HIV/AIDS Knowledge and Behavioural Change Among Migrant Workers: Evidence from a Cross-Border Intervention in India, Bangladesh, and Nepal

Abstract

This paper evaluates the Enhancing Mobile Populations’ Access to HIV and AIDS Services, Information and Support (EMPHASIS) programme implemented by CARE International across Bangladesh, India, and Nepal. Using individual-level data, we estimate the programme’s impact on HIV-related knowledge and preventive behaviours among migrant workers. Results show that participation in EMPHASIS significantly increased correct knowledge of HIV transmission, reduced misconceptions, and improved partner communication. These informational gains translated into higher condom use and fewer unsafe sexual practices, with stronger effects among women. The findings provide evidence that peer-led, information-based interventions can improve health behaviours among mobile populations. Integrating such approaches with gender empowerment and mobile health services offers a promising model for addressing HIV vulnerability in cross-border migration settings.

1. Introduction

International migration has long been a central topic in political discussions in both the Global North and South. In 2024, the number of international migrants worldwide reached approximately 304 million, representing approximately 3.7 per cent of the global population [1]. Although significant media attention has focused on South–North migration, movements within the Global South, known as South–South migration, are also on the rise. Indeed, in 2020, the number of people born in the Global South living within the same region was estimated at 98 million compared to 92.4 million residing in the Global North [2].

Irregular migration, unlike legal migration, is difficult to quantify due to its elusive nature. The last global estimates available suggest that there were at least 50 million irregular migrants in 2010, many of whom rely on smuggling networks [3]. Furthermore, the International Labour Organization (ILO) [4] reported that approximately 27.6 million people are victims of forced labour worldwide, with a significant portion suffering from human trafficking for sexual and labour exploitation, particularly in the Asia–Pacific region.

Within South Asia, major labour migration corridors include Bangladesh–India and Nepal–India, where irregular migration—often of an economic nature—and human trafficking are prevalent issues. This is particularly concerning given the large-scale trafficking of women and children across these borders [5,6].

The intersection of labour migration and health is critical. Migrant workers often face barriers to accessing healthcare services, including lack of insurance, language difficulties, and social stigma. These challenges can lead to increased vulnerability to infectious diseases, such as the Human Immunodeficiency Virus (HIV). In particular, the precarious living and working conditions faced by migrants frequently lead to high-risk behaviours, including unprotected sex and substance use, increasing the risk of HIV transmission [7].

The Acquired Immune Deficiency Syndrome (AIDS), caused by HIV infection, remains a significant public health challenge worldwide. Since the epidemic began in the 1980s, approximately 88.4 million people worldwide have been infected with HIV, resulting in approximately 42.3 million deaths [8]. Although India, Bangladesh, and Nepal are classified as low prevalence regions—each with less than 1 per cent of the adult population infected—recent estimates indicate that in 2023, approximately 6.6 million people in the Asia–Pacific region lived with HIV, with 2.5 million residing in India alone [8]. These statistics point to an ongoing regional challenge where mobility, limited access to health services, and stigma intersect to sustain transmission risks.

Behaviours and attitudes toward HIV transmission are heavily influenced by the knowledge and information of people about the disease [9,10,11]. However, there is limited research on HIV knowledge and its spread among migrant workers, a particularly vulnerable group. A few exceptions are the case studies [12,13] on Mexican migrant workers in the United States; on rural Chinese migrants workers in urban China [14,15,16]; on Burmese workers in Thailand [17]; on Tajik workers in Russia [18]; and on Nepalese and Indian domestic migrant workers [19,20], respectively. Yet, despite these contributions, quantitative assessments of how targeted interventions influence migrants’ HIV-related knowledge and behaviours remain scarce.

Understanding the factors associated with migrants’ HIV/AIDS knowledge, attitudes, and practices is crucial from a public health perspective, especially given the mobility and social vulnerabilities of this population.

This paper contributes to the literature by evaluating the impact of a large-scale project—Enhancing Mobile Populations’ Access to HIV and AIDS Services, Information, and Support (EMPHASIS)—implemented by CARE International from 2009 to 2014. The initiative aimed to improve HIV/AIDS knowledge and awareness among Bangladeshi and Nepali migrants working in India. Using Propensity Score Matching (PSM), we estimate the average treatment effects of EMPHASIS on migrants’ HIV/AIDS knowledge at both the source and destination locations and examine how increased knowledge influences behavioural changes such as more consistent condom use during sexual intercourse with regular partners or spouses.

Our findings demonstrate the positive effects of the intervention, emphasising the importance of information campaigns to raise awareness of HIV risks, transmission, and misconceptions. Such efforts can reduce unprotected sexual practices among vulnerable migrant populations in South Asia.

The remainder of this paper is organised as follows. Section 2 reviews the related literature. Section 3 provides the programme description. Section 4 describes the data used in the analysis and presents the empirical methodology. Section 5 presents the results. Section 6 discusses the main findings and Section 7 presents the results from the sensitivity analysis and limitations. Finally, Section 8 concludes the paper.

2. Related Lliterature

The link between HIV risk and migration is well documented. In general, the HIV risk for migrants is found to be a combination of risky behaviours and socioeconomic and cultural factors [21]. Risky behaviours that increase migrants’ vulnerability to HIV are linked, among others, to poverty [22,23,24,25] and separation from the migrant’s home community or spouse, which can cause feelings of loneliness and stress [26,27]. In a study of Mexican migrant men, Munoz-Laboy et al. [28] concludes that feelings of loneliness contribute to risky sexual behaviour and that the more lonely men perceive themselves, the less likely they are to use condoms with their new partners in the US. Other studies also find that separation from spouses leads migrants to resort to sex workers [23,29,30,31], which poses a significant risk of HIV or other sexually transmitted infections (STIs). Inaccurate knowledge of HIV and STIs and ignorance of the risks of contagion are found to be prevalent among migrant workers in Bangladesh [32] and India [33], with women in particular in a vulnerable position to contagion from returning husbands [34,35].

