Willingness to Pay for Health Insurance among Urban Poor: An Evidence from Urban Primary Health Care Project in Bangladesh

Background and Objective: With the growing concern over treatment cost in health care and the desire to improve the effectiveness and equality of healthcare financing and the quality of the care, policy-makers have turned their attention to health insurance, especially, for the poor. This study attempted to determine the willingness to pay for health insurance among the mothers who utilized the urban primary health care clinic (UPHCC) for maternal and child health. Methods: This cross-sectional study was carried out in the working areas of UPHC Project in Bangladesh following two-stage cluster sampling technique to select the participants. Data were collected from 3949 women aged 15-49 years having at least one child aged two years or less. The data on willingness to pay for health insurance was collected using the contingent valuation method with bidding style. Data analysis was done by SPSS 22.0 version. Two generalized linear models with binary logit link function and normal identity link function were developed to identify the potential predictors for willingness to pay for monthly health insurance. Results: Three-fifths (67.5%) of the respondents agreed to pay for monthly health insurance. The median monthly premium for health insurance was BDT 15.5. Multivariate analysis revealed that utilization of UPHC clinic, quality of life, family size, age, wealth index, level of education, husband and respondent’s occupation, ownership status of the house, religion and family income appeared to be potential predictors for health insurance (p<0.05). However, utilization of UPHC clinic and quality of life appeared to be important predictors across all the models. Conclusion: A large proportion of the community agreed to pay premium for health insurance. Based on the finding of the current study the policy makers might consider introducing a scheme for health insurance especially among the urban poor.


Introduction
There are many countries implementing health insurance systems, but they characteristically cover only civil servants and other formal-sector employees. Health insurance systems are then reviewed with the poor in attention 1 . Some initiatives have been involved to encompass coverage to the segment of poor who are able to pay at least certain premiums, or whose care can be shielded by cross-subsidies using funds upraised from those who are less poor in Latin America. In Colombia and Mexico, to identify poor households, simplified forms of proxy mean testing are used. Seguro Popular, a motivated set of reforms that has been introduced by Mexico targets to enroll the majority of the poor by the middle of the next decade 2-4 . Dercon, 5 explained that to expand the access of the poor to the monies they need for health care, health insurance is not the only way. Options embrace improving capacity of the poor to deposit their savings in banks and other formal sector commercial institutions and mounting access to short-term credit. Financial institutions can have a vital role to play in improving the health of the poor not only viewed as outposts of the health sector. Uzochukwu et al. 6 had argued that willingness to pay for Maternal and Child Health (MCH) services if the quality of care to be improved. The study reported a positive attitude towards willingness to pay for readily available services. A study in Ethiopia, WTP for injectable contraceptives was assessed using a dichotomous yes/no response and the follow up question was open ended to inquire how much they wanted to pay. The study revealed at least $0.65 spent for injectable contraceptives 7 . Lang and Lai 8 in Taiwan investigated the public willingness to pay for National Health Insurance (NHI) program by using contingent valuation method asking whether they would agree to pay at already fixed six price bids. The study found that the people were willing to pay NT$66 for NHI and NT$ 137 for added long-term care services per month. These referendum-like close ended questions were vulnerable to bias associated with the range of payment used in the study.
The introduction of social health insurance or copay proposal by the Ministry of Health would be inevitable. It was hoped that the reformation in funding would also bring an improvement in the service provided in term of quality, equity and timely accessibility. The issue of affordability would surely affect the acceptability of health care by the public. Therefore, one of the ways to roughly estimate the ability to pay by the public was by doing willingness to pay study 9 . Considering this view, this study attempted to assess the willingness to pay for health insurance among the urban mother attending the urban primary health care clinic and the factors associated with it.

Participants
A total of 3949 women were interviewed. The mean (SD) age of the respondents was 25.49 (5.1) years with a minimum 15 and maximum 45 years.
Most of the respondents were Muslim (90.4%) and remaining were Hindus (9.2%), Christian (0.2%) and Buddhist (0.2%). The mean (SD) family size was 4.71 (1.57) ranging from two to 15 members. About two-fifths of the respondents (38.2%) had completed primary education followed by secondary education (32%) and more than one-tenth (11.5%) had a higher secondary education. More than fourfifths (83.1%) were housewives. Only about 4.3% of the respondents were garment workers followed day labourers (3.1%). More than one-fourth (27.3%) of their husbands were engaged in small trade and 17.9% were engaged in private service followed by day labourer (16.5%). In Urban Primary Healthcare Project (UPHCP) working area, an entitlement cards (red card) were issued to the poor households to avail healthcare services free of charge from UPHCP facilities. Out of 3949 respondents, 20% entitlement card holder presented their cards to the interviewer However, another 5% did not show a card, and they claimed to have.

