HIV/AIDS transmission : the role of unemployment

(1)

Tracy Johnston (10603506) Supervisor: Erik Plug

HIV/AIDS Transmission

The Role of Unemployment

Tracy Johnston

Abstract

The continued transmission of HIV/AIDS, despite the general increase in incomes and

education levels, signifies that the driving forces behind the epidemic go beyond those

associated with poverty. Using nationally representative data from South Africa, I

create a pseudo panel and use usual panel techniques to examine the impact of

unemployment on HIV transmission. I find that a one percent increase in

unemployment leads to a statistically significant 0.07 and 0.1 percentage point increase

in the prevalence of HIV when observed and unobserved characteristics, respectively,

are controlled for. In a country with an HIV prevalence of 14.4% during the study

period, an increase of this magnitude is not trivial.

(2)

1. Introduction

Since its first official diagnosis in 1981, the prevalence of human immunodeficiency virus (HIV) has increased across the world at an unprecedented rate (Buvé et al. 2002). Estimations by the World Health Organisation (WHO) declare that, since its recognition, HIV has infected a total of 75 million people and has caused an estimated 36 million deaths worldwide. In 2011 AIDS, the disease eventually acquired from an HIV infection, was the 5th leading cause of death and figures indicate that in the year 2012 1.6 million people died of AIDS-related diseases (WHO/UNAIDS). The swift spread of the virus amongst heterosexual partners disproved earlier beliefs that AIDS was a disease exclusive to the male homosexual community and has heightened the urgency to define the origins, determinants, impacts, as well as potential treatments and cures of the virus.

Encouragingly, there has been a reduction in the number of new HIV infections and AIDS-related deaths over the last decade. Between 2001 and 2012, the global number of new HIV infections decreased by 33%, new child HIV infections decreased by 52% and, within the eight-year period of 2004-2012, incidences of AIDS-related deaths declined by 29% (UNAIDS). However, the implications of these figures are that, despite the reduction in HIV transmission rates, the overall number of people living with HIV (PLWH) is increasing (WHO). The availability of affordable antiretroviral therapies (ARTs) has raised the life expectancy and reduced the death rate of PLWH therefore, and because the only way to exit the “HIV pool” is through death, any new infections translate into an increase in HIV prevalence (Whiteside 2002:314). In other words, the increasing number of people living with HIV is a joint combination of reduced HIV infection and mortality rates.

In light of the aforementioned, a magnitude of studies have been undertaken to establish: in which environments the virus is most rampant; which members of society are most at risk of contracting HIV; and what interventions can be implemented to curb its transmission. An extensive literature now exists on HIV/AIDS, with researchers from almost every discipline striving to gain a better understanding of the behavioural, biological and socio-economic drivers and repercussions of the epidemic.

Poverty is alluded to be a primary contributor to the unprecedented spread of the epidemic (Bates et al. 2004) with various elements associated with poverty seemingly fuelling the transmission of the virus. For example, there may be low quality and sparse health care and social services, people may enter the commercial sex industry to make ends meet, low education attainment may render written awareness campaigns ineffective and insufficient local employment opportunities may take migrant workers away from their partners for a long period of time (Bates et al. (2004), Buvé et al. (2002)). As home to the poorest societies in the world, the grip of HIV in Africa further reinforces this theory. Of the 35.3 million people living with HIV in 2012, 25 million (71%) live in Sub-Saharan Africa (UNAIDS, WHO). To further emphasise the severity of the epidemic, the WHO advocates that roughly 1 in 20 people in Sub-Saharan Africa is HIV positive, the majority (58%) of whom are women (WHO). Nevertheless, it is evident that HIV is non-discriminating and not confined to the impoverished and uneducated. Countries such as Botswana and South Africa, for instance, have the highest incomes per capita in Africa and yet are also the location of some of the highest HIV prevalence rates in the world (WHO; Whiteside (2002)). 2 | P a g e

(4)

In view of increasing education and income levels, as well as the influx of various modes of awareness campaigns, more microscopic analyses have been required to understand which factors continue to promote the spread of the virus. There has been an increasing acceptance that the various characteristics of poverty, such has volatile income levels, unemployment, social security and education, act as drivers of the epidemic in their own right - factors which, if found to have a strong effect, may have the potential to undermine many HIV-prevention initiatives.

Early HIV programmes were often criticised for neglecting to consider macroeconomic influences. These programmes were commonly directed at high-risk individuals such as sex workers and substance abusers, however this form of targeting is felt to be imperfect, primarily because it “places the responsibility for change on individuals;…it ignores the social and economic factors that constrain the ability of an individual to change” (Bates et al. 2006:268). It is therefore crucial to identify the potential roles macroeconomic factors, such as unemployment, have on influencing risky behaviours.

The aim of this paper is to provide an economic analysis of the role unemployment plays in the transmission of HIV, an affiliation which is yet to be thoroughly explored in the literature. Unemployment has been proven to cause unhealthy behaviours such as increased substance use, poor eating habits and lifestyles, as well as having negative impacts on mental health, with a strong correlation existing between unemployment and depression/suicide incidences (Dooley et al. 1996). It is therefore plausible for unemployment to have an effect on sexual behaviour/practices and thus HIV prevalence. In addition to this, unemployed individuals have more spare time than their employed counterparts to interact with other people in informal settings. Therefore the main hypothesis of this thesis is that the availability of time increases the likelihood of contracting HIV.

Establishing whether there is a link between the two factors would give interested parties a greater understanding for why, despite widespread awareness, HIV continues to spread. Additionally, this could provide further insight into why some of the most educated and financially stable members of the population may still be at risk, namely young educated men and women between the ages of 19-30 years.

Using South African cross-sectional data spanning the period 2002-2005, a simple linear model is used to identify the relationship between unemployment and HIV. In addition to controlling for observed and unobserved characteristics, I also take into account the likelihood of reversed causality and, therefore, make use of time lags to approximate the direction of causality.

The results indicate that, prior to the inclusion of observed variables, a one percentage point increase in unemployment leads to a highly significant 0.19 percentage point increase in the prevalence of HIV. When observed and unobserved characteristics are included the effect decreases in significance (to the 10% level) and magnitude (o.o6 for observed characteristics and 0.11 for fixed effects). Women are found to have an HIV prevalence rate approximately 5 percentage points higher than their male counterparts and individuals in the 25-29 year old age category are realised to have the highest HIV rates. It is also established that members of society who have achieved at least a secondary level of education have roughly a one percentage point lower HIV rate than those with primary 3 | P a g e

(5)

schooling or less. Performing the regression with lagged unemployment and HIV variables completely eradicates any effects, concluding that unemployment has no effect on the HIV prevalence of cohorts and vice versa.

Due to the nature of the surveys used for the analysis, I can only speculate over the reasons behind the results found. It is tempting to conclude that being unemployed has no effect on sexual behaviours and the rate of HIV transmission, therefore other factors, such as being financially constrained, are the driving forces behind the epidemic. In other words, an individual who receives financial contributions from non-income sources, e.g. parents/relatives, and is therefore not destitute, will not engage in riskier sexual practices if he finds himself unemployed. Using more specified samples, additional identification strategies are undertake to estimate the heterogeneous effects within the population

The rest of this thesis is as follows: Section 2 gives an overview of the HIV/AIDS literature in which unemployment has been taken into consideration as a contributing factor; Section 3 describes the data and methodology used for the analysis; the results are highlighted in Section 4 with further discussion taking place in Section 5. I will then end with some concluding remarks in Section 6.