The different culture and living environments at destination sites might also encourage migrant workers to engage in sexual behaviour not accepted at home. In a study of rural migrants to Kathmandu, Puri et al. [19] finds that sexual behaviour becomes more risky—casual encounters and irregular condom use—among both men and women when they migrate outside of their community. Poudel et al. [30] also finds that Nepalese migrants in India are often pressured to drink and frequent brothels by peers, with increased alcohol consumption identified as a determinant of risky sexual behaviour, namely the failure to use a condom [36].

Research on the impact of HIV intervention programmes has focused largely on non-mobile populations compared to mobile populations. Kirby et al. [37] reviews studies on the impact of school education programmes on sex and HIV and finds that two-thirds of 83 studies reported positive effects in reducing risky behaviours and increasing the use of contraceptives and condoms in young adults under 25 years old. Dupas [38] uses a randomised experiment to test whether increased information could change sexual behaviour among adolescents in Kenya and reports positive behavioural results.

Few systematic reviews highlight the importance of information and behavioural interventions in reducing the risks of HIV contagion. Fonner et al. [39] conducts a systematic review of studies that look at the effect of voluntary counselling and testing (VCT) on reducing HIV risk behaviour in developing countries. They authors find that access to VCT reduces HIV risky behaviours although with limits. Wariki et al. [40] reviews experimental and quasi-experimental studies that assess the effect of behavioural interventions on the reduction in HIV infections among female sex workers and their clients in low- and middle-income countries. The authors find that behavioural interventions result in improvements in many outcome variables such as incidence and prevalence of STI and reported condom use. Another systematic review [41] of experimental studies on structural- and community-level interventions campaigns concludes that while these interventions increase the likelihood of condom use and knowledge of HIV, there is no clear evidence that these changes lead to decreases in HIV infections.

Experimental and quasi-experimental studies assessing the impact of HIV interventions for migrant workers are scarce, largely due to the difficulty of reaching mobile populations with irregular status. One of the very few studies is [42] that targets Nepalese migrants in India with letters containing information on HIV and its transmission. The authors find that participants improved their awareness of HIV and other STIs, were less likely to visit brothels, and used condoms more frequently than the control group.

In the context of rural-to-urban migration in China, Li et al. [43] conducts a social cognitive experiment that focuses on raising awareness of the vulnerability and severity of HIV infection. The study includes awareness training on HIV and risky sexual behaviour followed by the use of condoms after 6 and 12 months of the intervention. The authors find that the treated group was more likely to use condoms, relative to the control group, and had better knowledge about HIV and related perceptions.

Mass media campaigns have also been used to raise awareness of HIV for hard-to-reach populations such as undocumented migrants. Olshefsky et al. [44] designs an awareness programme that consists of broadcasting on Spanish-language radio, print media, and a toll-free HIV testing referral hotline to promote HIV testing among Latinos living on the California–Mexico border. The authors report an increase in HIV testing as a result of the campaign. In Bangladesh, there are HIV awareness programmes in the mass media, but limited access to television and radio among the rural poor restricts their effectiveness [45].

3. Programme Description

Enhancing Mobile Populations’ Access to HIV and AIDS Services Information and Support (EMPHASIS) was a regional programme implemented by CARE International in Bangladesh, India, and Nepal from August 2009 to July 2014. The main objective of the programme was to reduce HIV- and AIDS-related vulnerabilities among mobile populations migrating across the Bangladesh–India and Nepal–India borders [46].

EMPHASIS targeted both source and destination locations that experienced high labour migration. In Bangladesh, interventions were concentrated in Jessore and Satkhira districts; in Nepal, in Kanchanpur and Achham; while the main destination locations included Mumbai, Delhi, and West Bengal in India [47,48,49]. These sites were major cross-border mobility corridors identified by the programme.

By April 2014, the programme reached an estimated 351,423 migrants and family members through community-level interventions, awareness activities, and health service linkages. The programme referred 21,577 individuals for sexually transmitted infection (STI) treatment, voluntary counselling, and HIV testing. Capacity-building efforts trained more than 10,486 stakeholders, including healthcare providers, NGO staff, local authorities, and law enforcement representatives, to support safer mobility, HIV prevention, and migrant rights [46].

A set of interventions was implemented across six strategic areas (see

Table 1. Summary of EMPHASIS Programme Interventions.

