{{HIV}}\) and \({X}^{{GNI}}\) was one, but the order of integration in the other variables was determined to be zero.

Panel regression model

The Hausman test results showed that the random effect null hypothesis could be rejected at conventional significance levels (\(p < 0.01)\). This result revealed that a fixed effect model could be chosen. The results of the fixed effect model are given in Table 2.

Table 2 Fixed effect model results.
Full size table

All coefficients of the independent variables were found to be statistically significant at 1%. These results showed that the selected independent variables had a statistically significant effect on HIVD. Additionally, the coefficient of determination, which is the model’s explanatory power, was quite large at 99.98%. This result demonstrated that the selected independent variables explained 99.98% of the total variance in HIVD per 100,000 population. This explanatory power was considerably greater than expected. The high coefficient of determination may be because HIVD with one period of lag was included as an explanatory variable in the model. To see this, one-period-lagged HIVD values were excluded from the model, but it was determined that the coefficient of determination remained at 99.85%. When the fixed effect model was estimated with only cross-section intercept terms, the coefficient of determination was 96%. In our model, the effects specification was cross-section-fixed (dummy variables). The parameter estimation method was the estimated generalized least squares (EGLS) method. Both the fact that there was an increasing trend in HIVD according to time and the fact that this trend did not differ too much from country to country caused the coefficient of determination to be high in the fixed effect model containing only fixed terms. The Durbin-Watson statistic showed that there was a positive autocorrelation between the error terms. Therefore, standard errors were corrected using cross-section weights (PCSE) for standard errors and covariances.

The signs of coefficients for the independent variables were estimated consistently with our prior expectations. The effects of DD and UD included in the model as risk factors for HIVD were determined to be positive. Additionally, the coefficients of both MYS and GNI were estimated as negative and consistent with our expectations. As the values of the risk factors for HIV increased, the number of HIVD cases also increased, while the number of HIVD cased decreased, MYS and GNI, which are leading indicators of economic development, increased. If DD increased by 1%, HIVD would increase by 0.2146%. If UD increased by 1%, HIVD would increase by 0.7363%. In contrast, if MYS and GNI increased by 1%, HIVD would decrease by 0.1439 and 0.0197%, respectively.

The one-year lagged values of HIVD had a statistically significant positive effect on the current value of HIVD. This result can be interpreted as the spread (transmission) effect of HIV. It is noteworthy that the coefficient of determination in the logarithmic model was quite large. This result may have been observed due to non-stationarity in the time-series data. Therefore, panel regression was repeated by taking logarithmic first-order differences corresponding to the annual change rates of the variables. Through logarithmic first-order differences, the data would be more stable over time so that stationary data could be obtained. Thus, it was aimed to obtain robust results. Robustness was also checked for panel regression analysis.

Alternative panel regression as a logarithmic first-order differences model was defined as:

$$\begin{array}{l}{\Delta {LnY}}_{{it}}^{{HIVD}}={\alpha }_{i}+{\beta }_{1}{\Delta {LnX}}_{{it}}^{{DD}}+{\beta }_{2}{\Delta {LnX}}_{{it}}^{{UD}}{+\beta }_{3}\Delta {{LnX}}_{{it}}^{{MYS}}\\\qquad\qquad\qquad+{\beta }_{4}\Delta {{LnX}}_{{it}}^{{GNI}}+{\beta }_{5}\Delta {{LnY}}_{i,t-1}^{{HIVD}}+{u}_{{it}}\,\end{array}$$

Where \(\Delta\) is the first-order difference operator. The Hausman test results showed that the random effect null hypothesis could be rejected at conventional significance levels (\(p < 0.01)\). This result indicated that a fixed effect model could also be chosen for logarithmic first-order difference variables. The first-order logarithmic difference fixed effect model results are given in Table 2

In the fixed effect panel regression model established with the variables whose first-order logarithmic differences were taken, the coefficient of determination was found to be quite large at 99.74%. The coefficient of determination decreased to 55.77% when one-period-lagged HIVD values were excluded from the model. This result revealed a coefficient of determination that was greater than expected due to one-period-lagged HIVD values. Additionally, in the fixed effect model, where only fixed terms were included, the coefficient of determination decreased to 19.3%. The coefficients of DD, UD, and one-period-lagged HIVD were again found to be statistically significant at 1%, while the mean schooling rate coefficient had a significance rate of 5%, and the per capita income coefficient had a significance rate of 10%.