2. Literature Review

The HIV/AIDS literature is inundated with studies reaffirming the two-way relationship between poverty and HIV: Poverty leads to increased HIV incidences and HIV incidence leads to poverty (Whiteside 2002:330). Additionally, the discussion of unemployment in HIV-related studies is not exceptional to the literature either, however, it should be noted that where unemployment has been mentioned as a co-stimulus of HIV transmission, it is typically done in relation to poverty. In other words, unemployment is often taken as a cause of poverty and it is the condition of being poor that stimulates “risky” behaviours, i.e. behaviours that increase the likelihood of contracting HIV. By and large, it is thus implied that poverty induces risky behaviour and not unemployment itself.

In their analysis of poverty on HIV prevalence in Trinidad and Tobago, for example, Scott et al. (2011) use national unemployment as the explanatory variable in their estimation of poverty on HIV incidence. The assumption made by the authors is that unemployment leads to poverty and therefore it acts as a suitable proxy for poverty (Scott et al. 2011:65). Granger causality tests show that a positive two-way correlation between female unemployment and the incidence of HIV, however the actual magnitude of the relationship could not be determined (Scott et al. 2011:67). What they do find is that “[the] negative impact of HIV/AIDS on socioeconomic status...is slower than the reverse relationship. By way of explanation, when a person is unemployed and poor, he/she is more vulnerable to contracting HIV/AIDS, furthermore, the period within which the poor, unemployed individual may fall prey to the disease is shorter than the time it would take for a non-poor person who contracts the disease to become unemployed and poor” (Scott et al. 2011:63). The result is significant to the literature as it empirically indicates that unemployment, which can affect all members of working age in any society, could be an influential driver of the epidemic in its own right.

(6)

Whiteside (2002) and Nattrass (2004) makes a similar association between unemployment, poverty and HIV. Both authors agree on the non-discriminatory nature of the HIV virus, however believe greater emphasis should be given to the role economic downturns play in HIV transmission. What both authors do in particular, is to pick up on the fact that economic structural adjustments, such as trade liberalisation, led to large changes in the labour market composition of South Africa and simultaneously altered the rate of unemployment. The decline in the primary and secondary sectors and the rise in tertiary industries increased demand for capital and skilled workers and reduced demand for unskilled labour. This ultimately resulted in an overall decline in formal employment and, subsequently increasing poverty, which they believe raised the rate of HIV transmission (Whiteside 2002, Nattrass 2004). Illustrated in the example below, Whiteside further emphasises that employment status and poverty are not the only forces behind the HIV epidemic, and that “economic, social and cultural factors” also play a major role on an individual’s sexual behaviour:

Nevertheless, in light of this unemployment-poverty-HIV connection, both authors believe reducing unemployment will reduce the spread of HIV by eliminating the various HIV transmitting influences spurred by poverty.

Realising the potential drive behind macroeconomic factors, Oster (2013) analysed the relationship between exports and HIV incidence. Her estimations suggest that doubling exports results in an increase of new HIV infections by 10% - 70% (Oster 2013:1027). Extensive research has been done recognising the significant role the transportation industry plays in the spread of epidemic – an industry which would flourish with increased exports (Vass 2005:568). Truck drivers spend long periods of time away from home and on the road, which increases their likelihood of engaging in commercial sex. Oster (2013) therefore endorses the use of HIV-prevention programmes and states that given the large effect of exports on HIV transmission, programmes focussing on high-risk groups may be far more vital than originally anticipated.

Undoubtedly, one of the primary reasons that so many studies cite poverty as a main determinant of infection is due to the higher prevalence rates amongst poorer members of all societies, including those in the West (Whiteside 2002:314). Hallman (2005:37), in her behavioural study of men and women aged 14-24 years in the Kwazulu-Natal province of South Africa, found that having a low income increases the likelihood of an earlier sexual debut and of having multiple sexual partners. Moreover, it was established that low income reduced the probability of condom use and discussing safe-sex practices with partners, and

“....a truck driver on any of the major routes in Africa may be away from home for long periods. He might have sex with a commercial sex worker because he is bored, he feels his job is dangerous and he deserves some compensation, he is frequently away from his wife and family, he experiences peer pressure from his fellow drivers to engage in this activity and he has the necessary money. The commercial sex worker, on the other hand, is driven by poverty and the need to feed her family.” – Whiteside (2002:317)

(7)

increased a woman’s prospects of being forced to have sex (7 percentage points higher than women from wealthier backgrounds). All of these factors raise the likelihood of acquiring the virus (Hallman 2005:43).

Ibrahim et al. (2008) complementarily find the reversed relationship in their London-based study. They find that people living with HIV in London are more likely to suffer economic and social hardships. HIV infected individuals in London, particularly ethnic minorities, usually have low incomes and higher unemployment rates. Furthermore, the authors find that unemployment is significantly positively related to the number of years since diagnosis – thus ratifying the bi-causal relationship of HIV and poverty (Ibrahim et al. 2008:620).

Several studies find an association between unemployment and/or irregular income and HIV prevalence. Studies by Wilson (2012) and Burke et al. (2013) find that income fluctuations has a notable impact on HIV incidences. During the copper boom of 2003 to 2008, employment in Zambia’s copper industry rose by 180% and was estimated to have reduced transactional sex and multiple sexual partners in mining cities by 10% and 20% respectively (Wilson 2012). Comparable figures were found when households situated along copper transport routes, where the trucking industry would bias the results, were excluded. Correspondingly, using rainfall as an exogenous determinant of income in rural Africa, Burke et al (2013) estimate that episodes of drought increased the rate of HIV infection by 11% in HIV-endemic rural communities and that 20% of the variation in HIV prevalence in Africa could be explained by income shocks. One suggestion for their findings is that these negative income shocks increased the rate of transaction sex, though they could not dismiss other potential factors that would increase transmission rates, e.g. dropping out of school and migration (Burke et al. 2013:30).

To reinforce the idea that poverty is not the lone determinant of increased HIV transmission, a study conducted in Zambia established that, in the absence of employment and recreational facilities, people may engage in sexual activities to help pass time (Fetters et al. 1998), consequently increasing the probability of HIV transmission. Such findings highlight the importance of increasing youth employment opportunities; not only for increased taxes and poverty alleviation, but for the benefit of public health and HIV control. Bärnighausen et al. (2007), using longitudinal HIV-monitoring data from a poor rural South African community, make the unforeseen discovery that members of households making up the middle 40% of relative wealth had a 72% higher risk of acquiring HIV than individuals from the poorest 40% of the community (Bärnighausen et al. 2007:7). They further find that the risk of contracting HIV was reduced by 7% as a result of one extra year of education.

In summary, these studies suggest that risky sexual behaviour, and subsequently HIV rates, increase during economic hardships regardless of education status and HIV knowledge. Moreover, it is further implied that macroeconomic factors are indeed correlated to HIV incidences, which may have some serious implications on the efficacy of HIV-prevention initiatives - initiatives which primarily focus on altering individual sexual behaviour without considering that the economic situation of individuals may restrain their ability to alter this behaviour. The effectiveness of these approaches could be significantly reduced if macro and micro factors which influence the unemployment rate have strong linkages to HIV spread.

(8)

3. Data & Methodology

3.1 Data Description

The data used for the analysis is from The ‘South African National HIV Prevalence, HIV Incidence, Behaviour and Communication Survey’ (SABSSM), a nationally representative cross-sectional survey conducted and acquired from the Human Sciences Research Council (HSRC), South Africa. The survey has so far been conducted in the years 2002, 2005, 2008 and 2012, however at the time of writing, only the first two waves are available for scrutiny. The forthcoming analysis will therefore be conducted using only the 2002 and 2005 datasets.