Category	Specific Intervention	Immediate Objective	Target Population	Geographic Coverage
1. HIV Prevention & Behavior Change	Peer education and outreach through trained migrant peers	Increase HIV/STI knowledge and promote safer sexual practices	Mobile men and women, spouses, sex workers	Source, transit, and destination sites across Bangladesh–India and Nepal–India corridors
	Community awareness sessions and small-group discussions	Normalize HIV conversation; encourage testing and disclosure	Migrants, spouses, community members	All project locations
	Distribution of IEC materials (posters, leaflets, flip charts in Bengali, Hindi, Nepali)	Provide low-literacy HIV/STI information	Mobile populations and general community	Border crossings, workplaces, transport hubs
	Condom promotion and demonstration	Increase condom use rates	Migrant workers, sex workers	Source and destination areas
	Community Support Groups (CSGs)	Facilitate mutual learning and solidarity	Migrants and spouses	Source communities (Jessore, Satkhira, Kanchanpur, Achham)
2. Health Service Access & Referral	Cross-border referral system between HIV testing and counseling centers	Ensure continuity of care for migrants	PLHIV and migrants needing HIV services	Bangladesh–India and Nepal–India corridors
	Mobile health camps (testing, counseling, STI treatment)	Provide accessible health services to migrants	Migrant workers, local communities	Destination and border towns
	Linkages with national HIV programs (NACO–India, NASP–Bangladesh, NCASC–Nepal)	Institutionalize service continuity	National and district health facilities	Regional and national levels
	Partnerships with local NGOs and clinics	Expand outreach and service points	Local NGOs, health providers	All project areas
3. Livelihood Support	Savings and remittance counseling	Promote financial security of migrant households	Migrants and spouses	Source communities (Bangladesh, Nepal)
	Livelihood and financial literacy training	Strengthen resilience and safe remittance practices	Returnee migrants, spouses	Bangladesh and Nepal source sites
	Legal awareness and rights education	Empower migrants on labor and residency rights	Migrant workers, employers	Destination cities (Delhi, Mumbai, Kolkata)
4. Gender Empowerment & Anti-Stigma	Women’s support groups/solidarity circles	Increase self-confidence and agency of spouses	Female spouses and returnee women	Jessore, Satkhira, Kanchanpur
	Gender and HIV training	Address gender norms and reduce vulnerability	Male and female migrants	All countries
	Stigma-reduction workshops for service providers and community leaders	Reduce discrimination against migrants and PLHIV	Health staff, landlords, employers	Destination and transit areas
5. Capacity Building & Coordination	Training for NGOs and health workers on cross-border case management	Improve service quality and confidentiality	NGO staff, health providers	Bangladesh, India, Nepal
	Cross-border coordination committees and networks	Facilitate cooperation between border authorities and health offices	District officials, border security, NGOs	Petrapole–Benapole, Gaddachauki, Banbasa
	Joint review meetings and learning exchanges	Share best practices and monitor progress	CARE country teams, local partners	Regional secretariat (Kathmandu)
6. Advocacy & Policy Influence	Policy dialogue on migration, HIV, and rights	Improve enabling environment for mobile populations	Policymakers, donors, regional bodies	National and SAARC levels
	Dissemination of research (baseline, endline, case studies)	Generate evidence for scaling and replication	Governments, NGOs, researchers	Regional forums and conferences

). At the prevention and community level, EMPHASIS promoted peer education, outreach, and behaviour change communication to improve HIV knowledge and condom use among migrants, spouses, and sex workers [47]. In health access, cross-border referral systems and mobile health camps were established to ensure continuity of care and linkages to national HIV/AIDS programmes [46]. Livelihood components supported migrants’ families at source locations through financial literacy, remittance counselling, and linkages with banks and cooperatives [48]. Gender and anti-stigma interventions aimed to empower women, particularly spouses of migrants, by facilitating solidarity groups, gender training, and engagement with employers to improve rights and entitlements at workplaces in destination cities [47,49]. Capacity building and cross-border coordination aimed to strengthen partnerships among government, NGOs, and border authorities, while advocacy and evidence generation informed national and regional policy dialogues on migration, HIV, and human rights [46].

4. Materials and Methods

4.1. Data

The EMPHASIS programme included an impact evaluation protocol that gathered both qualitative and quantitative data from vulnerable migrant populations at their points of origin and destination. The analysis in this paper draws on individual-level data collected from Nepali and Bangladeshi migrants in Indian cities—Mumbai, New Delhi, and Kalkota—as well as from circular or returnee migrants and their spouses in Nepal’s Kanchanpur and Achham districts and in Bangladesh’s Jessore and Satkhira districts. The baseline survey was conducted between November 2010 and March 2011, with a follow-up endline survey carried out after four years of programme implementation, between February and March 2014.

At source, the programme focused on districts with high emigration rates and long-established cross-border mobility patterns. In Bangladesh, the districts of Jessore and Satkhira were selected for their proximity to the Benapole–Petrapole border crossing—South Asia’s busiest land port—and for their dependence on migration as a livelihood strategy. Both districts experience acute livelihood constraints, including salinization of agricultural land from shrimp cultivation, recurrent flooding, and limited non-farm employment opportunities, which have intensified household reliance on migration [47,50]. In Nepal, the districts of Kanchanpur and Achham were chosen for their economic marginalisation, geographic remoteness, and entrenched patterns of seasonal and circular migration to India via border points such as Gaddachauki and Banbasa. These areas were also prioritised under Nepal’s national HIV and AIDS strategy due to the elevated risk of HIV transmission among returnee migrants and their spouses [48,51].

At destination, the programme was implemented in Delhi, Mumbai, and Kolkata, India’s principal urban centres for Nepali and Bangladeshi migrants. These cities were identified through a mapping analysis of migrant concentrations and HIV-related vulnerabilities [46,49]. Within each city, sub-locations were identified through a structured mapping process that classified areas by migrant density, occupational profile, and housing conditions, ensuring diversity across socioeconomic contexts. These sites were selected for their large migrant inflows, high levels of HIV/STI vulnerability, and the presence of CARE partner organisations and local health facilities that enabled consistent service delivery and monitoring.

Due to the transient nature of migrant populations, tracking the same individuals over time was not feasible. Instead, the endline survey focused on locations where EMPHASIS provided services for several years, designated as treatment sites. Additionally, similar locations with comparable socioeconomic and migratory characteristics—yet not receiving EMPHASIS services due to budget limitations—served as control sites. To minimise spillover effects, treatment and control locations were separated by 2 to 6 km.

The inclusion criteria for participant selection were established to ensure that the survey captured individuals most representative of the mobile populations at risk of HIV vulnerability across the Nepal–India and Bangladesh–India corridors. Eligible participants included male and female migrants aged 15–49 years who had resided or worked in India for a minimum of three months and a maximum of ten years. At destination sites, the survey targeted Nepali and Bangladeshi migrants engaged in informal employment sectors such as construction, manufacturing, domestic work, transportation, and small-scale trading. At source locations, the sample included circular or returnee migrants, spouses of migrants, and adults from households with at least one migrant member. Women accompanying migrant husbands were also included, even if not economically active, in order to capture household-level effects of migration and information diffusion.

