Human rights and social justice in labour migration is more than a written entitlement to work. Also they have to do with whether workers get paid, whether they have personal freedom, whether they work in safe conditions, and whether they can have someone to help them if a conflict arises. The United Arab Emirates (UAE) 2021, 2022 present a valuable context in which to explore these intersections given the experiences of migrant workers in the employment and migration environments and the sectoral regulation of labour. This article looks at those experiences as reported in individual-level records of surveys, not as reflected in national population statistics as proxies for workplace conditions (Arifuzzaman et al., 2021).

Previous studies on labour protection in the UAE have been linked to institutional accountability and worker health. Keane and McGeehan, (2008) studied the enforcement of migrant workers' rights both domestically and internationally. Sönmez et al. (2011) talked about the link between the reported rights violations and health inequalities. These contributions have made it salient to explore protection in practice. The conclusions that they reach in the past, however, do not necessarily represent the present or the impact of later legislation. Regulatory environment has shifted. The main federal private-sector labour framework is governed by Federal Decree-Law No. 33 of 20213 which includes non-discrimination provisions, and domestic workers are covered under Federal Decree-Law No. 9 of 2022, as amended 2019. The country brief provided by the International Labour Organization (ILO) outlines key rules related to recruitment charges, documents, working conditions and grievance procedures for categories of migration. A difference between legal regimes does matter: a mere survey answer cannot provide sufficient evidence of a violation of a legal rule, a contract, an event, and jurisdiction.

Empirical work can be useful in disentangling a number of questions that are lumped together. What is the proportion of workers in a specified sample who say they have an employment concern? Are there differences among groups in reports? Is legal awareness associated with using complaint channels? Responding to one question does not answer the other questions. Frequent reports of a problem in a sample, even though there is no demographic difference. Similarly, complaints may be made without any resolution and non-filing may be due to a variety of reasons. In the present study, the dataset of the survey with 400 records from which the author's thesis was based is used. It poses questions regarding both the remuneration and work-condition and mobility issues reported in the dataset and the distribution of awareness of rights, access to legal resources and complaints filed; and asks whether selected associations and adjusted complaint-filing relationships are visible. Its contribution is a reproducible reanalysis that clarifies descriptive burden from statistical association, maintains uncertain responses and compares the reported results with the individual-level records. It also explains what social justice can and cannot be taken for granted from short-term indicators of employment.

Conceptual Framework

Social justice is used as a normative interpretative framework rather than a measured dependent variable. Fraser, (2005) links the concepts of redistribution, recognition and representation to discussions of justice between borders. In this context, redistribution relates to remuneration and the distribution of recruitment expenses, and recognition relates to treatment, dignity and autonomy while representation refers to claiming an entitlement and to being heard. These dimensions are used to interpret the Total score and do not form part of the Total score.

UN Guiding Principles on Business and Human Rights categorize the state's responsibility to protect, company's responsibility to respect and access to remedy (Office of the United Nations High Commissioner for Human Rights [OHCHR], 2011). This is a framework that links institutions and business relations to workplace practices. It has also brought an element of scrutiny to what occurs when a grievance is raised. The principles are not intended to replace the governing law but are rather intended to be used as an analytical benchmark. Other factors to consider are employment related dependence. Parreñas and Silvey, (2011) examine disciplinary mechanisms impacting migrant domestic workers in the UAE, and Silvey and Parreñas, (2020) recount the financial and employment precarity along migration pathways. These studies provide some indication of how the options of the worker can be limited. Those mechanisms can not be directly tested with the current survey, which does not include detailed migration histories or verified immigration records. The items it sponsors are thus a reflection of the reports and perceptions of the respondents.

“Access to Justice” is viewed as a multi-stage process of recognising a concern, identifying assistance, starting a claim and achieving an outcome. The dataset collects information about awareness, access and filing, not about adjudication and compensation. This separation is the empirical scope. The article is not permitted to evaluate the fairness, speed or effectiveness of the outcome, but it may include a description of a complaint process.

