Presentation on theme: "An Investigation of the Job Preferences of Mid-Level Healthcare Providers in Sub- Saharan Africa: Results from Large Sample Discrete Choice Experiments."— Presentation transcript:
An Investigation of the Job Preferences of Mid-Level Healthcare Providers in Sub- Saharan Africa: Results from Large Sample Discrete Choice Experiments in Malawi, Mozambique and Tanzania Dr Eilish McAuliffe, Centre for Global Health, Trinity College, University of Dublin & HSSE team Supported by: Irish Aid & Ministry of Foreign Affairs, Denmark UNIVERSITY EDUARDO MONDLANE Faculty of Medicine
Partners to the Project Centre for Global Health, University of Dublin, Trinity College, Dublin (Eilish McAuliffe, Susan Bradley) Averting Maternal Death and Disability Program (AMDD), Heilbrunn Department of Population and Family Health, Mailman School of Public Health, Columbia University, USA (Lynn Freedman (Helen de Pinho, Samantha Lobis, Rachel Waxman and Sang Hee Won) Realizing Rights: the Ethical Globalization Initiative, USA ( Mary Robinson, Peggy Clark, Ibadat Dhillon, Naoko Otani) Regional Prevention of Maternal Mortality network, Accra, Ghana (Angela Sawyer, Dora Shehu) Ifakara Health Institute, Mikocheni, Dar Es Salaam, Tanzania (Godfrey Mbaruku, Honorati Masanja, Tumaini Mikindo, Neema Wilson, Debby Wason, Abdallah Mkopi, Aloisia Shemdoe) University of Malawi, College of Medicine, Centre for Reproductive Health, Malawi (Francis Kamwendo, Mwizapanyuma Simkonda, Wanangwa Chimwaza, Andrew Ngwira, Effie Chipeta, Linda Kalilani) Department of Community Health, Faculty of Medicine, Eduardo Mondlane University, Mozambique (Mohsin Sidat, Maria de Fatima Cuembelo, Sozinho Daniel Ndima)
Project objectives Expand the evidence base in support of effective use of mid-level health workers within an enabling environment through the generation of new evidence and a critical analysis of existing evidence; Increase recognition and effective use of mid-level health workers among national, regional, and global policymakers to address the human resources crisis in district health systems based on project evidence; Advocate for an enabling environment that optimises performance of mid- level providers in order to strengthen health systems; and In partnership with African institutions, deepen local capacity to research and analyse human resource and health systems problems, develop innovative solutions, influence policymakers at local and global levels, and sustainably implement new strategies; and build the capacity of northern institutions to successfully engage in and support partnerships of this kind. Research Advocacy
Review of previous DCE work Previous studies mostly with students Prior experience influences choice – important to focus on established health workers as their choices may be very different All except one previous study conducted with doctors and nurses – yet health systems staffed by mid-level providers Most studies - single country Previous DCE work tells us little about the factors that are important in motivating and retaining this majority component of human resources for health.
Distinctive features of this study Large sample (2,072) Across three countries (Malawi, Tanzania, Mozambique) Health workers in the health system Includes mid-level cadres Variables – human resource management and continuing professional development
Table 1:Facilities and Providers of EmOC MalawiMozambiqueTanzania Eligible providers approached 729622922 Providers consented 679607859 Provider questionnaires returned 631587854 Participation rate (among eligible providers) 87%97%93% No. of Facilities sampled 8413890
Inclusion Criteria Involvement in obstetric care defined as having completed at least one emergency obstetric care signal function in the past three months. The 9 signal functions assessed were: (i)administered parenteral antibiotics, (ii)administered uterotonic drugs (e.g. parenteral oxytocin, parenteral ergometrine), (iii)administered parenteral anticonvulsants for pre-eclampsia and eclampsia (e.g. magnesium sulphate), (iv)performed manual removal of placenta, (v)performed removal of retained products (e.g. manual vacuum aspiration, dilation and curettage), (vi)performed assisted vaginal delivery (e.g. vacuum extraction, forceps delivery), (vii)performed neonatal resuscitation (e.g. with bag and mask), (viii)performed surgery (e.g. caesarean section), (ix)performed blood transfusion.