In treatment areas, survey teams used existing lists of migrant workers who had received services to implement a two-stage cluster sampling approach. In control areas, respondents were selected via random walk sampling. For details on the endline survey sampling frame, see [52]. The sampling frame was designed to enable both longitudinal comparisons—between baseline and endline—and cross-sectional comparisons—between treatment and control groups at endline. Ultimately, 3528 interviews were conducted at endline, with 1734 respondents belonging to the treatment group (see

Table 2. Sample size of baseline and endline surveys.

	Baseline (2011)	Endline-Control (2014)	Endline-Treatment (2014)
Source country
Nepal	550	466	448
Bangladesh	550	444	443
Destination country
Nepali migrant population in India	573	473	417
Bangladeshi migrant population in India	472	411	426
Total	2145	1794	1734

).

Both surveys collected comprehensive data on socio-demographic characteristics, worker rights and entitlements, income and remittance flows, access to services, sexual behaviour, family planning, and HIV/AIDS knowledge, attitudes, and practices among migrant workers in India and their families in Bangladesh and Nepal. The surveys were conducted in Hindi, Nepali, and Bengali, respectively.

4.2. Migrant Profiles and Context of Mobility

Migration from Bangladesh and Nepal to India has historically served as a crucial coping and survival strategy for many individuals. The primary motivation for migration across our sample is the pursuit of better livelihood opportunities. Respondents consistently highlighted that India offers significantly greater economic prospects compared to their home countries. This economic drive stems from limited employment and income-generating opportunities at source, prompting migrants to seek better prospects abroad.

Once in India, the migrants engage in a variety of occupations based on their gender, skills, and opportunities available. Nepali-speaking migrants predominantly work as restaurant workers (25%), house servants (23%), watchmen (13%), and factory workers (12%). Conversely, Bangladeshi migrants are mostly employed as casual labourers (22%), house servants (14%), and petty traders (7%). An interesting demographic detail is that 29% of Bangladeshi migrants are housewives or unemployed, compared to only 7% among Nepali migrants. However, the data do not sufficiently distinguish between housewives and unemployed individuals, making it difficult to determine who faces higher unemployment levels in India.

While the quantitative data underscore the economic motivations for migration, qualitative insights from focus group discussions reveal additional complexities, especially concerning women migrants. Although 99% of respondents acknowledged that India offers greater livelihood potential, women often migrate to follow their husbands or based on promises from relatives of better marriage prospects. However, the reality upon arrival frequently diverges from these expectations. Bangladeshi female migrants in particular report being forced by relatives into sex work or abandoned by their husbands, forcing them to fend for themselves. Some women also fall victim to trafficking, kidnapping, and sale into brothels or exploitation by pimps, as detailed in [53].

Education levels among migrants vary, with notable differences between Bangladeshi and Nepali populations. Among the Bangladeshi-speaking migrants in India, 24% have no formal schooling at the time of the survey, 32% have primary education, and 29% have completed middle school. For Nepali-speaking migrants, the figures are slightly better, with 19% having no schooling, 39% with primary education, and 25% having completed middle school. The remaining individuals in both groups have attained secondary or senior secondary education. Regarding gender distribution at the time the survey, 37% of Nepali migrants and 47% of Bangladeshi migrants in India are women, indicating a significant female presence in migrant populations from both countries.

Overall, economic necessity seems to drive migration from Bangladesh and Nepal to India, with migrants engaging in a range of occupations that probably reflect their skills and circumstances. However, the migration process, particularly for women, is often fraught with risks and unmet expectations, underscoring the complex realities behind the economic motivations. Understanding these dynamics is essential for policymakers aiming to address migrant vulnerabilities and improve their integration and protection in host countries.

4.3. Estimation

We resort to Propensity Score Matching (PSM) estimators to measure the effect of EMPHASIS on various outcomes related to knowledge, attitudes, and behaviour of HIV/AIDS. The idea behind PSM is to identify a pool of potential comparison observations that closely resemble the treated units [54]. Although the set of conditioning factors necessary to identify valid comparison units is typically required to be of high dimension, Rosenbaum et al. [55] shows that instead of conditioning the matching on the whole set of individual characteristics, it suffices to concentrate on a single-index variable, i.e., the propensity score (PS).

Using this approach, we matched migrant workers in our sample who were exposed to the EMPHASIS programme to a subset of workers from the control locations on the basis of their observable characteristics. For a detailed discussion on the properties of the PSM estimators, see [56,57]. To achieve close balancing, we used a Kernel matching with the optimal bandwidth of 0.06. In addition, we imposed common support by dropping treatment observations whose propensity score is higher than the maximum or less than the minimum propensity score of the controls.

Assessing Matching Quality

Standard balancing tests were performed to check the quality of the matching, i.e., t-tests and Standardised Differences between the treated and untreated groups. We considered the the sample to be balanced on observables when the standardised difference was less than 20% as suggested in [58].

Table A1. Characteristics across matched and unmatched samples, Group 1.

	Unmatched Sample				Matched Sample
Variable	Treated	Control	$p>t$		Treated	Control	$p>t$
Gender (male = 1)	0.80	0.47	0.00		0.77	0.75	0.50
Age	32.00	31.40	0.95		32.10	32.69	0.39
Married	0.97	0.95	0.00		0.97	0.98	0.47
Lives with family at destination	0.51	0.79	0.00		0.54	0.66	0.01
Level of education	2.29	2.32	1.05		2.33	2.42	0.54
Disposable income	6494.60	4881.40	0.55 *		6568.50	6628.10	0.88
Frequency visits home	1.87	1.11	0.90		1.71	1.90	0.02
Sample	Pseudo $R^2$	Mean bias	Median bias
Unmatched	0.181	38.1	33.9
Matched	0.023	10.1	5.5

Note: Pseudo $R^2$ of probit model for the selection of treated units. Group 1 refers to the sample at destination country (Nepali speaking in India). Significance level * $p<0.05$. Source: authors.