Design, source and analytic sample

This study is quantitative secondary analysis of the cross-sectional survey data that were collected for the author's thesis on migrant labour in UAE. Non-probability recruitment is described in the parent research documents; this involved mainly purposive sampling and snowball sampling to access workers across sectors of employment. The final respondent file has 400 records of construction, domestic, hospitality, retail and “Other” files. No new respondents were recruited for this analysis and the cases available were used as the analytic sample. The aim was to reach migrant workers in the UAE, as targeted in the parent study. The file participants are divided into 4 adult age groups starting with age 18-25 years. The sample is not representative of the UAE migrant workforce, as there is not a sample probability nor a sampling frame. Documentation of the number approached, refusals, and fieldwork dates and distribution by emirate is not available in the analytic file. Therefore, a response rate and population weighting and comparison with the reform period were not computed. These reporting limits are relevant to the interpretation, as recom-mended by the guidance for observational studies.

Measures and operational definitions

There are 25 columns in the file. There are 7 attributes describing the characteristics of the respondent: Age group, Gender, Nationality, Education, Sector of employment, Time in the UAE and Monthly salary (AED). Working-condition items relate to perceptions of safety, hours worked per week, health and safety training, lack of leave, satisfaction and denial of leave. Others focus on the timeliness of pay, labour law awareness, filing of complaints, perceived restrictions on freedom and access to legal resources due to sponsorship, salaries paid on time, and labour law awareness.

The substantive items were analyzed individually. Items that had a single answer choice (Yes/No) were kept. The responses of “Not Sure” to the questions about safety, training, and legal-resource access were left in place as separate responses. Rights awareness was entered as Not Aware, Somewhat Aware, and Fully Aware in the correlation analysis as well as entered categorically in the logistic model. This three-level measure was not replaced by a separate item on labour-law awareness (Yes/No). Sometimes or Never was considered irregular salary payment while Always was considered as regular salary payment. Accommodation dissatisfaction was combined into one descriptive summary (Dissatisfied and Very Dissatisfied). All of the 400 records have the same wording in the last column: “Changes for Legal Protection Improvement,” “Open-ended.” No substantive responses and was excluded. No interview transcriptions or focus-group material were analyzed and the present article is not reported as mixed-methods research. There is no validated composite measure of perceived social justice, discrimination or satisfaction with justice mechanisms. Thus, no scale reliability coefficient, latent-variable model or hypothesis test with regard to an invented social-justice score was conducted.

The meanings of the recorded item labels are retained for reports of wage theft, confiscation and restrictions. There is no data on the amount or on the date of the event or the date of adjudication of the case. Therefore, a positive answer does not mean that a specific employer has committed a specific statutory breach, but rather means that a specific employer has self reported a violation of a specific statute. There is no identification of the channel, number of complaints, legal issue or legal outcome in the complaint item.

Data reconciliation and quality checks

The spreadsheet with numbers on the respondent level was the source of numbers for analysis. All 400 rows were saved. There were no blank values in the substantive columns and no exact duplicate complete rows. The total number of items in categorical fields were verified and the values recorded in numeric fields were checked. Monthly salary takes four values, duration five values and weekly hours four values. These were analyzed as recorded (assuming no finer underlying precision).

A reconciliation with the results of the thesis chapter revealed two discrepancies in materials. The distribution of its grouped working-hours did not correspond to the spreadsheet: that has only 40, 50, 60 and 70 hours, and no group below 40. In addition, the separate dispute-resolution-awareness table restates the labour-law-awareness counts, without having a separate variable for awareness of dispute mechanisms. This table was not reproduced. Rather, the legal-resource-access item which was available, was reported in its correct name. There were no items that were not recorded as is due to cross item discrepancies. For instance, 119 records relate to a ‘Yes' response to perceived restrictions related to sponsorship and a ‘No' response to being tied to an employer. Not always logically impossible, items may represent different experiences or understandings. They, however, restrict the use of the sponsorship items as legal-status measures established. They were unable to resolve their interpretation due to a lack of documentation of validation of the questionnaire.

Statistical analysis

Throughout the counts and percentages used N = 400. Recorded weekly hours, duration and salary were summarized by means and sample standard deviations, also medians were explored. For selected binary summaries, nominal 95th percent confidence intervals10 were computed. These intervals measure binomial uncertainty given the sample analyzed, and do not reflect non-probability sampling and selection, clustering of sample recruits, or reporting error; they are not national margins of error. Six exploratory bivariate analyses were stated, corresponding to the research questions of wage theft by sector; perceived safety by gender; rights awareness and level of filing a complaint; legal resource access and level of complaint filing; perceived mobility restrictions and wage theft; and ordinal rights awareness and length of time in the UAE. The first five were analyzed using Pearson's chi-squares without a continuity correction with Cramér's V as an effect size measure. Count of cells were verified. Duration and ordinal awareness were analyzed using Spearman's rank correlation. Two-tailed p values are presented along with Benjamini-Hochberg, (1995) corrected p values for this six-test family. Analyses were exploratory and were not preregistered.