Signs of the coefficients were estimated as consistent with our expectations. If the annual change in DD and UD increased by 1 point, the annual change in HIVD would increase by 0.2212 and 0.7602 points, respectively. If the change in MYS increased by 1 point, the annual change in HIVD would decrease by 0.0181 points. If the annual change in GNI increased by 1 point, the change in HIVD would decrease by 0.0046 points. The change in one-period-lagged HIVD had negative effects on the change in HIVD in the current period. This result contradicted the result obtained using the double logarithmic model. The findings showed that HIVD increased over time, but the rate of change in this increase decreased with time. In other words, HIVD values were still increasing, but the growth rate of this increase was decreasing. These results showed that the transmission effect of HIV increases over time, but this increase has a diminishing rate. These empirical findings also indicated that the treatment of HIV has improved over time, and the growth rate of deaths has been reduced.

VAR model: variance decomposition

In this study, VAR models were considered separately for each country by using annual time-series data. In these models, the appropriate lag structure was estimated as 4 for each country using the akaike information criterion (AIC), with a maximum lag of 4. For the regions included in the study, the results on the variance decomposition of HIVD were found for four forecast periods. Because of the large area of coverage in the study, the variance decomposition results for every country could not be included. Instead, mean results are reported for the examined countries overall and for certain sub-regions. The results on the variance decomposition of HIVD for all countries are given in Table 3.

Table 3 Results of variance decomposition for HIV-related deaths (all countries and certain sub-regions).
Full size table

The results on variance decomposition showed that DD stood out with a share of 49.9% in the first forecast period of all countries. The effect of DD on HIVD decreased in the following periods and declined to 28.42% in the fourth period. While the share of MYS in the forecast error variance decomposition of HIVD was 23.93% in the first period, this rate increased to 30.09% in the fourth period. The same trend was seen for per capita income. The results regarding the second risk factor, UD, demonstrated that it had little effect on HIVD values in the short-term, but this effect increased in the following periods. Although the share of UD in the forecast error variance decomposition of HIVD was quite low in the first period at 4.14%, this share increased in the following periods and rose up to 9.63% in the fourth period. These results showed that DD was the most effective variable on HIVD in the short-term, but this effect decreased in the long-term. The results on variance decomposition indicated that the effects of educational and economic indicators on HIVD were moderate in the short-term, but they increased in the long-term.

The results on variance decomposition in this study varied according to countries and regions. Therefore, the variance decomposition results of HIVD for certain sub-regions are reported in Table 3. Within the variance decomposition of HIVD, the share of DD for the first forecast period was always large for each sub-region compared to the other variables. These rates were quite large in developed European countries and the Middle Asia regions, with shares at 63.25 and 59.46%, respectively. The regions with the lowest rates of these shares were found as South America and Sub-Saharan African regions, at 42.17 and 42.58%, respectively. Although these shares decreased noticeably in the following forecast periods, they remained substantial, and the fourth-period share for developed European countries remained at a very large level, 45.82%.