,

Table A2. Characteristics across matched and unmatched samples, Group 2.

	Unmatched Sample				Matched Sample
Variable	Treated	Control	$p>t$		Treated	Control	$p>t$
Gender (male = 1)	0.59	0.65	0.00		0.60	0.63	0.39
Age	30.87	32.91	1.08		31.04	31.44	0.49
Married	0.98	0.96	0.00		0.98	0.98	0.81
Lives with family at destination	0.95	0.94	0.00		0.95	0.95	0.80
Level of education	2.74	2.88	1.11		2.76	2.76	0.99
Disposable income	6909.10	7571.00	1.06		6984.60	7337.10	0.47
Frequency visits home	1.10	0.96	1.37 *		1.06	1.06	0.98
Sample	Pseudo $R^2$	Mean bias	Median bias
Unmatched	0.02	12.4	11.3
Matched	0.001	3.2	2

Note: Pseudo $R^2$ of probit model for the selection of treated units. Group 2 refers to the sample at destination country (Bangladeshi speaking in India). Significance level * $p<0.05$. Source: authors.

,

Table A3. Characteristics across matched and unmatched samples, Group 3.

	Unmatched Sample				Matched Sample
Variable	Treated	Control	$p>t$		Treated	Control	$p>t$
Gender (male = 1)	0.42	0.42	0.00		0.41	0.43	0.74
Age	28.50	31.09	0.67 *		28.41	28.49	0.90
Married	0.95	0.98	0.00		0.95	0.95	0.99
Level of education	2.26	2.27	0.78 *		2.27	2.18	0.55
Sample	Pseudo $R^2$	Mean bias	Median bias
Unmatched	0.024	12.8	8.7
Matched	0.001	2.3	2

Note: Pseudo $R^2$ of probit model for the selection of treated units. Group 3 refers to the sample at source country (Nepal). Significance level * $p<0.05$. Source: authors.

and

Table A4. Characteristics across matched and unmatched samples, Group 4.

	Unmatched Sample				Matched Sample
Variable	Treated	Control	$p>t$		Treated	Control	$p>t$
Gender (male = 1)	0.60	0.81	0.00		0.63	0.62	0.90
Age	32.85	31.62	1.83 *		32.39	31.78	0.74
Married	0.98	0.93	0.00		0.98	0.98	1.00
Level of education	2.77	2.60	1.06		2.65	2.69	0.90
Sample	Pseudo $R^2$	Mean bias	Median bias
Unmatched	0.052	23.9	19.5
Matched	0.001	3	2.6

Note: Pseudo $R^2$ of probit model for the selection of treated units. Group 4 refers to the sample at source country (Bangladesh). Significance level * $p<0.05$. Source: authors.

in the Appendix report the results of the balancing tests. We observed a substantial improvement in the quality of the selected control after matching, as shown by both the reduction in the mean absolute standardised bias and in the pseudo $R^2$ of the probit model for the selection of treated units. Table A1, Table A2, Table A3 and Table A4 also show the p values of the mean differences for each of the observed characteristics we are controlling for. We note, however, that t-tests and other statistical tests of hypothesis are influenced by the sample size; thus, we expected few significant differences between the treatment and control groups to remain after the matching.

5. Results

Prejudices, misconceptions, and limited knowledge about HIV transmission mechanisms remain the main obstacles to preventing the spread of the disease [59]. EMPHASIS implemented a series of interventions aimed at improving HIV/AIDS knowledge among migrant workers and their spouses. These policies included peer education and awareness raising activities conducted through door-to-door outreach, thematic group sessions, and permanent and mobile drop-in centres at both source and destination locations. These activities included modules to influence sexual behaviour and the promotion and demonstration of the correct use of condoms.

In this study, we examine the first-order effects of EMPHASIS awareness raising activities on factual knowledge of HIV transmissions, misconceptions about the disease, and whether migrant workers are able to discuss HIV-related issues with their spouses. However, raising awareness about HIV/AIDS does not per se translate into changes in sexual behaviour.

Limited access to, or unaffordable market prices of, condoms [60,61], unequal power relationships—and violence—between wives and husbands [62,63,64], and excessive use of alcohol and drugs [65,66] are among the factors that can undermine the effectiveness of HIV awareness raising campaigns. Thus, in this study we also examine the second-order effects of EMPHASIS on actual condom use among migrant workers and their spouses. For a meta-analysis on the correlates of condom use, see [67].

5.1. HIV/AIDS Knowledge

In this subsection, we present the Average Treatment Effects on the Treated (ATT) estimates for HIV/AIDS knowledge. We also computed the Average Treatment Effects (ATE) estimates and found qualitatively similar results. In the following sections, we only report ATT estimates, although the ATE results are available on request from the authors. The first two columns of

Table 3. ATT estimates of the impact of EMPHASIS on HIV/AIDS knowledge by the country of origin of migrants.

	Endline: C-T		Treated: BL-EL
	Source	Destination	Source	Destination
Panel A: Nepal
Transmission of HIV	0.129 ***	0.244 ***	0.269 ***	0.258 ***
	(0.08–0.18)	(0.18–0.30)	(0.22–0.32)	(0.16–0.36)
Misconceptions about HIV	0.084 ***	0.240 ***	0.177 ***	0.138 ***
	(0.04–0.13)	(0.18–0.29)	(0.13–0.23)	(0.06–0.21)
Discussion of HIV	0.573 ***	0.345 ***	0.300 ***	0.092
	(0.31–0.83)	(0.23–0.47)	(0.16–044)	(−0.02–0.20)
Panel B: Bangladesh
Transmission of HIV	0.535 ***	0.384 ***	0.361 ***	0.323***
	(0.48–0.59)	(0.32–045)	(0.29–0.43)	(0.25–0.39)
Misconceptions about HIV	0.512 ***	0.367 ***	0.418 ***	0.602 ***
	(0.45–0.57)	(0.30–0.43)	(0.33–0.50)	(0.52–0.68)
Discussion of HIV	0.389 ***	0.378 ***	0.265 ***	0.294 ***
	(0.30–0.48)	(0.31–0.45)	(0.17–0.36)	(0.17–0.42)

Note: Statistical significance at conventional levels: *** $p<0.01$, ** $p<0.05$, * $p<0.1$. 95% confidence intervals are reported in parentheses.

show the differential effect between treated and control groups cross-sectionally, i.e., at endline, while the third and fourth columns show the differences in the outcome measures between the baseline and the endline. The two sets of results are very similar, so we focus on the results between baseline and endline.