The sample for the surveys were selected using the 2001 South Africa census’ enumeration areas as Primary Sampling Units (PSUs), for which there are 1000 spanning the entire country. 11 (15) households in the 2002 (2005) survey were then chosen in each enumeration area, within which a randomly selected, eligible individual would be selected for the interview. Individuals aged 2 years and over were eligible for the survey, however for the purpose of this thesis only the youth and adult data are used, i.e. persons aged 15 years and older, resulting in a sample size of 9,788 and 16, 398 individuals in 2002 and 2005 respectively.

The survey consists of biographical characteristics such as the respondent’s age, gender, race, and marital, education and employment status. It also comprises of information regarding: marital and sexual practices; media and communication; knowledge and views of HIV and HIV related factors such as government policies, campaigns, stigma, male circumcision, condoms and sexual debuts etc.; as well as health related questions, including substance use and mental health (Shisana and Simbayi, 2002). Additionally, a non-invasive HIV test (an oral swab) was conducted on willing participants. 6,118 respondents (62.5%) in 2002 and 12,032 (73.4%) in 2005 consented to HIV testing in 2002 and 2005, respectively.

Despite the range of topics and quality of survey, there are some concerns regarding the data. Firstly, the surveys are not completely uniform in their content and structure, however for the purpose of this analysis the dissimilarities do not greatly impact the investigation between unemployment and HIV. For example, the 2005 survey goes into greater detail regarding knowledge and perceptions of specific HIV campaigns/programmes and has more explicitly defined sections on alcohol consumption, condoms, and other topics. Secondly, the availability of only two surveys limits the certainty of any findings.

3.2 Data Processing

To acquire a better understanding of the unemployment-HIV relationship, only observations containing information for both variables are retained for analysis. I further exclude people over the age of 65 years as they are more likely to be retired rather than unemployed. Additionally, respondents whose HIV test results are listed as ‘indeterminate’ or ‘no specimen was received’ are also excluded from the sample. The reason for this is that, firstly, identifying the correlation between unemployment and HIV would not be observable using these individuals; secondly, these individuals constitute a small proportion of the 2002 7 | P a g e

(9)

population (0.7%) therefore their exclusion would not bias the analysis; and thirdly, no similar HIV status entries are found in the 2005 data. The remaining observations were expanded using individual sample weights already provided in the datasets. The final sample sizes for the 2002 and 2005 surveys are therefore 21,664,579 and 28,756,985 respectively.

As Verbeek (2007:2) clearly states, “...the major limitation of repeated cross-sectional data is that the same individuals are not followed over time” and therefore usual panel data techniques cannot be applied. To mitigate this, I created a pseudo panel, an estimation strategy proposed by Deaton (1985), whereby individuals are grouped together into cohorts based on a common characteristic which can then be traced in each survey, e.g. birth year, gender, etc. Cohort averages then serve as pseudo panel observations upon which panel data methods can be utilised. The cohorts in this thesis are grouped according to four characteristics, namely, province, gender, age category and education status.

Due to issues with sample size, even after expansion, several variables have to be generated to make both datasets comparable and to achieve an equal number of cohorts for each year. For example, zero ‘no schooling’ observations in various provinces in the 2002 dataset made it unfeasible to group individuals according to education levels at that level of education. Similar problems exist at the tertiary level. Therefore two broad education levels are used – ‘primary schooling or less’ and ‘secondary schooling or more’. Five age categories are also defined, with the first three age categories increasing in five-year increments (15-19, 20-24 and 25-29) and the final two covering the ages 30-45 and 46-65. The reasons for this are three-fold. Firstly, youth are over represented in the sample; secondly, due to lower HIV levels as age increases, grouping these ages together will allow for any correlation between HIV and unemployment to be highlighted; and thirdly, based on previous literatures, it is expected that the individuals between 20-29 will mostly be affected by unemployment (youth unemployment) and therefore the correlation between HIV and unemployment will be at its greatest in this age range.

Unemployment is defined as individuals who are out of work, willing to work and actively seeking employment, i.e. those deemed officially unemployed. This means that students, housewives and those who are unemployed but are not searching for a job, all of whom I have retained in the sample, will not considered as unemployed. I therefore generate an unemployment binary variable which gives a value of 1 if the individual is

officially unemployed and a value of 0 otherwise.

In total, 180 cohorts are established, containing an average of 120,3591_{and 159,761} observations each in the 2002 and 2005 data respectively, with which panel data methods can be performed. Table 1 provides general descriptive statistics at the national level for the two years of data. Even though the overall HIV prevalence and unemployment rate remain more or less unchanged between the two surveys, when the statistics are disaggregated into various categories, it is clear that the epidemic is not homogenous within the country.

In both years: women experience the highest HIV prevalence and unemployment rates, the 25-29 year old age category has the highest HIV and those with primary schooling or less comprise the highest levels of unemployment. Women make up approximately 64%

1_{The number of observations per cohort range from 1,394 to 654,483 in the 2002 data, and from 3,417 to}

897,415 in the 2005 data.

8 | P a g e

(10)

of the unemployed population and 65.5% of the people living with HIV, similar to WHO estimations. Provincially, in both 2002 and 2005, Mpumalanga has the highest HIV prevalence while Limpopo is home to the highest rates of unemployment in both surveys.

In 2002, the Eastern Cape and Northern Cape had the lowest HIV and unemployment rate respectively, however by 2005, Western Cape had successfully achieved a population with the lowest HIV and unemployment rate. As expected, individuals in the 20-29 age band continue to have the highest HIV prevalence and unemployment rates in both surveys.

For the context of this study, it is encouragingly to note that there appears to be some correlation between HIV and unemployment. For six of the nine provinces, the change in unemployment and HIV share the same sign. In other words, both factors move in the same direction. This also appears to be the case for women at the national level and for two of the five age categories. Based on this, I will now proceed to identify their relationship.

Table 1: HIV prevalence and unemployment rates

2002 2005

Total Total

# HIV(%) U(%) # HIV(%) U(%)

Total 21,664,579 14.44 25.98 28,756,985 14.45 27.5 Gender: Male 9,087,171 11.59 21.37 13,460,448 10.67 22.01 Female 12,577,408 16.50 29.31 15,296,537 17.78 32.34 Province: Western Cape 2,604,489 11.39 22.71 2,726,403 2.85 13.46 Eastern Cape 3,018,879 9.53 28.94 4,079,336 13.67 31.55 Northern Cape 517,782 11.13 18.49 531,160 8.15 24.94 Free State 1,550,319 18.72 24.09 1,813,769 18.18 28.90 Kwazulu Natal 3,822,074 15.29 26.99 6,580,345 20.06 27.06 North West 1,574,767 13.37 27.03 2,354,023 16.08 33.18 Gauteng 4,705,397 17.20 21.76 5,744,207 12.83 25.18 Mpumalanga 1,608,654 19.61 21.24 1,944,353 20.69 28.99 Limpopo 2,262,218 12.23 38.51 2,983,389 10.40 34.44 Age: 15-19 3,809,433 6.15 13.78 4,949,051 5.91 9.55 20-24 3,212,793 12.85 42.67 4,488,539 15.08 40.67 25-29 2,620,182 28.08 41.22 3,687,586 23.51 45.08 30-45 6,984,500 18.49 28.04 9,515,439 20.16 31.73 46-65 5,037,671 9.03 13.78 6,116,370 6.56 15.20 Education status: Primary or less 7,480,388 13.58 29.55 8,472,426 15.69 28.76 Secondary or more 14,184,191 14.90 24.10 20,284,559 13.93 27.00 9 | P a g e