The region with the highest share of UD in the variance decomposition in the first forecast period for HIVD was the Middle East and North Africa, at 7.42%. For the region, there was a very low degree of decrease in the subsequent forecast periods, and this share remained at 8.74% for the fourth forecast period. Middle Asia was the region with the lowest share of UD in the variance decomposition of HIVD. In this region, shares increased from 1.08% in the first forecast period to only 5.82% in the fourth period. In other regions, shares that were at a low level in the first forecast period increased in subsequent forecast periods. The region where this increase was the most evident was the Central America and the Caribbean region. In this region, the share of only 1.39% for the first forecast period increased to 17.82% in the fourth forecast period. In the examination of the share of HIVD in the variance decomposition of MYS, it was noteworthy that the region with the highest share in both short and long periods was Sub-Saharan Africa. Additionally, while this share was 18.64% in the first forecast period in the Middle East and North Africa (MENA) countries, the share increased to 41.06% in the fourth forecast period. In other regions, this share tended to increase, albeit at a low rate, between the first and last periods, indicating that education in the fight against HIV will be effective in the long-term, for especially developing and underdeveloped countries. In addition, the data suggest that income and level of schooling make for a large portion of the explanatory shares in explaining HIVD in the majority of regions, particularly over the long-term.

In the examination of the share of GNI, which was selected as an indicator of economic development, in the variance decomposition for HIVD, it was seen that the regions with the highest share were South America and Central America, and the Caribbean. The regions with the lowest share were MENA and the Far East. Notably, the share of 19.40% in the first period in the Sub-Saharan Africa region increased to 31.83% in the fourth period. Furthermore, while the share of per capita income in all sub-regions was lower in the first period, this share increased noticeably in subsequent forecast periods. These results showed that per capita income, albeit at different levels of importance for the selected sub-regions, had a particularly long-term impact on HIVD.

The results on the variance decomposition of HIVD in the four developed G7 countries and four underdeveloped countries where HIV is observed at high rates are given in Table 4. In the comparison of the variance decomposition values of HIVD for developed and underdeveloped countries, it was seen that the most substantial differentiation was in the share of DD. For developed countries, the share of DD in the first forecast period was relatively high at 75.38, 88.01, 73.03, and 81.02%, respectively, for Canada, France, Germany, and the United Kingdom. Although these rates decreased in subsequent forecasting periods, they remained significant. However, in the first forecast period, except for Zimbabwe, these shares were relatively low in underdeveloped countries. For example, these shares were 18.10, 18.50, 2.00, and 53.20% for Botswana, Lesotho, Zambia, and Zimbabwe, respectively. In subsequent forecast periods, these shares had an increasing trend in Zambia but decreased in other countries. It was seen that the shares of UD remained at low levels for all countries. It was noteworthy that the share of Germany increased from 8.75% in the first forecast period to 47% in the third and fourth periods. This result indicated that UD values are also crucial in the fight against HIV in the long-term, if not in the short-term, for Germany.

Table 4 Results of variance decomposition for HIV-related deaths (certain countries).
Full size table

In the evaluation of the shares of mean schooling years and per capita income in the variance decomposition of HIVD, it was seen that per capita income had a large share in Botswana, Zambia, and Zimbabwe in both the short and long terms, while mean schooling years had a substantial share for Lesotho. In developed countries, it was seen that these shares remained at much lower levels.

Considering these findings as a whole, it can be understood that reducing drug use is extremely important to lower the rates of HIV infections in developed countries. On the other hand, in underdeveloped Sub-Saharan African countries, where HIV infection is prevalent, it is of great importance to increase the mean schooling year of individuals along with per capita income in addition to the fight against drugs in the fight against HIV.

Panel Granger causality test results

The Fisher test statistic value combining the p-values of countries is derived to assess an overall hypothesis. The critical bootstrap values are received at the 1, 5, and 10% levels based on the bootstrap distribution of Fisher test statistics obtained from 2000 replications.

The results of the Panel Granger causality test for all countries included in the study are given in Table 5. These results showed that there was causality in the Granger sense at a significance level of 1% from mean schooling years, per capita income, and DD toward HIVD. However, it was determined that there was no causal relationship in the Granger sense from UD toward HIVD. The Granger causality test results indicated the short-term causal relationship between two variables, for it is defined based on one-step-ahead predictions. Therefore, the findings indicated that there was a short-term causal relationship between HIVD and the other variables, except for UD. The results of the panel Granger causality test were consistent with the results obtained based on the variance decomposition analyses.