Table 3 confirms the earlier literature showing that awareness campaigns can have large and significant effects on HIV/AIDS knowledge, misconceptions, and sexual behaviours [68,69]. The results show that, in general, treated migrants and their spouses at both the source and destination locations (i) were more likely to correctly identify two or more modes of HIV transmission, (ii) observed a higher probability of identifying two misconceptions about HIV, and (iii) were more likely to engage in discussions about HIV/AIDS with their partners than the control group.

Looking at the results by country of origin of the migrants, we find that while Nepali and Bangladeshi migrants exhibited very similar size effects in identifying modes of HIV transmission, Bangladeshi migrants appeared more apt to identify misconceptions about HIV than Nepali migrants. At first glance, it was unclear to us what factors were driving the results. Variation in education would usually be suspected as a key underlying factor, but while a small but statistically significant difference in education levels was observed between treatment and control groups within the Bangladeshi population, our matching estimator was able to remove any effect from this observed heterogeneity.

We also consider religion as another potential underlying factor. There is a growing literature that emphasises the role of religion in shaping views and understanding of HIV transmissions that enforce misconceptions [70,71,72]. The influence of religion, especially in traditional cultural settings, has been identified as a determinant of HIV spread [73]. We tried to examine this issue empirically, but unfortunately we observed very limited variation in our sample in terms of the faith the migrant workers professed (62 per cent of Bangladeshi migrants were Hindu and the remaining 38 per cent were Muslim, whereas 99 per cent of Nepali migrants were Hindu).

We then examined the role of gender; see

Table 4. ATT estimates of the impact of EMPHASIS on HIV/AIDS knowledge by the gender of migrants.

	Endline: C-T		Treated: BL-EL
	Source	Destination	Source	Destination
Panel A: Female
Transmission of HIV	0.274 ***	0.595 ***	0.320 ***	0.299 ***
	(0.23–0.32)	(0.52–0.67)	(0.26–0.38)	(0.20–0.40)
Misconceptions about HIV	0.271 ***	0.564 ***	0.276 ***	0.404 ***
	(0.23–0.32)	(0.49–0.64)	(0.22–0.33)	(0.31–0.49)
Discussion of HIV	0.382 ***	0.499 ***	0.331 ***	0.445 ***
	(0.26–0.50)	(0.41–0.59)	(0.20–0.46)	(0.28–0.60)
Panel B: Male
Transmission of HIV	0.438 ***	0.223 ***	0.251 ***	0.220 ***
	(0.35–0.51)	(0.19–0.28)	(0.16–0.34)	(0.07–0.37)
Misconceptions about HIV	0.343 ***	0.215 ***	0.281 ***	0.119
	(0.27–0.39)	(0.17–0.27)	(0.20–0.36)	(0.03–0.20)
Discussion of HIV	0.484 ***	0.238 ***	0.202 ***	0.014
	(0.39–0.57)	(0.22–0.34)	(0.10–0.30)	(−0.22–0.24)

Note: Statistical significance at conventional levels: *** $p<0.01$, ** $p<0.05$, * $p<0.1$. 95% confidence intervals are reported in parentheses.

. Previous scholarly work has found that differences in gender roles, particularly in traditional societies, reinforce HIV misconceptions and increase the risks of HIV contagion, especially—and disproportionately–in relation to women [74]. Inequitable gender norms can undermine women’s position and their voice when discussing HIV prevention and sexual practices [75]. Thus, targeted interventions such as EMPHASIS could in principle have differentiated effects between women and men that ultimately improve gender imbalances.

Looking at the results by sex of the migrant, we found that, indeed, at destination locations—the areas where the risks of HIV infection among migrant workers are most significant—the ATT estimates for misconceptions about HIV and discussions about HIV were large and statistically significant at the 1 per cent level for female migrant workers, but small and insignificant for male migrant workers (see the last column of Table 4). Our results indicate that EMPHASIS, through outreach and peer sex education and referral services, was particularly effective at empowering women with information and tools to discuss HIV issues with their spouse. Given the initial disadvantaged position of women, the positive and large effect observed after treatment appears to have influenced the overall results.

5.2. Condom Use

As noted earlier, raising awareness of HIV/AIDS might not directly translate into changes in sexual behaviour. Limited access to or unaffordable market prices of condoms [60,61], unequal power relationships and violence between wives and husbands [62,63,64], and excessive use of alcohol and drugs [65,66] are among the factors that can undermine the effectiveness of HIV awareness campaigns. Thus, it is imperative to examine the second-order effects of EMPHASIS on actual condom use among migrant workers and their spouses.

An initial analysis of the data shows that between 80 and 90 per cent of migrant workers at both the source and destination locations were sexually active; however, a very low percentage reported to have had sex in the previous 12 months with non-regular partners, including sex workers (see

Table A5. Sexual behaviour and practices. Figures in percentages, except samples.

