How learning about infection risk shapes the HIV epidemic

Painted health advices against HIV illness at a wall on side road between Lomé and Kpalimé in Togo, West Africa

The HIV epidemic in Sub-Saharan Africa (SSA) remains a pressing global health challenge. At its core lies a web of risky sexual encounters, often driven by factors like lack of awareness and economic disparities. As risky sexual behavior fuels the spread of the virus, it is critical to understand how this factor shapes the disparities in HIV as the epidemic evolves.

In SSA, education emerges as both a shield and a sword. In Barcelona School of Economics Working Paper 1418, “A Quantitative Theory of the HIV Epidemic: Education, Risky Sex and Asymmetric Learning,” Christian Aleman-Pericon, Daniela Iorio, and Raül Santaeulàlia-Llopis use micro data for more than 30 SSA countries and uncover a U-shaped relationship between education and HIV positivity across the epidemic stages.


Specifically, highly educated individuals are more likely to test positive for HIV during the early and late stages of the HIV epidemic, but not so in the middle stage. 

The authors also find that this pattern between HIV and education is consistent with an also U-shaped relationship between education and risky sexual activity (e.g. number of extra marital partners) across epidemic stages. 

Anchored in this micro evidence, the authors propose learning about the odds of HIV infection from risky sex as a new mechanism behind the HIV epidemic in SSA. They build a model of asymmetric learning in three stages:

The myopic stage

In this stage, ignorance prevails, and people remain unaware of the risks associated with unprotected sex.

The learning stage

When the epidemic matures, educated individuals, attuned to emerging information, adapt faster. They adjust their behavior, reducing risky encounters. Meanwhile, those with limited education lag behind, unaware of evolving risks. This asymmetric learning accounts for the turning point between the myopic stage and the learning stage.

The treatment stage

In this third stage, antiretroviral (ARV) treatment enters the scene, and their introduction can be a double-edged sword, altering once again the course of the HIV infections.

ARVs revolutionized HIV management, they extend lives, and reduce transmission risk. However, as governments roll out them, unintended consequences emerge: When ARVs become more accessible, people perceive HIV as a manageable chronic condition.

This perception may lead to riskier behavior, and HIV exposure rises again (Greenwood et al. 2019), which the authors empirically show is specifically the case among the highly educated individuals.

As ARVs reach new corners of Sub-Saharan Africa, they save lives but inadvertently stoke the flames and in an unequal fashion. 

The authors’ model reveals that the proposed learning mechanism is powerful: a mere 5-year advancement in education could significantly reduce AIDS-related deaths by nearly 45%

Tracking the evolution of the HIV epidemic

To track the evolution of the epidemic, the authors analyze Demographic and Health Survey (DHS) data from approximately 30 Sub-Saharan African (SSA) countries, spanning up to three years of data for each country.

To approach this data limitation, the authors leverage the idea that these surveys were conducted during different stages of the HIV epidemic.

First, the authors define a stylized (or reference) epidemic path (e.g. an average across countries).

Second, they use a transformation of coordinates (a normalization) that maps the country-specific epidemic paths onto the stylized path.

To illustrate this point, Figure 1 visually represents the country-specific and normalized paths for three SSA countries. 

Panel (a) shows the HIV prevalence paths from UNAIDS and corresponding HIV prevalence from the DHS country-year observations for three countries. Panel (b) shows the stylized path of an epidemic.
Figure 1. Demographic Health Surveys (DHS) and Epidemic Stages: An Illustration

After mapping the country-year pairs onto the stylized epidemic path, the authors exploit the rich variation in the position of these pairs across the three critical stages—myopic, learning, and treatment—to unravel the dynamic interplay between education, risk perception, and behavioral adjustments, shedding light on the pandemic’s shifting dynamics.

Figure 2 provides a concise overview of the main empirical findings, illustrating the evolution of the HIV-education gradient and the risky sex-education gradient across the stages of the epidemic.

This graph plots the benchmark estimates of the HIV-Education gradient using the full sample (with year controls).
Figure 2. HIV-Education and Risky Sex-Education Gradients across Epidemic Stages

Panel (a) shows that in the Early Epidemic Stage, each additional year of education is linked to a 1.1 percentage point increase in the likelihood of being HIV positive. Hence, the educated population faces higher risks during this initial phase.