(11)

3.3 Methodology

As stated previously, repeated cross-sectional data disables us from calculating individual effects of unemployment on HIV prevalence over time, therefore I create pseudo panel with 180 cohorts grouped according to four characteristics (province, gender, age category and education status). Grouping according to four characteristics is done so that cohorts are as detailed as possible which will allow me to see the heterogeneity between individuals more clearly. Cohort averages are calculated for HIV prevalence, unemployment and other control variables, resulting in the following linear model:

HIVct = α + βUnemploymentct + εct (1)

where HIVct denotes the HIV prevalence of cohort c in time t and Unemploymentct is the cohort rate of unemployment and εct is the error term. Since the number of observations in each cohort is large, as is the case in this study, the intercept, αc, is assumed to be the same in both periods, but not across cohorts. The coefficient β measures the effect of a change in unemployment on the HIV prevalence in the same cohort. One of the primary concerns of estimating β from (1) is that the error term contains all the remaining observable and unobservable variables that effect the dependent variable. If one of these variables is also correlated with unemployment, the regression will suffer from omitted variable bias. To solve for this I include observable characteristic to the regression, resulting in:

HIVct = αc + βUnemploymentct + γGenderct + фAgecategory ct + δEduc. ct + εct (2)

where γ, ф and δ estimate the impact of gender, age and education status on HIV prevalence respectively. Henceforth, gender, age and education will be defined as vector ‘X’ in the forthcoming regressions.

As suggested by Moffit (1993) and Ridder and Moffit (2009), if the asymptotics of the pseudo panel are such that number of individuals in the survey (and within each cohort) tend to infinity and if the number of cohorts is fixed, then performing a fixed effects estimation which will provide us with a consistent β estimate. I therefore perform the following regression which will control for any unobservable omitted variable bias:

∆HIVct = α’c + β’∆Unemploymentct + Ω∆Xct + ∆εct (3)

For the reason that cohorts are based on all the available observable covariates and that the intercept is time invariant, these two parameters are are reduced to zero when fixed effects is performed.

A second concern in any study of unemployment and HIV is the issue of reverse causality. Instead of unemployment causing HIV prevalence to rise, the relationship may run in the opposite direction. For example, an HIV diagnosed individual may be restricted from entering the work force for health or discriminatory reasons. An empirical model with lagged parameters will additionally be estimated to test for this:

(12)

HIVct = α’’c + β’’Unemploymentct-3 + Ω’’Xct + εct (4) and the reverse relationship,

Unemploymentct = α’’’c + β’’’HIVct-3 + Ω’’’Xct + εct (5)

Regressions (4) and (5) make use of lagged unemployment and HIV variables to identify the impact of unemployment and HIV in 2002 on HIV and unemployment in 2005. It is expected that changes in unemployment will precede changes in HIV (regression (4)), producing a significant β’’ coefficient. If, however, if the reverse is true and changes in HIV precede changes in unemployment then the coefficient β’’’ in (5) will be greater than that in (4). Running the above regression will shed light on the impact of unemployment on HIV prevalence.

The format and number of surveys available severely restricts the variety of estimation methods plausible. Only one time lag is possible, but given that there is a three year period between the surveys, it may provide us with some insight which will form the basis of future investigations.

4. Results

Table 2 presents the OLS estimations for the effect of unemployment on HIV prevalence. Without including observable characteristics, column (1) suggest that a 1 percentage point increase in unemployment positively raises HIV prevalence by 0.19 percentage points. Though the increase appears to be small, in reality it is actually a considerable amount and is highly significant at the 1% level. Strictly speaking, if 14% of the population has contracted HIV, a 10 percentage point increase in unemployment raises the HIV prevalence by essentially 2 percentage points – that constitutes a 15% increase in HIV.

As stated previously, this first result is not very informative. Not only does it suffer from omitted variable bias, it additionally does not inform us of the heterogeneity within the population. By adding variables for gender and education status to the regression (columns 2 and 5), the impact of unemployment decreases slightly to 0.18 percentage points, however it continues to be highly, statistically significant. Women are found to have an HIV prevalence 4.45 percentage points more than their male counterparts (significant at the 1% level) and column (5) indicates that higher educated individuals have a 0.26 percentage point lower HIV prevalence than those with primary schooling or less, though this is not statistically different from zero.

The impact of unemployment on HIV practically disappears when age is controlled for. A 1 percentage point increase in unemployment now leads to 0.07 percentage point increase in the prevalence of HIV (significant at the 10% level). With the 15-19 year old age category as the baseline age, it appears that 20-24 year olds have a 5.5 percentage point higher HIV prevalence than their younger population. As expected, the 25-29 year old age bracket has the highest HIV prevalence (14.72 percentage points more) compared to the baseline.

(13)

Table 3 presents the results from performing a fixed effects regression (column 2), as well as carrying out the regression with lagged unemployment (columns 3 and 4). For ease of comparing regression outputs, column 1 in table 3 is a replication of column 4 in table 2. Column 2 shows that with fixed effects, the coefficient for unemployment increases to 0.11 and is statistically significant at the 10% level. Having grouped the cohorts based by time invariant variables i.e. by province, gender age category and education status, all these variables are captured by the fixed effects.

Table 2: Effect of unemployment on HIV prevalence

(1) (2) (3) (4) (5)

OLS

HIV

Unemployment

ct

0.191* 0.177* 0.072*

0.066*

0.176***

(5.86)

(5.44)

(1.94)

(1.73)

(5.33)

Gender

4.447* 5.228*

5.273* 4.454*

(2.89)

(3.86)

(3.89)

(2.89)

Age category:

20-24 years old

5.287**

5.464**

(2.25)

(2.31)

25-29 years old

14.543* 14.723*

(6.16)

(6.20)

30-45 years old

12.071* 12.164*

(5.57)

(5.60)

46-65 years old

1.783

1.772 (0.85)

(0.85)

Education Status

-1.069

-0.259

(-0.79)

(-0.17)

_cons

8.294*** 2.037

-2.754

-1.125

2.444 (6.71)

(0.82)

(-1.11)

(-0.35)

(0.71)

N

360

Note: t statistics in parentheses, * p<0.05, ** p<0.01, *** p<0.001

Columns (3) and (4) show the effect of unemployment in 2002 on the HIV prevalence in 2005, excluding and including observable variables respectively. Similar to column (1) in table 2, the exclusion of observable characteristics gives a very statistically significant unemployment estimate of 0.15. This would mean that unemployment in one year would cause a 0.15 percentage point increase in HIV prevalence three years later. As before, the inclusion of observed characteristics completely diminishes the effect - unemployment in year t-3 has no effect on HIV in year t.

(14)

Table 3: With Fixed & Lagged Effects

(1) (2) (3) (4)

OLS FE OLS OLS

HIV HIV HIV HIV

Unemploymentct 0.066* 0.112* (1.73) (1.97) Unemploymentct-3 0.154*** -0.023 (3.61) (-0.48) Gender 5.273*** 6.919*** (3.89) (3.93) Age category: 20-24 years 5.464** 9.527*** (2.31) (3.17) 35-29 years 14.723*** 17.735*** (6.20) (5.91) 30-45 years 12.164*** 14.787*** (5.60) (5.33) 46-65 years 1.772 0.472 (0.85) (0.17) Education Status -1.069 -3.088* (-0.79) (-1.75) _cons -1.125 10.624*** 9.440*** 0.272 (-0.35) (5.95) (5.97) (0.06) r2 0.02 0.07 0.33 N 360 360 180 180

The reversed relationship produces results with a similar pattern. Table 4 shows the a 1 percentage point increase in HIV raises the unemployment rate by 0.14 percentage points (column 2). This is somewhat intuitive given that HIV-diagnosed individuals may become too ill to work or are perhaps discriminated against in the job market once employers discover their status. Again, when the lagged HIV variable is used, the impact of HIV on unemployment disappears (column 4). In other words, there is no effect of HIV on unemployment after a three year period.