Table 5 Granger causality test results for all countries and certain sub-regions.
Full size table

As with the variance decomposition results, the panel Granger causality test results varied according to regions and countries. In the developed European region, which includes 17 of the countries examined in this study, a Granger causality was found from DD toward HIVD at a 5% significance level. Among the other explanatory variables, no Granger causality was identified for HIVD. This result showed that drug use was the most effective factor in preventing HIVD in developed European countries. In the South America region, which includes eight of the countries examined in this study, the causality relationships in the sense of the Granger test were determined from DD and UD to HIVD at 1 and 10% significance levels, respectively. These results indicated that risk factors, rather than economic factors, impact HIVD in South American countries. In the Central America region, which includes 12 of the countries examined in this study, none of the explanatory variables were a cause of HIVD. In the MENA region, which includes 16 of the countries examined in this study, Granger causalities were found to be at the 1% level from per capita income toward HIVD and from DD toward HIVD, and at the level of 5% from mean schooling years toward HIVD. These results indicated that both per capita income and mean schooling years were effective as socioeconomic factors, and DD was effective as a risk factor for HIVD in the MENA region. In the Far East region, which covers ten of the countries examined in this study, Granger causalities were found in DD and mean Schooling years toward HIVD at levels of 1 and 5%, respectively. These results showed that in the Far East region, drug use as a risk factor and mean schooling years as an educational indicator were the leading indicators in the fight against HIV.

In the Central Asia region, which covers five of the countries examined in this study, only DD had a causal effect on HIVD at a significance level of 1% in the Granger sense. This result showed that drug use was the most effective factor in preventing HIVD in Central Asia. In the Sub-Saharan Africa region, which includes 31 of the countries examined in this study, Granger causalities toward HIVD were identified from mean schooling years at a 10% significance level and from other variables at a 1% significance level. These results indicated that risk factors and economic indicators were the leading determinants of HIVD in the Sub-Saharan Africa region.

The Granger causality test results for certain developed countries and underdeveloped Sub-Saharan African countries where HIVD is high are given in Table 6. For France and Germany as developed countries, the Granger causality relationships from DD toward HIVD had significance levels of 5 and 10%, respectively. A Granger causality from per capita income toward HIVD at 10% was determined for Canada. UD and mean schooling years were not causes of HIVD for the selected developed countries, according to the Granger tests. In Zambia and Zimbabwe, where HIVD values were the highest, HIVD values were affected by both DD and UD according to the findings of Granger causal tests. Furthermore, Granger causalities were found from GNI toward HIVD for Zimbabwe, and MYS toward HIVD for Lesotho.

Table 6 Granger causality test results for certain countries.
Full size table

Discussion

The results of this research were in line with the findings of a large number of other investigations that had been conducted before. There is a strong correlation between a nation’s economic standing, the quality of education it offers its population, risky behaviors, and the manner in which it treats HIV-positive individuals, particularly those who need antiretroviral therapy. Life expectancy, average number of years spent in education, projected number of years, and GNI are all connected with HIV/AIDS mortality, prevalence, and incidence rates (Lou et al., 2014). These findings are also compatible with the results drawn from this paper. The panel regression analysis and variance decomposition showed that income has a significant role in explaining HIVD in both developed nations and developing countries. Poverty makes individuals more susceptible to HIV infection, and both the loss of a productive population as a result of HIV and the high expense of treatment for infected persons further contribute to poverty. This can create a vicious cycle. According to the findings of this research, the economic level of a country’s population has a significant influence on the progression of HIV/AIDS in the long-term.