Complaint filing was explored using a binary logistic model. Categorical rights awareness, categorical legal-resource access, age group, sex, sector, reported wage theft and perceived mobility restrictions were predicted. Demographic variables were included to be able to compare within the model, the employment indicators were used to capture potential reasons for a complaint. The number of non-intercept parameters were 15, and the number of records in each outcome category were 200. Adjusted odds ratios (aORs), Wald 95% confidence intervals and unadjusted coefficient p values were calculated. A likelihood-ratio test and McFadden's pseudo-R² were used for overall model performance in comparison with an intercept-only model. This is a model that is explanatory and exploratory, and there was no predictive/validation or causal claim. Python and libraries pandas, NumPy and SciPy were used for analyses. The logistic coefficients were estimated by maximum likelihood with a trust-region optimizer, with the covariance estimated from the observed information matrix. The rank of the full design matrix and convergence were checked. The coefficients were reproduced within 0.00001 by an independent BFGS optimization. Stepwise variable selection and search for statistically significant models were not performed.

Data handling

The analysis file does not include any personal identifiers. A record of the sources used was kept, and there are no individual employment records or combinations of sensitive attributes reproduced in the article. All reported results are collated. This secondary analysis did not involve any new participant contacts, administrative linkage or employer identification. Ethics and provenance statement. This article is a secondary analysis of an existing, de-identified survey dataset originally collected for the author's thesis. No additional participants were recruited and no new individual-level personal identifiers were obtained for this analysis. The available manuscript does not provide an institutional ethics committee approval number, the wording of participant consent, or the parent study's data-sharing permission; accordingly, no claim of ethics approval or consent waiver is made here. The author should confirm and document the applicable permissions and original study consent arrangements before submission.

Respondent profile

All 400 records were analyzed. Respondents aged 46 years or above comprised 115 records (28.75%), followed by ages 36–45 years, 104 (26.00%). The recorded gender categories were Male, 150 (37.50%); Female, 133 (33.25%); and Other, 117 (29.25%). The Other category is kept without interpretation as a gender identity, but this does not mean that there is no other gender identity of that person. 314 records (78.50%) was from nationalities Bangladesh, India, Pakistan, and the Philippines. The biggest recorded sector was retail, with 88 (22.00%) and construction the smallest with 75 (18.75%). These distributions are not a reflection of the national employment structure, but rather on the composition of the sample. The mean recorded duration in UAE was 3.0 years (SD = 1.4). Mean monthly salary was AED 4,015 (SD = 2,305.0), with a median of AED 3,000. The values of the file are only at levels of AED 1,000, 3,000, 5,000, and 7,000 and should not be taken as a detailed representation of the distribution in earnings in the broader workforce. The complete profile is presented in Table 1.

Table 1: Characteristics of the analyzed survey sample.

Note. N = 400. Categories and discrete values are retained as recorded. Percentages are sample-specific. “Other” is not reinterpreted.

Working conditions and remuneration

Perceived working conditions were reported as safe by 126 respondents (31.50%), unsafe by 140 (35.00%), and uncertain by 134 (33.50%). Health and safety training was reported by 141 respondents (35.25%); 121 (30.25%) reported none and 138 (34.50%) were unsure. The lack of a definite "no" was not considered a negative response or unsafe workplace and did not count as a lack of training. Weekly hours averaged 54.6 (SD = 11.6). There were 116 records at 40 hours, 84 at 50, 99 at 60 and 101 at 70. Thus, 25.25% recorded more than 60 hours and 50.00% recorded 60 or 70 hours. A dataset will be unable to differentiate between contractual hours, breaks and overtime arrangements. It does not therefore provide independent evidence of non-statutory compliance but sets extended recorded schedules instead. The corrected distribution is plotted in Fig. 1.

Fig. 1: Recorded weekly working hours. (Note. N = 400. No record reports fewer than 40 hours. Values are reproduced from the individual-level spreadsheet, not the discrepant grouped thesis table).