(15)

Table 4: Impact of HIV on the Unemployment Rate

(1) (2) (3) (4)

OLS OLS OLS OLS

U Rate U Rate U Rate U Rate

HIVct 0.385*** 0.138* (5.08) (1.88) HIVct-3 0.360*** -0.042 (3.28) (-0.46) Gender 6.624*** 7.209*** (3.25) (3.24) Age category: 20-24 years old 27.889*** 30.944*** (8.71) (8.88) 35-29 years old 27.087*** 32.544*** (8.00) (8.68) 30-45 years old 13.316*** 19.550*** (4.04) (5.42) 46-65 years old -1.994 1.08 (-0.63) (0.31) Education Status -7.122*** -6.313*** (-3.56) (-2.90) _cons 24.239*** 15.173*** 25.441*** 12.917** (14.08) (3.17) (11.88) (2.47) r2 0.056999 0.511677 N 360 360 180 180

In summary, the results indicate that the effect of unemployment on HIV is noteworthy and very significant. However, when observed variables are taken into account the impact disintegrates. I also find that there is no relationship between unemployment on HIV after a three year period. Speculation of these results will be conducted below.

5. Discussion

The results show that when observed and unobserved characteristics are taken into consideration, the effect of unemployment on HIV is a 0.07 and 0.11 percentage point increase respectively (significant at the 10% level). The positive relationship conforms to the findings of previous studies, suggesting that unemployment is correlated with HIV. Based on subsequent results, however, it is tempting to conclude that the state of being unemployed has no effect on an individual’s sexual behaviour and therefore such behaviour is induced by other contributory factors – be they social, economic or cultural. Due to the nature of the surveys, I can only speculate over which other driving forces could be at play, such as actual financial hardship.

(16)

For example, in highlighting the role public welfare schemes play, Campbell et al. (2005), find that low job prospects and the absence of unemployment benefits motivate some women to risk HIV transmission in order to fall pregnant, thus gaining access to child welfare schemes. Nattrass (2004) also concludes that, in the absence of other welfare schemes or low-wage employment opportunities, current HIV-prevention programmes will be weakened by macro factors such as poverty and unemployment. Therefore a closer look at a country’s welfare system could shed some light on the continued HIV transmission.

To get more insight into the role finances play in relation to unemployment and HIV, a wealth index would have been a useful tool to use, however questions and responses regarding income/wealth were not comparable between the surveys, consequently removing the possibility to analyse its interaction. Nevertheless, education status has been used in other studies as a proxy for wealth, based on the assumption that poorer individuals will be unable to afford higher education (especially tertiary education).

The study of Fetters et al. (2001) posits that more spare time and the lack of recreational activities leads to a higher probability of contracting the virus. Though it would be complex to assess this idea using unemployed individuals (difficulties excluding factors such as poverty), disaggregating the employed population, in particular looking at HIV discrepancies between those that work full-time and part-time, provides us with some insight regarding the issue of time availability and HIV contraction. Moreover, given the evident heterogeneity within the population, identification strategies with more specified samples, e.g. by gender and age, further enlighten us to the unemployment-HIV relationship.

The results from conducting the various identification strategies mentioned above are presented in Table 5, highlighting the heterogeneity within society. For comparative ease, the first entry in the table are the results for the original base sample. Table 5 shows that increases in male unemployment have no effect when their HIV prevalence, whilst a one percentage point increase in female unemployment increase their HIV prevalence by a significant 0.12 percentage points when observed characteristics are taken into consideration. The parameter increases in magnitude and significance (0.2 percentage points, significant at 1 percent level) when unobserved characteristics are controlled for. This clearly highlights the gender bias of HIV epidemic and the vulnerability of women.

When the younger and older members of society are studied, the results are not significant when observed and unobserved covariates are included. An anomaly is that lagged unemployment reduced the HIV prevalence by a weakly significant 0.18 percentage points for the young population. One possible reason for this finding may be that unemployed individuals in the sample age range (15-29 years old) could still live at home or with older relatives, which can limit the ability to interact with peers whenever one may desire to. Moreover, members of the first two five-year age brackets (15-19 and 20-24 years old) are over-represented in both surveys, which makes the likelihood of still living at home more viable.

Table 5: Heterogeneous Effects of Unemployment on HIV prevalence 15 | P a g e

(17)

(1) (2) (3) (4) (5)

OLS OLS FE OLS OLS

Lagged Unemployment

without X with X without X with X

HIV HIV HIV HIV HIV

Base Sample 0.191*** 0.066* 0.112* 0.154*** -0.023 Gender: Men 0.04 -0.023 0.005 0.048 -0.013 Female 0.274*** 0.124** 0.201*** 0.192*** -0.08 Age: Young(15-29) 0.190** 0.06 0.126 -0.016 -0.181* Old (45-65) 0.14 0.018 -0.084 0.175** 0.102 Education/Wealth: No schooling 0.363** 0.505 0.537** 0.255 -0.226 Tertiary 0.223** 0.176 0.027 0.216** 0.11 Employment: Part-time -0.027 -0.053 -0.261** 0.087 0.072 Full-time 0.031 0.006 0.014 0.044 0.066

A one percentage point increase in unemployment increases HIV prevalence for individuals with no schooling by a significant 0.5 percentage points compared to an insignificant 0.18 percentage points for those with tertiary education. Unemployment therefore has a much greater impact on the HIV transmission rate of the poorest members of society. When I conduct the regressions using both ‘no schooling’ and ‘tertiary’ education simultaneously, and another using all schooling levels (Table A7 and A8 respectively (in appendix)), coefficients suggest that individuals with higher education have a higher HIV prevalence compared to those with no schooling. Nevertheless, only those with secondary level education (and thus with middle incomes) have a significant 5.86 percentage point higher HIV prevalence. Based on this, it appears that financial constraint is not a prevailing characteristic for high-risk individuals.

Interestingly, when estimating of the impact of employment status on HIV prevalence, working full-time has no effect on HIV prevalence, whilst part-time employment reduces the HIV prevalence by 0.26 percentage points (significant at the 5% level) when observed and unobserved cohort characteristics are controlled for. When the same regressions are run using all employment forms, the HIV prevalence of those who work part-time is not statistically difference to those who are unemployed. Full-time workers, on the other hand, have a highly significant 6.7 percentage point reduced HIV prevalence compared to their unemployed counterparts (Table A11 in appendix). I therefore conclude that the availability of time, as suggested by Fetters et al. (2001), also plays a leading role in the transmission of the virus.

As stated previously, using the lagged variables of HIV and unemployment completely eradicates an effects of unemployment on HIV and vice versa. This is somewhat surprising, especially the lagged HIV on unemployment relationship, as it is at least expected that an increase in HIV prevalence in one year would increase the unemployment rate the following year – as diagnosed individuals become more ill. Earlier mentioned, Ibrahim et al. (2008) find a strong correlation between unemployment and the number of

(18)

years since diagnosis . Again, I can only speculate that the primary reason for my zero-effect findings lies in limitations with the data.

