Data
The study used data from the 2007 Bangladesh Demographic and Health Survey (BDHS) and 2017–18 BDHS, which are nationally representative surveys conducted by the National Institute for Population Research and Training (NIPORT) of the Ministry of Health and Family Welfare [29, 30]. The 2007 BDHS survey used the sampling frame provided by the list of census enumeration areas (EAs) with population and household information from the 2001 Population Census. Bangladesh is divided into six administrative divisions. In turn, each division is divided into zilas, and each zila into upazilas. Rural areas in an upazila are divided into union parishads (UPs), and UPs are further divided into mouzas. Urban areas in an upazila are divided into wards, and wards are subdivided into mahallas. EAs from the census was used as the Primary Sampling Units (PSUs). The survey is based on a two-stage stratified sample of households. At the first stage of sampling, 361 PSUs were selected. The selection of PSUs was done independently for each stratum and with probability proportional to PSU size in terms of a number of households. The urban areas of each division were further subdivided into three strata: statistical metropolitan areas (SMAs), municipality areas, and other urban areas. In all, the sample consisted of 22 strata because Barisal and Sylhet do not have SMAs. The 361 PSUs selected in the first stage of sampling included 227 rural PSUs and 134 urban PSUs. On average, 30 households were selected from each PSU, using an equal probability systematic sampling technique. A total of 10,996 ever-married women aged 15–49 from 10,400 households were interviewed with a response rate of 98.5% and 99.5%, respectively.
The 2017–18 survey used a sampling frame from the list of enumeration areas (EAs) of the 2011 Population and Housing Census of the People’s Republic of Bangladesh, provided by the Bangladesh Bureau of Statistics (BBS). The primary sampling unit (PSU) of the survey is an EA with an average of about 120 households. The survey is based on a two-stage stratified sample of households. In the first stage, 675 EAs (250 in urban areas and 425 in rural areas) were selected with probability proportional to EA size. In the second stage of sampling, a systematic sample of an average of 30 households per EA was selected to provide statistically reliable estimates of key demographic and health variables for the country as a whole, for urban and rural areas separately [31]. A total 20,127 ever-married women aged 15–49 years from 19,457 households were interviewed with a response rate of 98.8% and 99.4%, respectively.
The BDHS obtained detailed information on fertility levels, marriage, fertility preferences, awareness and use of family planning methods, breastfeeding practices, nutritional status of women and young children, childhood mortality, maternal and child health, and knowledge and attitudes regarding HIV/AIDS and other sexually transmitted infections. The detailed information on the survey is given elsewhere [30]. The effective sample size for the current analysis was 1157 (2007) and 1660 (2017–18) women aged 15–49 years who had given birth at home or a health facility during three years preceding the survey.
Variable description
Outcome variable
The outcome variable was whether the child received all basic vaccination or not. Full vaccination includes one dose of BCG against tuberculosis, three doses of DPT (diphtheria, whooping cough, and tetanus), three doses of oral polio vaccine and one dose of measles vaccine among 12–23 months children [32]. The variable was coded as 0 “received full vaccination” and 1 “did not receive full vaccination”. The DHS collected the full vaccination status of children from the two sources. Primarily immunisation record cards were provided by mothers, but if these were absent in the DHS the data collectors used mothers’ verbal reports of children’s immunisation status [33].
Equity stratifier
The wealth index was the equity stratifier in the present study. The wealth index has a natural ordering and known as an ordered stratifier used in several socioeconomic-related inequality studies and has a high predictive value in low and middle-income countries [34, 35]. It was coded as poorest, poorer, middle, richer, and richest [32, 34]. Households were given scores based on the number and kinds of consumer goods they own, ranging from a television to a bicycle or car, and housing characteristics such as the source of drinking water, toilet facilities, and flooring materials. These scores are derived using principal component analysis. National wealth quintiles are compiled by assigning the household score to each usual (de jure) household member, ranking each person in the household population by their score, and then dividing the distribution into five equal categories [32, 34].
Explanatory variable
The study added the explanatory variables based on the literature available [8, 23, 36]. The sex of the child was coded as male and female, the mothers’ age was coded as 15–19, 20–14, 25–29 and 30 + years, mothers’ and fathers’ educational status was coded as not educated, primary, secondary and higher. Mothers’ and Fathers’ working status was coded as not working and working. Media exposure which includes exposure to television, radio and newspaper, was coded as exposed to anyone and not exposed [37]. Parity was coded as 1, 2, 3 and 4 and above. Preceding childbirth interval was coded as less than 24 months, 24–36 months and more than 36 months [33, 38]. Antenatal care visits were coded as no visit, 1–3 visits and 4 or above visits [38]. Postnatal care within two days of delivery was coded as no and yes [33]. The place child delivered was coded as home and health facility.