Unpaid overtime was reported by 197 respondents (49.25%), and denial of leave by 207 (51.75%). There was some variation in satisfaction with the accommodations: 166 (41.50%) were Satisfied or Very Satisfied; 87 (21.75%) were Neutral; and 147 (36.75%) were Dissatisfied or Very Dissatisfied. No objective housing-quality classification was made for satisfaction ratings. 123 (30.75%) respondents reported that salary was Always paid on time. Sometimes was recorded for 132 (33.00%) and Never for 145 (36.25%). Combined salary-payment irregularity was therefore 277/400 (69.25%; nominal 95% CI [64.56%, 73.57%]). Wage theft was reported by 202 respondents (50.50%), recruitment-fee payment by 204 (51.00%), and passport confiscation by 216 (54.00%). Wage theft and unpaid overtime occurred together in 101 records and in at least one of the two in 298 (74.50%). The percentages, therefore, cannot be added together as if they are responses from different groups. Selected concern indicators are in Table 2 and their sample proportions are compared in Fig. 2.

Table 2: Selected self-reported labour concerns.

Note. N = 400. CI = nominal Wilson, (1927) binomial confidence interval. Intervals do not adjust for non-probability selection, clustering or reporting error. Indicators can overlap. Salary irregularity combines Sometimes and Never.

Awareness, sponsorship perceptions and access to justice

Rights awareness was Fully Aware in 121 records (30.25%), Somewhat Aware in 144 (36.00%), and Not Aware in 135 (33.75%). The separate labour-law-awareness item was Yes in 208 records (52.00%) and No in 192 (48.00%). These measures are not the same in terms of legal competence and are only a list of questions that have been recorded. 168 respondents (42.00%) indicated they were dependent on their employers under the sponsorship item. 203 (50.75%) stated that there were perceived mobility restrictions due to being sponsored, and 202 (50.50%) stated that there was an impact on their freedom. These ratios refer to people's reported perceptions. They do not prove that the same immigration provision was made to all the respondents.

Fig. 2: Selected labour concerns in the survey sample. (Note. N = 400. Percentages describe the analyzed sample and are not estimates of national prevalence. Respondents can report more than one concern).

Access to legal resources was Yes for 135 respondents (33.75%), No for 139 (34.75%), and Not Sure for 126 (31.50%). 200 people said “Yes” to complaints, while 200 said “No.” The data do not reveal if there was a dispute for non-filers or if there was help, a decision or compensation for filers. There still remains the distinction between awareness and access to and initiation of a complaint as shown in Table 3.

Table 3: Rights awareness, sponsorship perceptions and legal access.

Note. N = 400. Rights awareness and labour-law awareness are distinct items. Legal-resource access is not a measure of dispute-resolution awareness. Complaint outcomes are not observed.

Exploratory associations

The sector–wage-theft association was small, χ²(4) = 2.58, p = .630, V = .080. Additionally, there was no significant association between gender and perceived safety, χ²(4) = 4.07, p = .397, V = .071. When it comes to reported duration in the UAE, there was only a slight rank association with rights awareness, ρ = −.012, p = .814. 45.93% of Not Aware respondents, 52.78% of Somewhat Aware respondents and 51.24% of Fully Aware respondents reported that they filed complaints. There was no overall awareness association (χ²(2) = 1.42, p = .493). Proportions were 48.20% for No legal-resource access, 45.24% for Not Sure and 56.30% for Yes; the test was not significant χ²(2) = 3.46, p = .177. The restriction–wage-theft test was not significant, p = .057 (p = .342 after adjustment). All six analyses are reported in Table 4. The lowest number of cells in the five chi-square tables was 36.855, which is greater than the conventional value of 5.

Table 4: Exploratory bivariate associations.

Note. N = 400. Five Pearson chi-square tests and one Spearman correlation. Adjusted p values use Benjamini-Hochberg, (1995) correction across the six analyses. Cramér V is reported for chi-square tests. All tests are two-sided.

Adjusted complaint-filing model

All 400 records were used in the logistic model, which converged. It did not show significant improvement over an intercept-only model, likelihood-ratio χ²(15) = 14.06, p = .521; and McFadden's pseudo-R² was .025. The value is a likelihood based value and is not a percentage of variance of the outcome explained. Relative to Not Aware, the aOR was 1.38 (95% CI [0.85, 2.24], p = .195) for Somewhat Aware and 1.35 (95% CI [0.81, 2.25], p = .253) for Fully Aware. Relative to No legal-resource access, the aOR was 1.42 (95% CI [0.87, 2.31], p = .163) for Yes and 0.91 (95% CI [0.55, 1.49], p = .697) for Not Sure. The reported wage theft and perceived mobility restrictions were also not significant. All of the coefficient confidence intervals contained the number 1. Table 5 presents all the adjusted comparisons, there is no selective reporting of favourable comparisons.