In the Mid-Epidemic Stage, as the epidemic matures, disparities among educational groups gradually fade. The gap narrows, and education seems less predictive of HIV status. At this juncture, the virus affects individuals across educational backgrounds more uniformly.

Surprisingly, in the later Epidemic Stage, the educated individuals once again exhibit a higher likelihood of being HIV positive. Dynamic disparities in HIV go hand in hand with disparities in risky sexual behavior, as shown in Panel (b). The relative number of extramarital partners of educated people compared to non-educated people is positive in the early stages of the epidemic, virtually zero in the middle stages, and reverts to positive in the later stages of the epidemic.

As the epidemic evolves, understanding these dynamics becomes crucial for effective interventions and policy decisions to fight HIV.

Asymmetric learning at work: a quantitative theory of the HIV epidemic

To understand the mechanisms behind the U-shape of relative HIV prevalence among the educated at the three stages of the epidemic, the authors build a model of endogenous HIV infections through risky sex that is exchanged in a centralized market.

The main theoretical contribution of the model is the introduction of asymmetric learning during the middle stage of the epidemic: educated individuals learn about the odds of infection through risky sex faster than their uneducated counterparts.

The dynamics implied by the model are as follows:

At the beginning of the epidemic (myopic stage), people are unaware of the risks of extramarital sex, and the prevalence of HIV is higher among the educated simply because the educated engage in more risky sex.

During the learning stage, the educated learn faster about the odds of infection and reduce their risky sexual behavior at a faster rate than the uneducated, which narrows the gap in HIV prevalence.

Finally, the introduction of ARVs (treatment stage) brings back the positive relative prevalence of HIV among the educated since they have better access to treatment, thus reducing the relative risks of HIV infection.

The model can closely replicate the dynamics of the HIV epidemic, as shown in Figure 3 below. Panels (a) and (b) show new infections and deaths in the data and in the model, whereas panels (c) and (d) show risky sex behavior by education group and beliefs about the odds of infection by education group respectively.

Panels (a) and (b) show the evolution of the new infections and new deaths respectively in our benchmark model and in the data. Regarding with panels (a) and (b) note that the ’observed’ data refers to the UNAIDS model-generated data for the HIV epidemic. Panel (c) shows the model evolution of risky sex across education groups. Panel (d) shows the latent beliefs on the odds of infection due to risky sex by education group.
Figure 3. Model Outcomes Compared to the Data

Finally, using their calibrated model, the authors ask what would have happened if the learning stage of the epidemic, the point at which people start updating their beliefs about odds of infection and adjusting their behavior accordingly, had started earlier.

This figure compares benchmark results with the effects of counterfactual experiments
Figure 4. The Effect of Earlier Learning on Epidemic Outcomes

Figure 4 shows the results from this quantitative experiment. If instead of 1990 (continuous lines), learning had begun in 1984 (thick dashed lines) peak prevalence of HIV would have been 60% lower than the one observed. Similarly, new deaths at the peak would have been approximately 40% lower.

Conclusions

Education plays a pivotal role in shaping the course of the HIV epidemic. Aleman-Pericon, Iorio, and Santaeulàlia-Llopis reveal a fascinating pattern: the prevalence of HIV among educated and less educated individuals follows a U-shape throughout the epidemic. But what drives this arc?

They explain this pattern through a model of asymmetric learning and differences in sexual behavior and access to ARVs.

In the early days of the HIV epidemic, ignorance fueled its spread. People are unaware of the risks of extramarital sex, and the prevalence of HIV among the educated is higher due to higher engagement in risky sexual behavior.

Then, as people begin to learn about the risk, educated ones learn faster and adapt swiftly, narrowing the gap in HIV prevalence until it disappears. Yet, as the epidemic waned, a new twist emerged: the advent of ARVs, which are more accessible to educated people, and the gap reappears.

These findings underscore the critical role of education in combating epidemics. Implementing targeted awareness campaigns early on can significantly reduce infection rates.  According to their simulations, a hypothetical scenario with earlier learning of the risks of infections could have slashed prevalence by 60% and saved countless lives at the peak of the epidemic with a reduction of new deaths of 40%.