Religion was coded as Islam and others, and residential status was coded as urban and rural. Administrative divisions were provided in the survey as Barisal, Chittagong, Dhaka, Khulna, Rajshahi and Sylhet. To be noted in Bangladesh Demographic and Health Survey 2017–18 had eight administrative divisions, namely Barisal, Chittagong, Dhaka, Khulna, Mymensingh Rajshahi, Rangpur and Sylhet. For analytical reasons, Mymensingh was merged in Dhaka and Rangpur was merged in Rajshahi as these were divided from these regions at certain points after 2004.
Statistical analysis
Descriptive (percentage) along with bivariate analysis was used for carving out preliminary results. The Chi-square test was used to resemble the significance level (p-values) during bivariate association. Along with that, binary logistic regression analysis [39] was performed to estimate the association between outcome and explanatory variables.
Concentration Index (CCI)
The concentration curve is obtained by plotting the cumulative proportion of outcome variables (vaccination status) on y-axis against the increasing percentage of the population ranked by the socioeconomic indicator (wealth index) on x-axis. The curves show that whether the socio-economic status related inequality in the outcome variable (on x-axis) prevails or not [32, 37]. If the curve is above the line of equality (45 degree line) that means the index value is negative; hence it shows that the outcome variable is disproportionally concentrated among the poor and vice-versa [32, 37]. Income-related inequality in the vaccination status was measured by the concentration index (CCI) and the concentration curve (CC), using the wealth score as the socioeconomic indicator and binary outcome as vaccination status [32, 37]. The concentration index is defined as twice the area between the concentration curve and the line of equality. The concentration index measures the inequality of one variable (vaccination status) over the distribution of another variable (wealth index) [40]. The index ranges from -1 to + 1, where the index value of 0 (zero) shows no socioeconomic inequality [40]. However, the positive value of the index shows pro-rich inequality and vice-versa. Additional on either scale higher the value, the higher the extent of socioeconomic inequality.
CCI (concentration index), WI (Wagstaff's index), and EI (Erreygers index) are all binary variables that condition the level of absolute inequality on the most unequal society, although their definitions of that state differ [41]. CCI responds to the issue of how far the society has progressed from a state in which the wealthiest individual owns all of the society's health units (without considering the upper bound of the variable) [41]. WI and EI, on the other hand, acknowledge the boundedness of the health variable; WI answers the question of how far the society is from a state where only the top half of the income distribution is healthy, regardless of prevalence, while EI answers the question of how far the society is from a state where only the top half of the income distribution is healthy, regardless of prevalence [41].
The study used Wagstaff decomposition analysis to decompose the concentration index. Wagstaff’s decomposition demonstrated that the concentration index could be decomposed into the contributions of each factor to the income-related inequalities [42]. For any linear regression model on health outcome (y) (vaccination status), such as.
$$y=\alpha +{\sum}_k{\beta}_k{x}_k+\varepsilon$$
(1)
The concentration index for y, C, can be written as follows,
$$C={\sum}_k\left({\beta}_k{\overline{x}}_k/\mu \right){C}_k+G{C}_{\varepsilon }/\mu$$
(2)
Where \(\mu\) is the mean of y, \({\overline{x} }_{k}\) is the mean of \({x}_{k}\), \({C}_{k}\) is the concentration index for \({x}_{k}\) (defined analogously to C), and \(G{C}_{\varepsilon }\) is the generalized concentration index for the error term (\(\varepsilon )\). Equation (2) shows that C is equal to a weighted sum of the concentration indices of the k regressors, where the weight for \({x}_{k}\) is the elasticity of y with respect to \({x}_{k}\)
\(\left({\eta }_{k}= {\beta }_{k}\frac{{\overline{x} }_{k}}{\mu }\right)\). The residual component captured by the last term reflects the socioeconomic status related inequality in health that is not explained by systematic variation in the regressors by income, which should approach zero for a well-specified model [32]. Each contribution is the product of elasticity with the degree of economic inequality [32]. Moreover, the percentage contribution is obtained by dividing each absolute contribution by total absolute contribution multiplied by 100 to obtain the estimates [32, 38]. The positive contribution indicates the role of factors in the extent of higher inequality, and the negative contribution explains the extent of reduction in inequality. Multicollinearity was assessed using variance inflation factor (VIF) [43]. Svyset command was used in STATA 14 to account for complex survey design. Further, individual weights were used to make the estimates nationally representative. STATA 14 [44] was used to analyse the dataset.