Table 5: Exploratory logistic model of reported complaint filing.

Note. N = 400; outcome = Complaint Filed (Yes = 1); 200 Yes and 200 No. All listed predictors enter one model. aOR = adjusted odds ratio; CI = Wald confidence interval. Coefficient p values are unadjusted. Overall LR χ²(15) = 14.06, p = .521; McFadden pseudo-R² = .025.

Frequent reports do not establish demographic or causal explanations

The most striking empirical result found in this sample is the incidence of reported employment issues, particularly issues related to irregular payment of wages and confiscation of passports. The analysis of associations does not lead to a corresponding explanation of a population or a measurable relationship between the chosen indicators of consciousness and complaint filing. Descriptive frequency is different from statistical association. Keeping that distinction does not allow for the normative significance of labour protection to be a vacuous statistical assertion (Alam et al., 2021).

The non-significant results also failed to provide an equal treatment between the sectors or between the gender. Short scales may fail to detect meaningful differences in severity or exposure, and confidence intervals are consistent with a variety of effects. For instance, a Yes/No report of wage theft does not differentiate between a small amount of salary that may have been deducted and larger amounts of salary that might have been unpaid. The study focuses on the reported concerns for further investigation, not on the calculated legally verified losses. A review of the history of labour law enforcement and health in the UAE aids in understanding these reports. No such studies are directly replicated in the current results nor are any measure of change from the time of the publications. Instruments, recruitment and time periods vary. In order to be able to evaluate reform, there needs to be a documented time, eligibility, and similar outcomes.

Remuneration and recruitment: linking monitoring to recovery

The irregularity in salary payments was the most common concern studied (69.25%). The number is for the samples, and includes the “Sometimes” and “Never” categories. It does not specify the time lapse or use of the Wage Protection System (WPS) by any respondent. The distinction between the two is important because the visibility of payments and full recovery of earnings are two different outcomes. The regional analysis of the ILO highlights the coverage of the wage system, data quality, detection and linkages with enforcement and compensation. The survey can help explore these dimensions and cannot identify which part of WPS accounts for a reported delay. Future research could also safely connect worker reports to payroll reports and track the development from detection to actual payment, given consent and confidentiality protections.

Fifty-one percent of recruitment-fee payment records were reported. ILO's principles on fair recruitment indicate that recruitment fees and related costs should not be paid by workers. Information about the amounts, the recipient and debt information is not included in the present item. Thus, it is not an accurate measure of indebtedness. One transnational approach that Silvey and Parreñas, (2020) advocate is viewing the relationship between recruitment and family financial needs in a cross-border manner. That mechanism is worthy of an instrument on its own rather than a single item of the feepayment.

Occupational protection and autonomy

Recorded 70-hour schedules and reports of overtime without pay give clues to occupational evaluation that may need to be done more closely. The number of hours per week is not a determining factor, nor is safety perception. Any compensation, contract details, exemptions and rest periods should also be analyzed. It is particularly important when there is a difference between the UAE and the domestic sector (private sector) for employment (UAE, 2021, 2022). The example of the UAE occupational research is an illustration of the importance of measuring conditions directly. Ahmed et al. (2020) evaluate thermal stress by several heat-stress indices. In a small study of construction site hydration, Bates and Schneider, (2008) found that hydration was maintained in situations where water was available and self-paced. All these studies confirm the need to study exposure and usable protection, not to take it for granted that all workplaces are alike. The current data set does not include information on heat or injury outcomes or validated safety audits and is unable to address those occupational questions.

Other measures of autonomy are passport confiscation and perceived mobility restrictions. ILO UAE regulatory brief on protections for workers' documents. An important yardstick is a legal prohibition, survey answers should be checked in context, but not court findings. Additionally, research on domestic work shows the linkages between the regulation of employment and vulnerability 19. The current null gender comparison cannot prove that risks for women described in that literature have been eliminated: not all of these risks are measured directly in the survey data.