Sample and sampling procedure
This was a cross-sectional study conducted in working areas of Urban Primary Healthcare Project (UPHCP) areas in seven divisions of Bangladesh. A two-stage cluster sampling technique was adopted to select the ever-married women aged 15-49 years having at least one child aged within two years. For determination of sample size, an anticipated population proportion of urban poor was considered as base criteria with 10% relative precision and 95% confidence interval. The initial sample size was 896. This was then inflated by multiplying nonresponse, design effect. Thus, the final sample size was 4124. However, 3949 completed questionnaires were collected with the response rate was 95.8%. The respondents who did not consent or unwilling to participate, visitor or guest residing in household were excluded from the study.

Data collection instruments and procedure
Willingness to pay is a methodological tool that seeks to estimate the capacity to pay for certain social groups in a search to find out the hypothetical monetary value for programs and specific medical interventions and treatment 9 . Willingness to pay in this study used contingent valuation method. There were two most common methods of eliciting willingness to pay which are conjoint analysis and contingent valuation method 10 . Contingent valuation method with closed ended bidding technique was the most commonly used and easier to apply [11][12][13][14] . The approach asked respondents about how much they were willing to pay for the goods (usually public goods). To be able to elicit the willingness to pay value from the respondents, a description of the commodity or scenario needed to be given first. The nature of the description and the clarity of the information given could influence the willingness to pay pattern by the respondents and therefore, must have as minimum bias as possible 15,16 . There was a payment scenario describing the willingness to pay for monthly health insurance. After the explanation of the scenario and the payment methods, the question on whether they agree to pay would be asked first. If they agree, then the bidding process would commence. The answers to the bidding are close ended with four options ('yes', 'may be yes', and 'may be no 'and 'no'). However, if the respondents answer 'no' or 'may be no', they would be asked about the reason and the interview stopped there. This is to differentiate the 'protest no' and the 'real no'. The bidding would start with the lowest value which would be obtained from the pre-test and afterward greater value and stopped if the participants opined that they were reluctant to pay the given amount. Again, the answer options are 'yes', 'may be yes', 'may be no' and 'no'. Once the respondents respond no or yes to the whole bidding, open ended question would ask about what the uppermost amount that they were willing to pay.

Statistical analysis
A completed data was entered into computer for analysis. Incomplete and inaccurate and missing information in the main components of the questionnaire was discarded. Before data analysis, data was cross-checked for any unusual findings, outliers and missing values 17 . Statistical analysis was done using Statistical Package SPSS, version 22.0 18 . A p-value less than 0.05 was considered as statistically significant.
In the present study, an econometric model for willingness to pay for health insurance was developed. The most preferred method was generalized linear model with binary logit link function [19][20][21] . A generalized linear model (or GLM) consists of three components: 1. A random component, specifying the conditional distribution of the response variable, Yi (for the ith of n independently sampled observations), given the values of the explanatory variables in the model. In the initial formulation of GLMs, the distribution of Yi was a member of an exponential family, such as the Gaussian, binomial, Poisson, gamma, or Inverse-Gaussian families of distributions.

2.
A linear predictor-that is a linear function of regressors,

A smooth and invertible linearizing link function g(·)
, which transforms the expectation of the response variable, μi = E(Yi ), to the linear predictor: The binomial distribution for the proportion Y of successes in n independent binary trials with probability of success μ has probability function ………………… (3) Here In our first model, generalized linear model with binary logit link function was used. In the scenario 1, the respondents were asked whether they want to participate for maternal health care coverage with monthly premium. In the first scenario, whether they agree or disagree for health insurance as a dichotomous variable, i.e. yes vs. no. Significant independent variables were fed into the model one by one and the best fitted ones were chosen. Assumptions of adequate sample size, multicollinearity and absence of outliers were examined. A total of 164 data were removed due to extreme outliers. In this analysis, robust estimation was choose to minimize the potential outliers and to produce stable model 22 . Finally, the estimated marginal means were calculated for the mean response for each factor, adjusted for any other variables in the model that is the estimated marginal means adjust for the covariate by reporting the means of Y for each level of the factor at the mean value of the covariate. Pairwise comparison with Bonferroni adjustment was done to minimize the type I error. The data were interpreted into non-poor and poor strata to compare the results.
However, after removal of '0' from the data set, the model fitting information (value/df) of Pearson Chi-Square and Deviance was much smaller than 0.05 which indicate poor fitting model in zero truncated Poisson regression. Categorical and continuous independent variables were fed into the model one by one and the best fitted one was chosen. Assumptions of adequate sample size, multicollinearity and absence of outliers were examined. Finally, 445 data were removed from the due to extreme outliers.

