Economic shocks and risky sexual behaviours in low- and middle-income countries: a systematic review of the literature

ABSTRACT

We document the extent to which economic shocks through economic incentives explain HIV transmission through risky sexual behaviours in LMICs. We include 35 papers containing 31 unique negative and 11 unique positive economic shocks combined with over 320 health and risky sex outcomes. We find a diverse literature with varying empirical approaches showing increases in risky sexual behaviours are more sensitive to negative shocks than decreases are to positive shocks. Those already at risk of engaging in transactional sex are particularly vulnerable to increasing risky behaviours. Protecting against negative shocks is likely to most effective in preventing HIV transmission.

KEYWORDS:

• Economic shocks
• risky sexual behaviours
• HIV
• transactional sex
• systematic review
• literature review

Acknowledgments

All errors are our own.

Disclosure of potential conflicts of interest

No potential conflict of interest was reported by the author(s).

Notes

¹ Defined as behaviours that increase the likelihood of contracting HIV and other STIs (Dimbuene, Emina, and Sankoh 2014). See Table 3 to Table 5 for all included risky behaviours.

² Transactional sex is defined as: ‘non-commercial, non-marital sexual relationships motivated by the implicit assumption that sex will be exchanged for material support or other benefit’ by Stoebenau et al. (2016) and the STRIVE network. In this paper, we consider ‘transactional sex’ that is motivated by economic needs under the assumption that other transactional sexual relationships are on average unaffected by economic shocks. See Stoebenau et al. (2016) for full explanations of transactional sex, including motivations driven by transactional sex as ‘material expressions of love’ or transactional sex for ‘improved social status’.

³ Sex explicitly in exchange for money or gifts which has many overlapping features with the broad definition of transactional sex and can be viewed as sub-category of all transactional sex.

⁴ Results of sub-group analyses were not collected but important lessons from the sub-group analyses were noted for discussion in the results where appropriate.

⁵ All pertinent estimates were gathered from included papers, i.e. those models where the effect or association of a valid shock was estimated on a valid outcome. Where an author specified multiple versions of models, the estimates from the main model were extracted. There was one replication study, the estimates of which were only collected where they differed in statistical significance or direction from the original study (Baird et al. 2012; Smith, Hein, and Bagenda 2019).

⁶ Forty-seven per cent of papers directly study shocks pre-2017, the median year for published studies, and 56 per cent post-2017.

⁷ Excluding repeat analysis of shocks across papers and combined shocks where analysis is also done of their constituent parts.

⁸ Four of five are studies of the impact of a long-term cash transfer programme in Kenya (Handa et al. 2014, 2017, 2015; Rosenberg et al. 2014).

⁹ Cash transfers that were conditional on STI status were excluded because they change the existing risk/reward decision women are making. In other words, it is impossible to distinguish between the income effect and price effects in these cases. One exception is Kohler and Thornton (2012) where we include the cash transfers, but only include the analysis of outcomes after the cash is paid and the conditionality ceased.

¹⁰ With a total of 16 different cash transfer arms analysed against relevant outcomes, there are many estimates on slightly different sub-samples or definitions of cash transfers that do not occur in studies of naturally occurring shocks which have far fewer estimates.

¹¹ These shocks are: one agricultural, nine health shocks, four household shocks, and three shocks are combinations of various shocks.

¹² See Dupas and Robinson (2012) and Tequame and Tenikue (2017), the former makes inference analysing temporal changes at the person level, addressing well concerns with time trends. The latter attempts to compare, less successfully, those more and less affected by civil unrest.

¹³ Jones and Gong (2019) collect self-reported STI symptoms, all ten other studies collect health outcomes with a biological test, with four studies investigating positive shocks and seven negative.

¹⁴ Jones and Gong (2019) analyse self-reported STI incidence.

¹⁵ Contrary to most other transactional sex outcomes, the recall on this outcome in this study is 7 days which is appropriate for the shock.

¹⁶ A problem with many other observational studies is the recall is long or imprecise and there is potential bias with differential reporting for those who have suffered large economic shocks.

17 The authors based this on the hypothesis that drought differentially affects genders in rural and urban locations.

¹⁸ Reporting only answer to questions on why they have transactional sex.

¹⁹ Child marriage is an important factor within the extensive margin of sex for consumption smoothing too, but this arguably aligns more closely with the forced marriage literature than transactional sex due to the lower agency of the young women at risk. Therefore, we do not expand our discussion further on child marriage.

²⁰ Jones and Gong (2019) also find any sex took place after illness to another household member.

²¹ Cash is given to community groups of the siblings of orphan and vulnerable children in the programme who decide how it is to be distributed. The wider intervention administered alongside the cash transfers has ‘elements of psychosocial support, microfinance, entrepreneurial and other life skills training, cash transfer, and sex education in Kenya’ (Goodman et al. 2015).

²² Statistically significant estimate for the baseline dropouts only.

²³ There was no significant relationship for each gender separately but the size and direction of point estimates for men and women support the combined result.