I believe that the first limitation stems from the frequency of the survey. A small percentage of those who are unemployed in 2002 are likely to be unemployed three years later. It is particularly difficult to disentangle the extent unemployment in 2002 affected HIV prevalence in 2005 without having more information of regarding influential factors. To mitigate this, annual data would be more efficient in identifying a relationship similar to those found in existing literatures. Moreover, if individuals were unemployed for this amount of time, they may be discouraged from applying for jobs. Such individuals would, therefore, not be included in my estimation as I only use the official unemployment rate as opposed to the expanded unemployment rate which includes those who are ‘unemployed and not looking for work’. However, I have run the same regressions again using expanded unemployment and find that the effects remained unchanged. I therefore conclude that the three year lag plays a significant role to play in my findings.

The implementation of an antiretroviral therapy (ART) rollout programme by the South African government could be a reason for HIV having no impact on unemployment three years later. Between 2003 and 2004 the government of South Africa began providing free ARTs to HIV sufferers with a CD4 count of less than 350 (CDC 2011). Prior to this treatment was at the great expense of the individual which had debilitating consequences financially and medically. Those who could not afford the drugs would probably be less able to perform well at work three years later as their immune system begins to breakdown. The provision of these drugs enabled many HIV infected people to continue working efficiently and further reduced the impact of becoming poor as a result of medical payments. It is therefore extremely difficult to establish whether, in the three year period, people who were unemployed in 2002 became HIV positive, were able to access treatment and managed to secure a job by 2005. I consequently conclude that any analysis using lagged variables would not provide us with a reliable estimate – a lot can happen in three years.

What I do conclude from this analysis is that there is a positive relationship between the unemployment and HIV. Despite issues concerning the quality and frequency of the data, the results obtained are still noteworthy. This finding explains why the spread of HIV still continues among the educated youth of today. Moreover, it should encourage governments to pay particular attention to youth unemployment, not only for poverty and tax revenue reasons, but more importantly for public health concerns.

6. Concluding Remarks

Despite rising incomes, higher levels of education and intense awareness campaigns in South Africa, the persistent transmission of HIV, especially among the youth, continues to be a public health concern. Though earlier literature has attributed the spread of the virus primarily to poverty, global statistics and evidence from more recent studies enlighten us to the fact that HIV is non-discriminatory and that other influential forces also play a significant role in its transmission. Examples of this being economic booms and income fluctuations caused by exports, rainfall and copper prices - all of which been deemed as HIV influencers.

(19)

The connection between unemployment and HIV transmission is not new to the literature, however it is often discussed in connection with poverty. The aim of this study is therefore to look specifically at the role of unemployment on HIV transmission out of the poverty context. Previous research has found a positive causal relationship between the two entities, however issues with data and reverse causality often hamper the abilities of studies to make an efficient estimate of the effects. I therefore contribute to the literature by attempting to get a more meaningful estimate of the effect unemployment has on HIV transmission.

Using cross-sectional data from South Africa spanning the period 2002 to 2005, I generate comparable cohorts in each year in order to create a pseudo panel with which panel methods can be used. I find that unemployment leads to a significant (at the 10% level) 0.07 percentage point increase in HIV prevalence. When a fixed effects estimation is undertaken, thus controlling for unobserved variable bias, the effect increases to 0.11 percentage points (also significant at the 10% level). Though the result may seem small, it is significant and particularly meaningful when taking into consideration what this translates to in terms of actual numbers and HIV transmission. The positive result further ratifies findings found in previous works. I use lagged HIV and unemployment variable to estimate the causal relationship between the two factors, only to find that there is no effect of unemployment on HIV and vice versa. I speculate that one of the main reasons for this result is the quality and quantity of the data.

The availability of only two years of data makes finding a causal relationship very challenging. Moreover, the fact that there is a three year gap between the first and the second survey, means that is highly unlikely for a relationship to be seen. Someone who is unemployed in 2002 is unlikely to be unemployed in 2005. I therefore conclude that the use of lagged variable, regardless of the cohort strategy used, would provide us with unreliable results.

Using more specified samples, I undertaking the same study again to explore the heterogeneity within the population. I find that female unemployment plays a significant role in HIV transmission: an increase of one percentage in female unemployment leads to a 0.2 percentage point increase in HIV transmission. Using education status as a proxy for wealth, an increase in unemployment for those with no schooling raises the HIV prevalence by a significant 0.5 percentage points when observed and unobserved characteristics are controlled for. However, when all education levels are compared, people with a middle-level income (and have thus achieved high school qualifications) have a significant 5.9 percentage point higher HIV prevalence than those with no schooling.

By focusing on the disparities in HIV prevalence of full-time and part-time employment separately, a one percentage point increase in full time employment does not have a significant effect on HIV prevalence of the full-time population, while an equivalent increase in part-time employment will reduce the prevalence of HIV by a significant 0.26 percentage points for those working part-time. When the impact of all employment status’ on HIV prevalence are estimated simultaneously, it becomes apparent that compared to their unemployed counterparts, those working part-time have a 1.8 percentage point lower HIV prevalence (though this is not statistically significant), whilst those working full-time have a highly statistically significant 6.7 percentage points reduced HIV prevalence. 18 | P a g e

(20)

Compared to those who are unemployed, part-time workers have a higher income but still have more spare time compared to those who are employed full-time, therefore the availability of time appears to have a role to play, irrespective of financial situations.

What this study brings to light is that financial constraint and poor education are not the only drivers of the HIV epidemic, and that the availability of time has a role to play. Moreover, the results found do confirm that there is a positive relationship between unemployment and HIV transmission. In their mission to counter the spread of HIV, governments should therefore strengthen efforts to tackle joblessness in the country. Based on this study, the creation of jobs, particularly for young, well educated, women would have a positive impact on controlling the epidemic in South Africa.

19 | P a g e

(21)

7. References

Bärnighausen, T., Hosegood, V., Timaeus, IM. & Newell, M-L.(2007) “The Socioeconomic Determinants of HIV Incidence: Evidence from Longitudinal, Population-Based Study in Rural South Africa”. AIDS. November, 21(Suppl. 7), S29-S38.

Bates, I., Fenton, C., Gruber, J., Lalloo, D., Medina Lara, A., Bertel Squire, S., Theobald, A., Thomson ,R. &Tolhurst R (2004) “Vulnerability to Malaria, Tuberculosis, and HIV/AIDS Infection and Disease. Part 1: Determinants Operating at Individual and Household Level”. The

Lancet Infectious Diseases; Vol.4, Issue 5, pp 267-277

Burke M, Gong E& Jones K (2012, March) ”Income Shocks and HIV in Africa”. In Midwest

Economic Development Conference. Mimeo, pp 1-71

Buvé, A., Bishikwabo-Nsarhaza, K. &Mutangadura G (2002) The Spread and Effect of HIV-1 infection in sub-Saharan Africa. The Lancet, 359: 2011-2017.

Case, A. &Paxson, C. (2013) “HIV Risk and Adolescent Behaviours in Africa”.American

Economic Review, 103(3): 433-438

CDC (2011) “CDC's HIV/AIDS Care and Treatment Programs in South Africa: Antiretroviral Treatment and Services: Prolonging and Improving the Quality of Lives”. Centre of Disease Control

and Prevention. Available from:

http://www.cdc.gov/globalhealth/countries/southafrica/what/antiretroviral-treatment.htm

Deaton, A (1985) “Panel data from time series of cross-sections”. Journal of Econometrics. Vol.30, Issue 1-2, pp 109-126.