In the early 1990s, research conducted in countries in Sub-Saharan Africa found that a higher education level was connected with a greater HIV incidence (Kirunga and Ntozi, 1997). In contrast to the findings of this study, the panel regression analysis in this paper demonstrated that education can dramatically lower the risk of HIVD in all countries. According to the findings of this research, income has a significant impact on HIV/AIDS over the long-term. A significant proportion of HIV-positive people live in Sub-Saharan Africa, the world’s poorest region (Piot et al., 2007; Probst et al., 2016). As a result, HIVD continues to be a significant problem in this area. The paper indicates that schooling and income can significantly impact HIVD in African countries. As a result of the extensive transmission of knowledge on HIV prevention, researchers in the 2000s revealed the protective impact of higher education levels (Bärnighausen et al., 2007). In a comparable way, the findings of this research came to the conclusion that lengthening the average amount of time spent in school in both developed and developing nations is an extremely important factor in the fight against HIV. Preventing new HIV infections, increasing HIV testing rates, and encouraging adolescents to engage in safer sexual practices are all benefits of providing sexuality education in schools (George et al., 2022). After analyzing the correlation between years spent in secondary school and the likelihood of contracting HIV, researchers discovered that the more years spent in secondary school, the greater the protective impact against HIV risk, particularly for women. According to the findings of the same research, providing education is an efficient use of resources, and the influence of education indicators on HIV mortality was modest in the short-term but grew in the long-term (De Neve et al., 2015).

This research came to the conclusion that lowering drug usage is particularly essential in reducing the incidence of HIV cases in both industrialized nations and developing countries. As can be seen from variance decomposition that drug use is a significant explanatory variable in developed European countries and Canada. According to the findings of Mathers et al. (2008), it was projected that Eastern Europe, East and Southeast Asia, and Latin America had the largest number of HIV-positive patients that were associated with drug use. When an HIV-positive person also engages in risky behaviors, including smoking, heavy drinking, and drug use, the mortality risk almost doubles (Van and Koblin, 2009; Murray et al., 2012; Petoumenos and Law, 2016). Iakunchykova and Burlaka (2017) conducted a study on Ukrainian sex workers and found that the HIV prevalence rate was 5.6%, that 34.5% of participants used condoms inconsistently, and that those with lower incomes, older ages, drug use histories, histories of physical or sexual abuse, and those who sought clients on highways were more likely to contract HIV. Similar to this research, the paper indicated that unsafe sex or risky behaviors caused a positive impact on HIVD. In addition, political measures might be implemented to lower the prevalence of HIV. People living in low-income and middle-income nations may benefit from these policies, which may take the form of an affordable public health system or treatment options. Antiretroviral medicine has become more affordable, and increased spending on medical research and development can contribute to a decline in the incidence of HIV and related illnesses (Benzaken and colleagues, 2019).

As a consequence, panel regression analyses revealed that the impact of income and education is negative and statistically significant on HIVD, whereas the impact of drug use and unsafe sex is positive. The variance decomposition showed that in the nations, drug use continues to account for a major portion of HIV/AIDS, but income and level of education also play significant roles, and both are likely to experience a rise in importance over the long-term. According to the countries that have a high HIV prevalence, factors such as income and educational attainment are more significant for African countries, while drug usage is a significant factor for developed countries. These conclusions are supported by panel Granger causality tests; however, the extent and power of results differ depending on the country.

Poor nations are characterized by a prevalence of low-income populations. Within these nations, there is a lack of substantial human resources that can effectively counteract the adverse effects of HIV. Furthermore, due to limited financial resources, individuals have challenges in accessing education and medical facilities. Unlike these nations, industrialized nations possess highly advanced medical services that can effectively cater to the needs of individuals living with HIV. Furthermore, these individuals are able to access health services and medications due to the high-income levels in these countries. Hence, advanced countries possess substantial resources to combat HIV, whereas poor and emerging nations continue to grapple with issues such as illiteracy, inadequate education, and limited access to healthcare. Thus, in such nations, the significance of the economy remains crucial for human existence.