²⁴ The definition of non-spouse partnerships in this paper includes those ‘monogamous cohabiting union’ and single sexually active individuals, so a portion of this measure is capturing single partner relationships. Therefore, only a small number of these relationships might be considered risky and the result of the estimation could be interpreted as a delay in marriage due to shocks rather than representing an increase in risky behaviours.

²⁵ Note quality limitations.

²⁶ Outcomes include number of partners and number of sex acts, which in the context of commercial sex work fit conceptually with multiple partnerships and other intensity measures.

²⁷ Their shock definition includes several permanent shocks: Loss of job, loss of grant, loss of support from outside the household, and the shock must have had a moderate or severe financial impact which itself introduces potential endogeneity.

²⁸ The analysis is of behaviours since the conditionality has ceased (Mills et al. 2018).

²⁹ However, the result is only significant when those in the conditional arm are included. The conditionality is to remain STI free violates our definition of a shock. However, point estimates for the conditional inclusions and non-conditional only are very similar (0.206 vs 0.216) with a reduced sample size in the non-conditional arm raising the standard error sufficiently to be non-significant.

³⁰ Gong, De Walque, and Dow (2019) measure unprotected sex with non-primary partners without reporting changes in spousal condom use, and Dupas and Robinson (2012) study FSWs and find an increase the number of sex acts per client that are unprotected but equally a reduction in the number of clients that they have. They do not report the overall proportion of sex acts that are now unprotected, which would have been useful additional statistic.

³¹ There is no significant increase in condom use generally, a non-significant increase in frequency of sex overall, so the increase in unprotected sex appears to be capturing an absolute increase in unprotected sex driven by those who never use condoms and, thus, may be more risk loving. The authors note that this was unexpected because condom use was known by the sample to be protective against Ebola.

³² That is, samples that do not recruit based on likelihood or evidence of transactional sex.

³³ In RCTs prevalence is usually preferred to incidence to avoid potential Hawthorne effects from HIV testing at baseline.

³⁴ Largest sample analysed.

³⁵ Replication of (Baird et al. 2012) study.

³⁶ See Lépine, Treibich, and D’Exelle (2020) and Treibich and Lépine (2019).

³⁷ Measures of transactional sex in the literature often conflate it with commercial sex work. This is a limitation of this outcome measure and future research should recognise this distinction in their work (Wamoyi et al. 2019).

³⁸ See Venkataramani and Maughan-Brown (2013).

³⁹ For example, Molotsky (2019) transactional sex from ‘multiple survey questions’ and includes those who ‘reported ever receiving gifts or money from their partners’, implying their data could have been analysed to measure the intensive margin also.

¹ Initial searches included literature review databases, namely Cochrane and Campbell Collaboration, to find similar or duplicate reviews.

² EconLit, Medline, EMBASE.

³ Web of Science, JOLIS, IDEAS/RePEc, J-PAL, World Bank working paper series, Research4Development.

⁴ For example, engaging in commercial sex, condomless sex, anal sex, multiple partnerships, age-disparate relationships etc.

⁵ We differentiate results between condomless sex in the context of family planning and condomless sex in other contexts to reflect the difference.

Additional information

Funding

We gratefully acknowledge funding from the Bloomsbury colleges.

Notes on contributors

Henry Cust

Henry Cust is a PhD student in the Department of Global Health and Development at the London School of Hygiene and Tropical Medicine. He has four years of experience working in low- and middle-income countries in both the international development and academic research sectors. His research lies at the intersection of health and development economics with his thesis focusing on the role of economic incentives in driving risky sexual behaviours and transmission of HIV and other STIs.

Harriet Jones

Harriet Jones is a research degree student at the London School of Hygiene and Tropical Medicine. Her research uses routinely collected clinic data to understand trends in HIV epidemiology and HIV service engagement among female sex workers in Zimbabwe.

Tim Powell-Jackson

Tim Powell-Jackson is an Associate Professor in Health Economics in the Department of Global Health and Development and a member of the Centre for Health Economics in London (CHIL). He has 15 years of research experience working in low- and middle-income countries, applying economics to understand of how health systems work and how they can perform better. Much of his work involves the evaluation of policies that seek to increase patient demand for healthcare or improve the quality of care. He works on topics ranging from health financing and provider payment reforms to management practices and the private sector.

Aurélia Lépine

Aurélia Lépine is an Associate Professor in Health Economics at the Institute for Global Health, University College London. Her research is on the health behaviours of vulnerable women in low-income countries and she currently holds a UKRI Future Leaders Fellowship to explore the role of economic shocks on sexual behaviours of vulnerable women in Africa.

Rosalba Radice

Rosalba Radice is a Reader of Statistics at City University, London. Following the completion of her PhD, she worked as a research assistant and then a research fellow in the Department of Health Services Research and Policy, LSHTM. From 2012 to 2018 Rosalba worked as a Lecturer, Senior Lecturer and Reader in Statistics in the Department of Economics, Mathematics and Statistics, Birkbeck. Her main research interests are in the areas of statistical modelling, copula regression, flexible parametric modelling, distributional regression, penalized likelihood-based inference with applications in epidemiology and economics.