Ethical issues
The Before data collection, an informed written consent was obtained and they were briefed about a) the objectives, steps and estimated result of the study, b) welfares of the research, in term of welfares to the subject and benefits to others; c) absolute confidentiality of facts gained; and e) the right to take away from the study at any time wanting any way upsetting her current situation.

Willingness to pay for insurance
The respondents were informed by giving a scenario of improved service and facilities in PHC setting for MCH services with health insurance as the financial resources. They were asked whether they agree to pay for the monthly insurance or not and the amount they were willing to pay. Three-fifths of the respondents (67.5%) agree to pay for monthly health insurance. Among the respondents those who disagree, 29.3% were unable to afford the payment while 25.7% put the responsibility to the government. More than half of them agreed to pay around BDT 10-20 but the highest value that the 0.1% respondents were willing to pay was BDT 1000. The median monthly premium for health insurance was BDT 15.5 ( Table 1).   Table 2).

Generalized linear model with normal identity link function
Analysis revealed that occupation of the husband, utilization of urban primary health care clinic and age of the respondents appeared to be the important predictors for willingness to pay for health insurance (p<0.05) for both non-poor and poor catchment areas, however, level of education, family income (BDT) appeared to be significantly influence WTP for health insurance among the non-poor whereas, religion, ownership status of house, having a red card and quality of life significantly influence willingness to pay for the poor catchment area (p<0.05). Respondent having primary and secondary level of education in both non-poor and poor catchment area more likely to WTP for health insurance however, non-formal education and higher secondary level of education among poor likely to pay health insurance, but not among the non-poor (p>0.05). Similarly, those who never or partially utilized urban health care clinic were likely to pay health insurance premium compared to full utilization in both non-poor and poor catchment areas. Among the poor catchment area, the non-Muslim respondents were more likely to pay health insurance, similarly, those were engaged in small trade and manual job were more likely to WTP for health insurance (p<0.05). Those who were living in their own house or a rented house were also more likely to pay health insurance (p<0.05). But those who had no entitlement card (red card) were likely to pay health insurance premium in the poor catchment area. Analysis revealed that with increasing quality of life among the respondents, they were more likely to pay health insurance. In contrast to the family income, respondents with family income less than BDT 10,000 to 30,000 were less likely to pay health insurance premium among the non-poor catchment area (p<0.05) ( Table 3).

Discussion
In our study, more than three-fifths of the respondents (67.5%) agreed to pay for monthly health insurance. About 29.3% were unable to afford the payment while 25.7% put the responsibility to the government. More than half of them agreed to pay around BDT 10-20 (USD 0.12-0.24). The higher the bid was made, the lower number of respondents who were willing to pay.
In Bangladesh, Ahmed et al. 23 28,45 . In Iran, WTP decreased with increasing the family size as in the present study 28 .
In contrast, a systematic review in the past found that increases in family size were correlated with higher WTP for insurance, but similar with the present study with education level and income were consistently correlated with higher WTP for insurance and increases in age were correlated with reduced WTP 46 . On the effect of age on WTP, Oyekale 47 showed a strong negative correlation, but Babatunde et al. 48 found a significant influence of age on WTP for health insurance. The current study is in line with Oyekale's findings but inconsistent with Babatunde's assessment. Increase quality of life significantly influenced WTP and ownership of house had no significant effect on WTP which agreed with Smith and Cunningham 49 , although the purpose for WTP was not the same. On the other hand, respondents who resided in their own house or in rented house in the poor catchment area had a positive influence on WTP which might be due to living security they want to pay for monthly HI premium.
Our results are based on urban population and our study sample was not representative of the national population. Given that our sample was relatively poor and that the amount of willingness to pay perhaps may not reflecting the wealth and the characteristics of our sample limit the generalizability of the results. The sample population of general population may result in different values WTP. Next, the range of response choices for the payment of WTP question may have influenced participant responses to the open-ended question. However, since we pre-tested our WTP questions to choose ranges for the paymentscale question, it may also be that the paymentscale ranges accurately captured the range of values respondents had in mind for WTP. But it might be a response like nodding the head of "Ya Ya" distort the results.

Conclusions
Analysis revealed that utilization of urban primary healthcare clinic, occupation of the husband, family income (BDT), age in years appeared to be important predictors for willingness to pay for health insurance in both models. However, other variables such as wealth index, family size and ownership status of the house, entitlement card and quality of life varied in significance across the model. The goal for the Bangladesh government is to ensures quality of health service with equity. Thus, reaching the goal, the findings of this study might hold the key to strategic planning. Though Our findings on the determinants of WTP are, in this light, somewhat encouraging. WTP increase with increasing age and rises with more education. Respondents with higher education may be more encouraging to pay. A key task for policymakers is to implement a health insurance scheme on a pilot basis, particularly among the urban poor with low education.