Awareness, access and remedy are separate outcomes

One-third of records indicate access to legal resources, half report filing of complaints. This is not necessarily contradictory, as access and filing may be for different periods or resources and filing does not mean ongoing assistance. But the pattern illustrates why one complaint measure is not enough to gauge effective justice. Lowe et al. (2023) highlight the practical challenges to social protection in the GCC and Vlieger, (2014) associates the occupational status of domestic workers with access to justice and legal liminality. While these sources suggest plausible issues for follow-up, they do not indicate that people were not filed for because of fear, cost, or language. There was not a measurement taken for those reasons. The non-significant awareness and access results should thus be treated in a different manner from the significant results, and attributed to better measurement instead of a speculating of motives. The existing institutional processes need to be recognized as well. According to the Ministry of Human Resources and Emiratisation (n.d.), the ministry takes a decision on some of the issues regarding some labour disputes after a number of changes in the law. The survey does not specify if complaints were referred to that procedure. Research should document the complaint channel, legal basis, date, time to resolution and amount of money received (if applicable), so that analysis of institutional performance will not be based on a formal rule or filing rate.

Implications for social justice and subsequent research

The conceptual framework can help place the measured concerns in relation to remuneration and recruitment (redistribution); autonomy and working conditions (dignity and recognition); and awareness and complaint use (opportunities for being heard). The survey is not a validation of an extensive construct of social justice. It is empirical description of specific domains, having a clear distinction between what records reveal and what is extrapolated from that by a broader theory.

When considering the evaluation of a policy, ensure that workers are paid in full, documentation is kept, safety provisions are accessible and useful, and that the workers' outcomes of the remedy are recorded. The roles of the employer, recruiter and institution can be examined in the context of the protection, respect and remedy framework. From this point, intervention studies should be able to test for the likelihood of those outcomes occurring rather than assuming that they do because of awareness.

Future studies will need to indicate the exact wording of the questionnaires, recall periods, and to clarify ambiguous categories and include recruitment logs with sector and emirate information. Long term follow-up could explore the impact of a complaint on receipt of payment and the continuation of protection following an employment transition. Independent payroll or dispute records would enhance verification, where possible. A prospective qualitative component may be required to interpret access and sponsorship but must be grounded in the actual transcripts and safe recruitment not be universal narratives.

Strengths and Limitations

The study provides analysis of all available individual-level data, re-calculates thesis percentages instead of copying them, includes uncertainty categories, and reports all specified exploratory tests. It offers a complete set of adjusted results, as well as independent numeric verification for the fitted model. These features enhance transparency of the analysis, but not the quality of recruitment or measurement (Islam, 2025). National prevalence inference cannot be made with non-probability sampling. The dates and locations of field work, denominators and response rates for sampling of the parent survey are not available, which restricts the ability to reproduce the parent survey and to make comparisons with legal reforms. Demographic interpretation is needed for the broad Other categories, particularly gender. Numerical fields may take only discrete values and finer levels of measurement should not be assumed.

Self-reports are not independently verified, have not been validated scale properties and consistently documented recall periods. The discrepancies on the cross-item to the sponsorship question should not be interpreted as administrative legal status. There are some factors missing such as injury history, outcome of the complaint, compensation, and actual reasons why complaints were not filed. This is a cross-sectional file and cannot be used to draw a causal pathway or identify institutional reform effects. Neither nominal confidence intervals nor model-based p values correct for selection or measurement bias. Lastly, the questionnaire does not provide a broad measure of humanrights or social justice but focuses instead on individual employment issues. The article therefore presents the findings of indicators. There is no such thing as a null association establishing equality and null associations do not constitute absence of harm or failure of a particular policy and sample reports should not be used as evidence of legally adjudicated employer misconduct.

There are regular reports of salary concerns, paperwork, leave and mobility concerns, and some reports of access to legal resources in the 400-record data set. Bivariate exploratory analyses and an adjusted model for complaints filed did not reveal statistically significant relationships. The findings indicate that there is a need to investigate practical outcomes of protection and remedy more closely, but do not provide evidence of national prevalence, causal mechanisms and/or the effectiveness of reform. Empirical studies of migrant workers' human rights and social justice in the UAE would benefit from more robust instruments, a more robust and documented recruitment process, and verified outcome measures.

The author acknowledges the migrant workers whose survey responses formed the basis of the secondary analysis presented in this article.

The author declares no conflict of interest.