Dixon, S., McDonald, S. & Roberts, J (2002) “The Impact of HIV and AIDS on Africa’s Economic Development”. BMJ: British Medical Journal, Vol.324, No.7331, pp. 232-234.

Dooley, D., Fielding, J. & Levi, L (1996) “Health and Unemployment”. Annual Review of Public

Health. Vol. 17, pp. 449-465

Evian, C., Slotow, S., Rosen, S., Thea D., Fox, M., Macleod, B. & Simon, J. (2001). “Anonymous unlinked HIV prevalence surveys in 43 workplaces in South Africa, Botswana and Zambia 2000-2001”. Provisional results. Presentation

Fetters, T., Mupela, E. &Rutenberg N. (1998) “Youth talk about sexuality: a participatory assessment of adolescent sexual and reproductive health in Lusaka, Zambia”. Lusaka: CARE Zambia and Population Council, 1998.

Fortson, J.G. (2008) “The Gradient in Sub-Saharan Africa: Socioeconomic Status and HIV/AIDS”.

Demography.Vol.45, No.2, pp. 303-322.

Hallman, K. (2005)“Gendered Socioeconomic conditions and HIV risk Behaviours Among Young People in South Africa”. African Journal of AIDS Research, Vol.4, Issue 1, pp. 37-50.

Hunter, M. (2007) “The Changing Political Economy of Sex in South Africa: The significance of unemployment and inequalities to the scale of the AIDS pandemic”. Social Science & Medicine. Volume 64, Issue 3, pp 689-700.

(22)

Ibrahim, F., Anderson, J., Bukutu, C. &Elford, J (2008) “Social and Economic Hardships of People living with HIV in London”. HIV Medicine. Volume 9, Issue 8, pp. 616-624.

Kalichman, S.C., Simbayi, L.C, Kagee, A., Toefy, Y., Jooste, S., Cain, D. & Cherry, C. (2006) “Associations of Poverty, Substance Use and HIV Transmission Risk Behaviours in Three South African Communities”. Social Science and Medicine. Vol. 62, Issue 7, pp.1641-1649.

Moffit, R. (1993) “Identification and Estimation of Dynamic Models with a Time Series of Repeated Cross-Sections”, Journal of Econometrics, Vol.59, pp 99-123Nattrass, N. (2004) “Unemployment and Aids: the social-democratic challenge for South Africa”. Development Southern Africa, Vol. 21, No.1, pp 87-108.

Oster, E. (2013) “Routes of Infection: Exports and HIV Incidence in Sub-Saharan Africa”.Journal of

the European Economic Association, Vol.10, No.5, pp. 1025-1058. Business Source Premier,

EBSCOhost.

Ridder, G. and Moffit, R. (2009) “The Econometrics of Data Combination”. Cited in: Verbeek, M (2007) “Pseudo panels and repeated cross-sections”. In: The Econometrics of Panel Data. Advanced Studies in Theoretical and Applied Econometrics. Vol. 46 (2008), pp 369-383. Scott E, Simon T, La Foucade A, Theodore K, Gittens-Baynes K-A (2011) “Poverty, Employer and HIV/AIDS in Trinidad and Tobago”. International Journal of Business and Social Science, Vol.2, No. 15, pp. 63-70

Shisana, O. & Simbayi, L.C. (2002) “South African HIV/AIDS, Behavioural Risks, Sero-status, and Mass Media Impact Survey (SABSSM)”, 2002. > [Data set]. SABSSM 2002 Adult-youth. Cape Town South Africa: Human Sciences Research Council, HIV/AIDS, STI and TB programme (HAST) [producer] 2002, [distributor] 2011

Shisana, O. Simbayi, L.C., Rehle, T., Parker, W., Zuma, K., Bhana, A., Connolly, C., Jooste, S., Pillay, V. et al. (2005) South African National HIV Prevalence, HIV Incidence, Behaviour and Communication Survey, 2005. [Computer file]. SABSSM 2005 Adult-youth. Cape Town South Africa: Human Sciences Research Council, HIV/AIDS, STI and TB programme (HAST) [producer] 2005, [distributor] 2012

Vass, J. (2005) “The Relationship Between Labour Market Dynamics and HIV/AIDS Prevalence in South Africa – A literature Review”. South African Journal of Economics, Vol. 73., Issue Supplement s1, pp 564-577.

Verbeek, M (2007) “Pseudo panels and repeated cross-sections”. In: The Econometrics of Panel Data. Advanced Studies in Theoretical and Applied Econometrics. Vol. 46 (2008), pp 369-383. Whiteside, A (2002) “Poverty and HIV/AIDS in Africa”. Third World Quarterly, Vol. 23, No. 2, pp 313-332.

Wilson, N. (2012) “Economic Booms and Risky Sexual Behaviour: Evidence from Zambian Copper Mining Cities”. Journal of Health Economics, Vol. 31, Issue 6, pp. 797-812

(23)

8. Appendix

Table A1: Effect of gender and unemployment on HIV prevalence (Male)

(1) (2) (3) (4) (5)

OLS OLS FE OLS OLS

without X with X without X with X

HIV HIV HIV HIV HIV

Unemploymentct 0.04 -0.023 0.005 (0.83) (-0.47) (0.05) Unemploymentct-3 0.048 -0.013 (0.77) (-0.22) Age category: 20-24 years old 3.867 4.992 (1.40) (1.39) 25-29years old 14.318*** 14.217*** (5.23) (3.94) 30-45years old 16.868*** 18.640*** (6.55) (5.43) 46-65 years old 5.236** 4.866 (2.06) (1.41) Education Status -1.379 -3.687 (-0.83) (-1.62) _cons 10.024*** 5.683* 10.949*** 9.315*** 7.823* (6.18) (1.69) (4.65) (4.63) (1.69) r2 0.000033 0.006666 0.335639 N 180 180 180 90 90 22 | P a g e

(24)

Table A2: Effect of gender and unemployment on HIV prevalence (Female)

(1) (2) (3) (4) (5)

OLS OLS FE OLS OLS

HIV HIV HIV HIV HIV

Unemploymentct 0.274*** 0.124** 0.201*** (6.48) (2.23) (2.64) Unemploymentct-3 0.192*** -0.08 (3.31) (-1.15) Age category: 20-24 years old 6.909* 16.024*** (1.87) (3.39) 25-29years old 14.715*** 23.140*** (3.91) (4.95) 30-45years old 6.903** 12.092*** (2.07) (2.87) 46-45years old -1.79 -4.087 (-0.58) (-1.03) Education Status -1.255 -2.695 (-0.63) (-1.06) _cons 7.686*** 9.222** 10.147*** 10.975*** 14.456*** (4.34) (2.37) (3.82) (4.59) (2.92) r2 0.072683 0.110774 0.381692 N 180 180 180 90 90

Table A3: Effect of age and unemployment on HIV prevalence (Young)

(1) (2) (3) (4) (5)

OLS OLS FE OLS OLS

HIV HIV HIV HIV HIV

Unemploymentct 0.190** 0.06 0.126 (2.40) (0.77) (1.15) Unemploymentct-3 -0.016 -0.181* (-0.14) (-1.81) Gender 10.567*** 14.707*** (5.00) (4.87) Education Status -0.308 -5.441* (-0.15) (-1.73) _cons 7.157** -3.943 9.286** 14.686*** 6.286 (2.48) (-0.81) (2.49) (3.37) (0.84) r2 0.036667 0.000559 0.442617 N 72 72 72 36 36 23 | P a g e

(25)