VariableMalawi (N = 631) Tanzania (N = 825) Mozambique (N = 587) Average age Female Cadre Enrolled nurses Registered nurses Medical attendants / Medical assistants*/ Clinical officers Doctors Midwives Other Cadre Missing 34 (SD = 10.73) 65.6% (413) 8.6% (54) 62.3% (393) 26.1% (165) Medical assistants* & 1.7% (11) 1.3% (8) 39.69 (SD = 9.51) 75.3% (614) 20.8% (172) 36.5% (301) 40% (330) Medical assistants* & 26.)1% (8) 1.7 % (14) 32.49 (SD = 8.04) 81.79% (476) 60.8% (357) 16.9% (99) Medical assistants* && 18.6% (109) 2.6% (15) 1.2% (7) Table 2. Descriptive statistics for the demographic characteristics and cadre breakdown of participants in Malawi, Tanzania, and Mozambique
DCE - Basic Approach Present different composite jobs Respondents evaluate jobs relative to each other –Rate, rank, discrete choices Analyse choices –Infer underlying value system from the choices made about jobs Can provide estimation of: –Relative importance of different attributes –Willingness of respondents to trade-off between attributes –Relative benefit/utility scores of different combinations –Values of different subgroups
Selection of Attributes Key attributes which define job Limited by experimental design consideration Attributes and levels should be actionable Based on: –Literature review –Expert opinion –Key informant interviews –Focus group discussions –Surveys –Policy relevance –Findings from previous studies (MaxHR)
Job Attributes Geographic Location This attribute specifies whether your place of work is in an urban or rural area. Net Monthly Pay (including regular allowances) Base represents the base salary for a health worker at an “average” grade in the civil service pay scale, while higher levels are multiples (1.5 times and 2 times) of this average base level. Note that the base salary does not necessarily reflect your current actual salary. Government-provided Housing None means there is no housing provided by the government as part of the conditions of employment. Basic housing means the government provides housing for the health worker, but that it is rudimentary, having no electricity or running water, and with at best an outside toilet. Superior housing means the government provides housing of higher quality, including the presence of electricity and running water, including an inside flush toilet.
Job Attributes (cont.) Availability of Equipment and Drugs Inadequate is the standard of equipment and availability of drugs that you might expect in a poorly equipped public facility in the given location. Improved is the level of supplies that would result from a doubling of the budget currently spent on equipment and drugs. Access to Continuing Professional Development This attribute measures the availability of continuing professional development, in terms of access to further education and upgrading. Limited access means there are very few opportunities, with no clear guidelines on who can avail of them. Improved access means there are sufficient opportunities available, with clear policies on the criteria needed to qualify for places.
Job Attributes (cont.) Human Resources Management Systems Poor describes a management system with either no mechanisms or poorly administered mechanisms for staff support, supervision and appraisal. Functioning describes a system where there are transparent, accountable and consistent systems for staff support, supervision and appraisal.
Design Fractional factorial design 15 choice sets 6 attributes –4 with two levels –2 with three levels Job 1 held constant
Table 3: Coding format for the attribute levels (design) AttributeLevels Variable code Code format Location Rural0 Urban location 1 Net monthly pay Base pay1 0 1.5 x base pay2 1 2 x base pay3 2 Housing None houseno 0 Basic houseba 1 Superior housese 2 Equipment and Drugs Inadequate0 Improved equi 1 Professional Development Limited0 Improved pdev 1 Human Resources Management Poor0 Functioning hrm 1
Section L: Discrete Choice Experiment If your circumstances permitted it, which of the two jobs described would you choose? Tick one: Job 1 Job 2
Analysis Initially data was analyzed using the conditional logit model (CLM). The CLM allows observing how the characteristics of the alternatives affect individuals’ likelihood of choosing them; it has been extensively used in the discrete choice model literature (Louviere & Lancsar, 2009; Lanscar & Louviere, 2008; Guttman et al., 2009). The baseline model tested assumed linear effects across all attribute parameters.
Analysis (2) Additionally, to test for non-linear relationship between an attribute and utility, three dummy variables were included to represent each level of the three-level attributes (housing and net monthly pay). The design above was then merged with the dataset containing the choices made by respondents, and the other socio-economic and job related information. control variables representing socio-economic and demographic characteristics are also included in the final dataset that was analyzed: zone, gender, education, age and edu_level.
Dataset (Malawi as example) The original dataset contained 631 respondents and the DCE answers were identified by dce_1, dce_2,…, dce_15, indicating the respondents choices for each of the 15 choice sets presented to them. The final dataset has 9,465 choices made (15 X 631). 74.84% of the choices were for alternative one (job1, constant alternative) and 20.1% for alternative 2 (job 2). Approximately 5% of choice sets were not answered and these were dropped from the final dataset. The final dataset therefore contained 8,986 choices made.
Results All coefficients are statistically significant indicating all attributes have influence on the choice between job1 or job 2. They have positive values, indicating that increases in the level of the attributes increases the utility of choice. These are in accordance with the a priori expectations (external validity).
attribute Coef. zP>z location0.2154.090.0000 pay1.23329.470.0000 housing0.65217.410.0000 equi0.4027.070.0000 pdev2.03936.810.0000 hrm2.27629.890.0000 Number of obs 17972 Log likelihood -7814.56 Table 4: Conditional logit model results (Malawi) – baseline model The attribute human resources management has the highest absolute value (hrm =2.276) while the attribute location had the smallest absolute value (location=0.215).