Conclusion

This study employed panel data analysis, variance decomposition, and panel Granger causality tests to investigate the influence that factors such as income, schooling, drug usage, and risky behavior as unsafe sex have on the prevalence of HIVD throughout the globe. The investigation included the whole of the globe, including all areas and nations where HIV was frequent. The empirical data from the panel regression model demonstrated that greater levels of DD and UD were associated with statistically significant positive impacts on HIVD, but higher levels of MYS and GNI were associated with statistically significant negative effects. The findings of the variance decomposition analysis showed that the values of DD were the most important factor in determining HIVD in the short-term; however, this influence diminished throughout the course of the study. Though UD’s impact on HIV/AIDS was minimal initially, it became stronger with time. In the short-term, MYS and GNI had only a limited impact on HIVD, but in the long-term, their impacts were more significant.

The variance decomposition analysis produced varying findings depending on the nations and areas investigated. When compared to the levels of the other variables, the proportions of DD within the variance decomposition of HIVD were significantly higher for each sub-region. These rates were relatively high in developed European countries as well as Central Asian Countries. South America and Sub-Saharan Africa were shown to be the locations with the lower rates of these share percentages overall. In addition, the percentage of UD that contributed to the variance decomposition for HIV was low across all of the sub-regions. When looking at the proportion of HIVD in the variance decomposition of MYS, Sub-Saharan Africa stood out as the area with the largest proportion in both the short and long time periods. The findings that were acquired in other areas demonstrated that the use of education as part of the fight against HIV would be helpful over time, particularly in countries that are still in the process of developing.

When looking at the proportion of HIVD in the variance decomposition of MYS, Sub-Saharan Africa was one of the areas that had a larger proportion than the others. Education as a weapon in the battle against HIV can be beneficial over time, particularly in nations that are still considered developing or undeveloped. South and Central America were shown to have the greatest percentages of the HIVD variance decomposition attributable to per capita income. MENA and the Far East were the areas that had the lowest share percentages. Additionally, while the percentage of GNI contributed by all sub-regions was lower in the first year, this percentage significantly grew in the succeeding forecast periods. Based on these findings, it was determined that the amount of income per capita, although to varying degrees of importance depending on the chosen sub-regions, had an especially profound influence on HIVD over the long-term. In the comparisons of the results on the variance decomposition of HIVD by nation, the empirical results revealed that decreasing DD was particularly crucial to lower HIV infection rates in developed countries. It is essential to enhance MYS with GNI together with the battle against drugs in the context of the fight against HIV in underdeveloped nations where HIV is prominent, such as the area of Sub-Saharan Africa.

The findings of the panel Granger causality tests demonstrated that there were causalities at a significance level of 1% from MYS, GNI, and DD toward HIVD. However, it was found that there was not a Granger-type causal connection between UD and HIVD. When the findings of the panel Granger causality test were evaluated according to the different locations, it was shown that drug use was the primary predictor for HIVD in the developed regions of Europe and Central Asia. It was found that the primary variables in lowering the prevalence of HIV disease in South American nations were risk factors, whereas economic factors did not have a significant influence on the epidemic. It was revealed that factors such as per capita income, mean schooling years, and drug usage all had a substantial influence on the prevalence of HIVD in the MENA area. It was discovered that DD was one of the variables that increased the likelihood of HIVD in the Far East area. Moreover, HIVD was impacted in the Far East area as a result of MYS. The findings of the panel Granger causality test conducted in this research demonstrated that the prevention of HIV in the Sub-Saharan African area can be achieved by including all of the factors. In other words, the findings suggested that in economically underdeveloped nations in sub-Saharan Africa, which are known to have high HIV infection rates, equal weight should be given to the promotion of economic growth and to the mitigation of two risk factors in the fight against HIV. The research comes to the conclusion that the varied findings could be attributed to the use of three different empirical approaches. Both wealthy nations and developing countries continue to give HIV/AIDS a significant amount of attention. The education systems in the emerging nations, as well as their levels of income, should be the primary concern. There is evidence that drugs have a role in raising HIVD rates in the industrialized world. Despite this, the struggle against HIV should not be confined to a single nation or area; rather, a global war is required to make significant progress in dealing with this issue.