Table A4: Effect of Unemployment on HIV prevalence (Older)

(1) (2) (3) (4) (5)

OLS OLS FE OLS OLS

HIV HIV HIV HIV HIV

Unemploymentct 0.14 0.018 -0.084 (1.51) (0.18) (-0.47) Unemploymentct-3 0.175** 0.102 (2.18) (1.09) Gender -1.344 -1.293 (-0.78) (-0.74) Education Status -4.913** -2.86 (-2.55) (-1.42) _cons 5.954*** 12.074*** 8.971*** 4.100*** 8.359** (3.89) (3.49) (3.45) (3.14) (2.41) r2 0.006143 0.122319 0.184832 N 72 72 72 36 36

Table A5: Effect of unemployment on HIV prevalence (no schooling)

(1) (2) (3) (4) (5)

OLS OLS FE OLS OLS

HIV HIV HIV HIV HIV

Unemploymentct 0.363** 0.505** 0.537* (2.55) (2.38) (1.97) Unemploymentct-3 0.255 -0.226 (0.77) (-0.30) Gender 9.408 17.201 (1.38) (1.07) Age category: 20-24 years old 2.052 -15.053 (0.19) (-0.45) 25-29 years old 10.813 -31.301 (0.85) (-0.99) 30-45 years old 10.862 -28.895 (0.89) (-0.74) 46-65 years old 13.281 -38.818 (0.94) (-0.84) _cons 0.537 -26.284 -5.972 10.438 24.816 (0.09) (-1.56) (-0.55) (0.76) (0.41) r2 0.301486 0.06862 0.507557 N 20 20 20 10 10 24 | P a g e

(26)

Table A6: Effect of unemployment on HIV prevalence (Tertiary)

(1) (2) (3) (4) (5)

OLS OLS FE OLS OLS

HIV HIV HIV HIV HIV

Unemploymentct 0.223** 0.176 0.027 (2.50) (1.47) (0.13) Unemploymentct-3 0.216** 0.11 (2.56) (0.87) Gender 1.802 3.606 (0.69) (1.08) Age category: 20-24 years old -4.717 -3.027 (-1.22) (-0.62) 25-29 years old 5.094 5.906 (1.33) (1.26) 30-45 years old 0.419 0.307 (0.13) (0.07) 46-65 years old -2.834 -3.873 (-0.92) (-1.03) _cons 3.680** 1.985 6.195* 3.433* -0.505 (2.12) (0.47) (2.17) (1.94) (-0.09) r2 0.001745 0.450606 0.855136 N 20 20 20 10 10 25 | P a g e

(27)

Table A7: Effect of unemployment on HIV prevalence (No schooling and tertiary education comparison)

(1) (2) (3) (4) (5)

OLS OLS FE OLS OLS

HIV HIV HIV HIV HIV

Unemploymentct 0.323*** 0.344*** 0.495** (4.36) (3.13) (2.64) Unemploymentct-3 0.335** 0.069 (2.22) (0.31) Gender 4.752 11.249 (1.33) (1.57) Education status: Tertiary 0.904 -11.875 (0.21) (-1.36) Age category: 20-24 years old -3.64 -4.184 (-0.67) (-0.38) 25-29 years old 3.754 -8.557 (0.69) (-0.78) 30-45 years old 2.597 -8.15 (0.47) (-0.72) 46-65 years old 1.791 -13.771 (0.30) (-1.11) _cons 2.193 -6.801 -2.092 4.748 7.215 -0.88 (-0.93) (-0.42) (0.96) (0.47) r2 0.268416 0.214534 0.428011 N 40 40 40 20 20 26 | P a g e

(28)

Table A8: Effects of unemployment on HIV prevalence (all schooling levels)

(1) (2) (3) (4) (5)

OLS OLS FE OLS OLS

HIV HIV HIV HIV HIV

Unemploymentct 0.385*** 0.373*** 0.400*** (7.85) (6.08) (3.47) Unemploymentct-3 0.392*** 0.224* (4.56) (1.76) Gender 4.203*** 8.544** (2.58) (2.70) Education Status: Primary 3.15 -2.045 (1.29) (-0.42) Secondary 5.863** -1.556 (2.35) (-0.31) Matric 3.202 -4.107 (1.26) (-0.80) Tertiary 1.615 -8.207 (0.56) (-1.44) Age category: 20-24 years old -4.007 -0.093 (-1.47) (-0.02) 25-29 years old 6.268** 4.548 (2.40) (0.89) 30-45 years old 6.010** 5.186 (2.44) (1.07) 46-65 years old 1.087 -4.027 (0.43) (-0.79) _cons 2.575 -8.033** 2.15 3.857 -2.403 -1.55 (-2.31) (0.65) (1.38) (-0.34) r2 0.197425 0.302568 0.467567 N 100 100 100 50 50 27 | P a g e

(29)

Table A9: Effect of employment on HIV prevalence (Part-time employment)

(1) (2) (3) (4) (5)

OLS OLS FE OLS OLS

HIV HIV HIV HIV HIV

Part.timect -0.027 -0.053 -0.261** (-0.35) (-0.71) (-2.45) Part.timect-3 0.087 0.072 (0.78) (0.71) Gender 5.627*** 6.889*** (4.19) (3.97) Age category: 20-24 years old 7.551*** 8.763*** (3.56) (3.21) 25-29 years old 17.034*** 16.717*** (7.84) (6.01) 30-45 years old 13.596*** 13.974*** (6.20) (4.93) 46-65 years old 2.115 -0.075 (0.96) (-0.03) Education level -1.507 -2.978* (-1.13) (-1.73) _cons 14.157*** 0.112 15.923*** 13.268*** -0.369 (13.73) (0.03) (16.1) (10.26) (-0.09) r2 0.032551 0.003385 0.330846 N 360 360 360 180 180

Table A10: Effect of employment on HIV prevalence (Full-time employment)

(1) (2) (3) (4) (5)

OLS OLS FE OLS OLS

HIV HIV HIV HIV HIV

Full.timect 0.031 0.006 0.014 (0.84) (0.13) (0.21) Full.timect-3 0.044 0.066 (0.96) (1.18) Gender 5.847*** 7.730*** (3.93) (4.05) Age category: 20-24 years old 7.303*** 8.273*** (3.41) (2.99) 25-29 years old 16.529*** 15.541*** (7.19) (5.13) 30-45 years old 12.976*** 12.244*** (5.15) (3.67) 46-65 years old 1.504 -1.445 (0.63) (-0.45) Education level -1.605 -3.733** (-1.15) (-2.01) _cons 13.266*** -0.199 13.634*** 12.872*** -0.537 (11.37) (-0.06) (8.37) (8.78) (-0.13) r2 0.00024 0.005169 0.334272 N 360 360 360 180 180 28 | P a g e

(30)

Table A11: Effect of employment on HIV prevalence (employment status comparison)

(1) (2) (3)

OLS OLS OLS

without X without X with X

HIV HIV HIV

Employmentct -3.279*** (-3.04) Employment Status: Part-time -1.617 -1.794 (-0.74) (-0.87) Full-time -6.576*** -6.714*** (-3.05) (-3.20) Gender 3.943** (2.33) Age category: 20-24 years old 0.629 (0.23) 25-29 years old 12.298*** (4.45) 30-45 years old 7.350*** (2.69) 46-65 years old -2.65 (-0.92) Education Status 0.46 (0.13) _cons 21.601*** 17.796*** 7.607 (9.31) (11.76) -1.18 N 520 520 520 29 | P a g e

HIV/AIDS transmission : the role of unemployment