Population Subgroups	Baseline	Endline Control Group	Endline Treatment Group
Destination locations in India (Nepali migrants)
Sexual Intercourse	79.7	80.8	86.8
Sex with a non-regular partner (last 12 months)	11.2	1.5	5
Sample	455	323	383
Destination locations in India (Bangladeshi migrants)
Sexual Intercourse	85.7	78.5	84.1
Sex with a non-regular partner (last 12 months)	0.7	3	0.3
Sample	403	328	339
Source locations in Nepal
Sexual Intercourse	99.8	92.4	92.9
Sex with a non-regular partner (last 12 months)	0.7	2.2	3
Sample	450	414	433
Source locations in Bangladesh
Sexual Intercourse	93.2	88.9	91.3
Sex with a non-regular partner (last 12 months)	0.7	2.8	6
Sample	287	393	397

in Appendix A). Given the sensitive nature of this question, we cannot reject the possibility of a response bias. While we cannot draw any conclusions based on this small sample, we can still calculate ATT estimates for condom use with regular partners or spouses. We present the results in

Table 5. ATT estimates of the impact of EMPHASIS on condom use.

	Nepali		Bangladeshi
	Source	Destination	Source	Destination
Last intercourse	0.340 ***	0.110 **	0.326 ***	0.241 ***
	(0.18–0.51)	(0.11–0.21)	(0.25–0.40)	(0.16–0.32)
Never use	−0.330 ***	−0.029	−0.265 ***	−0.341 ***
	(−0.61–−0.06)	(−0.16–0.10)	(−0.36–−0.16)	(−0.42–−0.26)
Some times	0.113	0.086	0.038	0.305 ***
	(−0.11–0.34)	(−0.1–0.19)	(−0.04–0.11)	(0.22–0.38)
Always use	0.235 ***	0.016	0.227 ***	0.045 ***
	(0.12–0.35)	(−0.04–0.07)	(0.16–0.29)	(0.01–0.08)

Note: Statistical significance at conventional levels: *** $p<0.01$, ** $p<0.05$, * $p<0.1$. 95% confidence intervals are reported in parentheses.

.

In general, we found a large effect of EMPHASIS on condom use among Bangladeshi and Nepali migrants during the last intercourse with regular partners, in both the source and destination locations. The ATT estimates are on the order of 11 and 34 per cent, and significant at 1 per cent level. The increase in condom use is in itself an important outcome of the intervention, given the prevalence of unprotected customary sex practices in these socially conservative environments. However, an increase in condom use does not tell us how regular this practice actually is. In the last three rows of Table 5, we present the results of an ordinal scaled measurement. In general, we found a considerable reduction in unsafe sexual practices. Among Bangladeshi migrant workers in both source and destination locations, we observed, respectively, a 27 and 34 per cent reduction in the probability of never using a condom when having sexual intercourse with their regular partners or spouses. Among Nepali migrant workers, we also found a similar size reduction in unsafe sex practices, but only at the source locations.

We also ran the ATT estimates by sex of the migrant and found a sizeable reduction in the incidence of unprotected sex practices after programme treatment; see

Table 6. Estimates of the impact of EMPHASIS on condom use.

	Female		Male
	Source	Destination	Source	Destination
Last intercourse	0.314 ***	0.100	0.324 ***	0.176 ***
	(0.22–0.41)	(−0.01–0.21)	(0.22–0.42)	(0.09–0.26)
Never use	−0.315 ***	−0.167 ***	−0.257 ***	−0.178 ***
	(−0.43–−0.20)	(−0.29–−0.04)	(−0.38–−0.13)	(−0.27–−0.08)
Some times	0.113 **	0.160 ***	0.037	0.205 ***
	(−0.01–0.23)	(0.04–0.27)	(−0.09–0.16)	(0.12–0.28)
Always use	0.202 ***	0.035	0.252 ***	0.012
	(0.13–0.27)	(−0.04–0.11)	(0.17–0.33)	(−0.03–0.05)

Note: Statistical significance at conventional levels: *** $p<0.01$, ** $p<0.05$, * $p<0.1$. 95% confidence intervals are reported in parentheses.

. Male as well as female migrants workers at both the source and destination locations reported a reduction in the probability of never using a condom when having sexual intercourse with their regular partners or spouses.

Although condom use was a direct preventive response against HIV/AIDS after the EMPHASIS intervention, nearly 90 per cent of migrant workers also reported using condoms as one of their primary family planning methods. Family planning was not part of the core services provided by EMPHASIS, although family planning methods were discussed during peer education sessions at the source and destination locations. Thus, it seems that the appealing dual functioning of condoms (i.e., HIV prevention and family planning) could have strengthened the impact of the programme, a feature that could be emphasised in similar future interventions.

6. Discussion

The evidence suggests that community-based approaches—particularly peer education and participatory group meetings—are effective in diffusing health information and promoting behavioural change among mobile populations. Social learning theory posits that individuals update their behaviours by observing and imitating credible peers within their social environments [76]. In the context of the EMPHASIS programme, peer educators and community support groups seem to have facilitated repeated social interactions that reinforced learning and legitimised preventive practices such as condom use and HIV testing. Migrants often face social isolation, stigma, and structural barriers to health access, which constrain the flow of information. Peer-led diffusion helped to overcome these barriers by embedding preventive messages from the information campaign within existing social ties—an approach consistent with evidence from highway-corridor trucker programmes and network-based testing strategies in mobile communities [77,78,79].

Community structures may also have activated collective behavioural dynamics similar to those observed in coordination and public-goods models of health cooperation [80,81]. Once a critical mass of participants adopted safer practices, reputational incentives and evolving group norms reinforced compliance. Empirical research in HIV prevention settings supports the idea that community sanctioning and social commitment sustain cooperative behaviour even after external support ends [82,83].