Attribute Coef. zP>z location0.45711.40.000 pay0.47917.990.000 housing0.1023.80.000 equi0.0120.320.000 pdev1.19931.60.000 hrm1.18125.610.000 Number of obs 23034 Log likelihood -11894.99 Table 5: Conditional logit model results (Tanzania) – baseline model Attributes with highest part-worth utilities were professional development (pdev=1,199) and human resources management (hrm =1,181). An improvement in any of these two attributes impacts more on the utility than any other attribute in the design.
Attribute Coef. zP>z location0.3166.520.000 pay0.60117.960.000 housing0.2658.160.000 equi0.3076.330.000 pdev1.53432.720.000 hrm1.33222.710.000 Number of obs 16918 Log likelihood-8577.44 Table 6: Conditional logit model results (Mozambique)– baseline model Attributes with greater utility were professional development (pdev=1,534) and human resources management (hrm =1,332).
Testing for non-linear effects Two of the six attributes had 3 levels, net monthly pay and housing, To test for non-linear effects – including in the model the dummy variables for housing and pay attributes (Test for non-linear effects allows observing whether the effect on utility from an increasing in the salary level (or housing) from the basic salary to 1,5 the basic (or from no housing to basic housing) is different from an increase from 1,5 the basic to 2 times the basic (or from basic housing to superior housing).) They were included separately and the goodness of fit was compared with the baseline model of linear effect of each three levels attribute a Wald test was applied to check whether or not the dummy variables included were different from zero. If so, it implies that there are non-linear effects on the three levels attributes, i.e., the impact on utility is different when moving from pay1 to pay2 compared to a change from pay2 to pay3 (or houseno to houseba compared to houseba to housesu).
Results Expanded model did not provide a better fit for the data. Non-linearity detected for pay only in Malawi
Table 7: Conditional logit model results (Malawi) – Model 2 attribute Coef. zP>z location-0.123-2.170.0000 Pay 1.5 base1.99530.240.0000 Pay 2 base2.08622.540.000 housingba1.56222.210.0000 housingsu1.36118.040.000 equi1.01915.730.0000 pdev1.38923.520.0000 hrm1.81822.670.0000 Number of obs17972 Log likelihood-7814.56 Marginal diminishing return for housing i.e. moving from level 2 to level 3 has less influence on choice of job than moving from level 1 to level 2
Table 8: Conditional logit model results (Tanzania) – Model 2 attribute Coef. zP>z location0.2676.260.000 Pay 1.5 base0.93721.700.000 Pay 2 base0.62210.490.000 housingba0.90617.310.000 housingsu0.2985.30.000 equi0.49410.870.000 pdev0.74117,460.000 hrm0.89017.750.000 Number of obs23034 Log likelihood-11894.99 Marginal diminishing return for pay and housing i.e. moving from level 2 to level 3 has less influence on choice of job than moving from level 1 to level 2
Table 9: Conditional logit model results (Mozambique) – Model 2 attribute Coef. zP>z location0.1242.400.0000 Pay 1.5 base1.05819.990.0000 Pay 2 base0.84911.370.000 housingba1.01115.850.0000 housingsu0.6539.760.000 equi0.76213.690.0000 pdev1.11621.710.0000 hrm1.02316.090.0000 Number of obs16918 Log likelihood-8577.44 Marginal diminishing return for pay and housing i.e. moving from level 2 to level 3 has less influence on choice of job than moving from level 1 to level 2
In Summary Consistent results across three countries Strongest predictors of job choice - access to CPD and HRM Strong preferences for functioning HRM and available professional development that operates with clear policies Consistent with other studies – pay is important but perhaps not as fundamental as suggested by previous studies Further analysis – differences between cadres, demographic profiles of health worker.
Additional data Demographics Job title Employment status Employer type Employer location Gender Age Education Professional affiliations Length of time with employer Work pattern Payment patterns Provider survey Job satisfaction Burnout levels Work environment Commitment Intention to leave Organisational justice Supervision Career progression opportunities
Limitations of DCE Stated vs actual preferences –Artificial / hypothetical constructs may not predict real choices Limited number of attributes and levels –Significant design constraints Have the most influential attributes been selected? –Different results with different attributes In this study qualitative and quantitative data collected using a variety of instruments are consistent with DCE findings.
With Thanks HSSE Team: AMDD, Mailman School of Public Health, Columbia University, USA Centre for Global Health, Trinity College, University of Dublin Centre for Reproductive Health, College of Medicine, Malawi Dept. of Community Health, Eduardo Mondlane University, Mozambique Ifakara Health Institute, Tanzania Realizing Rights: Ethical Globalization Initiative, USA Regional Prevention of Maternal Mortality Network, Ghana Funders: IrishAid & Ministry of Foreign Affairs, Denmark email@example.com