The stronger gains observed among female participants—in both knowledge and adoption of safer sexual practices—can be interpreted through intra-household bargaining models [84,85]. These frameworks emphasise individuals’ relative bargaining power within the household, which is shaped by access to income, assets, and external opportunities. The EMPHASIS interventions, especially those promoting remittance management, savings, and financial literacy, seem to have strengthened women’s fallback positions and decision-making capacity. Thus, EMPHASIS enhanced women’s agency to negotiate condom use and participate in testing decisions. Moreover, behavioural research indicates that women often display higher risk aversion and stronger responsiveness to credible health information [86,87], which may explain the more pronounced adjustments in behaviour observed among female participants.

7. Sensitivity Analysis and Study Limitations

The results may be influenced by potential selection on unobservables or unmeasured differences between treated and control units that are not fully accounted for by the matching procedure. In order to assess the robustness of the estimated treatment effects to such hidden bias, we conducted a sensitivity analysis using the Rosenbaum bounds approach [88]. This method evaluates how strongly an unobserved factor would have to influence treatment assignment to invalidate the causal inference by calculating the critical level of hidden bias, denoted by $\Gamma$, at which the treatment effect would no longer be statistically significant. Although this approach does not test the unconfoundedness assumption itself, it helps us to determine how strongly an unmeasured variable would need to influence the selection process in order to alter the results of the matching analysis [89].

The analysis based on the Mantel–Haenszel test under the null hypothesis of no treatment effect for different values of unobserved selection bias, $\Gamma$, shows that the matching results remain significant at the 1% and 5% level (10% level in a couple of cases) even if there are unobservable characteristics that increase the likelihood of treatment by a factor of 1.8 or more. These results are available from the authors upon request.

A few results, however, remain sensitive to hidden bias at: $\Gamma =1.40$ for misconceptions about HIV among the nepali population at source, and $\Gamma =1.30$ for female condom use (sometimes). Thus, one must interpret these results with caution. It should be noted, though, as highlighted by Becker [89], that these values represent worst-case scenarios. A critical value of 1.3, for instance, does not necessarily indicate the presence of unobserved heterogeneity or that the treatment has no causal effect on the outcome variable. Instead, it suggests that the confidence interval for the estimated treatment effect would include zero if an unobserved variable caused the odds of treatment assignment to differ between treated and control units by a factor of 1.3.

Thus, the robustness of most estimates to substantial bias indicates that the treatment effects are unlikely to be driven solely by unmeasured factors such as motivation, risk preferences, or prior awareness. Even if unobservable characteristics doubled the odds of programme participation, the estimated impacts would still remain statistically significant. Consequently, while residual selection cannot be fully ruled out completely, the overall robustness of the estimates provides reasonable confidence that the main results reflect programme effects rather than artefacts of unobserved heterogeneity.

While the robustness checks give us confidence about the findings, some limitations remain. First, the absence of individual-level panel data has limited our ability to rule out composition changes between baseline and endline surveys. Furthermore, potential spillovers between nearby sites and contamination via media or migrant networks could have attenuated estimated effects. Finally, the results pertain to specific migration corridors, which may limit external validity to other mobile population settings.

8. Conclusions

Migration has long been identified as a driver of HIV transmission by connecting geographically distinct epidemics and shaping environments that facilitate risky sexual behaviour and substance use [21]. While early research—particularly the well-known case studies of Latino migrants in the United States—provided important insights into mobility-related vulnerabilities [90,91,92,93], recent evidence has expanded the geographical and thematic scope of inquiry. Studies have examined mobile populations in sub-Saharan Africa, South and Southeast Asia, and Europe, documenting elevated HIV prevalence, gaps in service access, and the influence of migration on testing and treatment outcomes [94,95,96]. However, despite the growing literature, most of the evidence remains qualitative, or based on cross-sectional and programmatic data. Rigorous quantitative evaluations of HIV prevention and behavioural interventions among migrants—especially those with irregular status or engaged in circular mobility—remain scarce due to methodological and logistical constraints, underscoring the need for robust quantitative research.

This study reports the results of the EMPHASIS programme, a large intervention implemented by CARE International in India, Bangladesh, and Nepal with the aim of reducing HIV risks among migrant workers in the South Asian region.

Nepali and Bangladeshi migrants in India are particularly vulnerable to HIV infection. The often exploitative conditions surrounding their migratory status, along with significant language barriers and legal status, can delay the early diagnosis and start of HIV treatment, which in turn has significant consequences in terms of transmission in both the source and destination countries. Our analysis indicates that the EMPHASIS programme had positive effects on HIV and AIDS knowledge, attitudes, and practices among Bangladeshi and Nepali migrant workers. The treatment population at both the source and destination locations was more likely to correctly identify two or more major modes of HIV transmission; they also had a higher probability of identifying at least two misconceptions about HIV and were more likely to discuss HIV/AIDS with their partner than the control group. Furthermore, treated migrants were less likely to engage in unsafe sexual practices as they used condoms with their partners or spouses more regularly.

In general, our results show that peer-led education and information campaigns with continuity-of-care structures can effectively improve HIV knowledge and increase condom use among vulnerable migrant workers. Although previous research has shown the positive impact of information campaigns in increasing HIV/AIDS knowledge, our study contributes to the literature by quantitatively establishing a causal link between a targeted intervention and sexual behaviours among vulnerable populations.

Awareness campaigns play a crucial role in reducing the risk of HIV transmission. However, there is limited knowledge about the level of awareness of HIV and disease transmission among mobile populations. In this context, identifying the factors associated with knowledge, attitudes, and practices about HIV and AIDS of migrants is essential from a public health perspective. Our findings also highlight the importance of linking peer-led activities with HIV prevention measures, testing, and treatment services. Furthermore, disseminating information about the multiple benefits of condom use—including family planning—and promoting easily accessible condoms are particularly important strategies to reduce HIV transmission risks among mobile populations.

Future research should deepen causal identification by following individuals through mobility. Promising directions include longitudinal cohorts or phased rollouts to measure dynamic and longer-run effects. Greater attention to heterogeneity by gender, identities, legal status, occupation, and migration modality would improve knowledge and policy design.