Data source and population

This study used data from the 2016 EDHS (Ethiopia Demographic and health survey) which was conducted from January 18 to June 27, 2016. Thus, a community- based cross-sectional study was employed to identify individual- level and community- level factors affecting exclusive breastfeeding among under-six month infants [14].

The source population for this study was all infants undersix months of age living in Ethiopia and all infants undersix months of age in the selected enumeration areas during the data collection period were the study population. Accordingly, a total of 1185 infants under six months of age who fulfill the eligibility criteria were included in the study using a stratified two stage cluster sampling technique. Enumeration areas (EAs) and households were the primary and secondary sampling units respectively. The detailed sampling procedure was published elsewhere [14,24].

We extracted the outcome and explanatory variables from the EDHS 2016 kids data after getting permission from Measure DHS International Program. A structure and pretested questionnaire was used for data collection during the survey. Comprehensive information about the 2016 EDHS data collection procedure has been published elsewhere [14,24,25].

Variables of the study and operational definition

The outcome variable of the study was Exclusive Breastfeeding (EBF) among under-six month infants in Ethiopia. Infant related variables (Age of the infant, sex of the infant, birth order, birth weight, birth interval, infant comorbidities, and time of breastfeeding initiation), maternal socio-demographic variables (age of the mother, marital status, educational status of the mother, occupational status of the mother, household wealth index, media exposure, number of underfive children, and household family size), and Obstetric and Healthcare associated variables (ANC utilization, PNC utilization, place of delivery, mode of delivery, delivery assistance, and parity) were individual- level predictors of the study. Also, community -level variables included in the study were place of residence, contextual region, community media exposure, community wealth index, community women education, community ANC utilization, community level of employment, and community PNC utilization.

The World Health Organization (WHO) defines exclusive breastfeeding (EBF) as when ’an infant receives only breast milk, no other liquids or solids are given not even water, except for oral rehydration solution, or drops/syrups of vitamins, minerals or medicines [4]. It was measured using a 24-hour recall among mothers with infants under 6 months of age [4].

In this study, infant comorbidities were generated by aggregating variables (diarrhea, cough, fever, and shortness of breath of infants under six months of age) in the last two weeks preceding the survey, subsequently if at least one of the comorbid was found the response coded as “Yes” otherwise the response coded as “No. The occupational status of the mother was generated by using the occupation of the respondent and categorized as currently working (includes all types of work) and currently not working.

Community -level variables were generated by aggregating the individual characteristics in a cluster since EDHS did not collect data that can directly describe the characteristics of the clusters except the place of residence. The aggregates were computed using the proportion of a given variables’ subcategory. Since the aggregate value for all generated variables not normally distributed, it was categorized into groups based on the national median values and based on previous related studies [24,25].

Community ANC utilization was the proportion of mothers within specific cluster who visited ANC some number of times. It was categorized using national level quartiles in to low ANC utilized community (when≤25% of women are utilizing ANC), middle (when 25–75% of women are utilizing ANC), and high (when >75% of women are utilizing ANC) [25].

The Community level of PNC utilization was the proportion of women within specific cluster who visit PNC some number of times. It was categorized as low (when≤50% of women utilized PNC) and high (when>50% of women utilized PNC) [25].

Community- level of media exposure was an aggregate respondent level of exposure for different types of media categorized as “<25% = Low”, “25%-50% = Moderate” and “>50% = high media utilized communities” [25].

Community level of poverty was an aggregate wealth index categorized as “<25% = High”, 25%-50% = Moderate” and “>50% = Low poverty communities” [25].

Contextual region Ethiopia is demarcated for administrative purpose into 11 regions, which are classified as an agrarian, pastoralist, and city based according to the living status of the population. The regions Tigray, Amhara, Oromia, SNNP, Gambella, and Benshangul Gumuz were categorized as agrarian. Somali and Afar regions were grouped to form pastoralist region and Harari region, Addis Ababa and Dire Dawa city administrations were grouped to form city- based populations [14,25].

Community- level of women education was the proportion of women in the community who have primary or higher education, which was categorized as low (when≤25% of women were educated), middle (when 25–75% of women were educated), and high (when >75% of women were educated) [25].

Community level employment status was the proportion of women who were employed (had to work) in the specific cluster. It was categorized as low (when≤50% of mothers were employed) and high (when>50% of mothers were employed) [24,25].

Data management and analysis

Sample weight was done to compensate for the unequal probability of selection between the strata that were geographically defined, as well as for non-responses. Weighing of individual interview produces the proper representation of exclusive breastfeeding and related factors. Coding, recoding, and exploratory analysis was performed. Categorization was done for continuous variables using information from different works of literatures and re-categorization was done for categorical variables accordingly. For data analysis, STATA version 14.1 was used and descriptive statistics were used to present frequencies, with percentages in tables and using texts.

Four models were considered in the multilevel analysis to determine the model that best fits the data; Model one (Null model) without explanatory variable was developed to evaluate the null hypothesis that there is no cluster level difference in exclusive breastfeeding practice that specified only the random intercept and it presented the total variance in exclusive breastfeeding practice among clusters. Model two adjusted for an individual variable which assumes a cluster level difference of EBF practice is zero. Model three to evaluate community level factors by aggregate cluster difference of exclusive breastfeeding practice. Model four included both adjusted individual and community level factors.

The log of the probability of Exclusive breastfeeding was modeled using a two-level multilevel model as follows:

Where, i and j are the level 1 (individual) and level 2 (community) units, respectively; X and Z refer to individual and community-level variables, respectively; πij is the probability of exclusive breastfeeding for the i^th infant in the j^th community; the β’s is the fixed coefficients. Whereas, β0 is the intercept-the effect on the probability of exclusive breastfeeding use in the absence of influence of predictors and uj showed the random effect (effect of the community on exclusive breastfeeding) for the j^th community, and eij showed random errors at the individual levels. By assuming each community had a different intercept (β0) and fixed coefficient (β), the clustered data nature and the within and between community, variations were taken into account.

Multilevel logistic regression analysis was used to analyze the data since it is appropriate for DHS data as it had a hierarchical nature. Multilevel modeling was providing unexplained variation in exclusive breastfeeding due to unobserved cluster factors called the random effect. All models included a random intercept at the cluster level to capture the heterogeneity among clusters.

The measures of association (fixed-effects) estimate the association between likelihood of infants to exclusively breastfeeding as the AOR with 95% CI of various explanatory variables were expressed. Crude association between independent variables and the dependent variable was done independently and variables having p ≤0.2 in Bi-variable analysis were used to fit multivariable analysis model. At multivariable analysis,variables with p≤0.05 with confidence interval not including the null value (OR = 1)were considered as statistically significant variables with exclusive breastfeeding practice.

The measures of variation (random-effects) were reported using Intra-cluster correlation (ICC), Median Odds Ratio (MOR), and Proportional Change in Variance (PCV). ICC was used to explain cluster variation while MOR is a measure of unexplained cluster heterogeneity [27]. The ICC shows the variation in exclusive breastfeeding of infants undersix months of age due to community characteristics. The higher the ICC (ICC>5%), the more relevant was the community characteristics for understanding individual variation in exclusive breastfeeding of infants. The ICC can be calculated as follows: $$ \left[\text{ICC}=\frac{{\delta }^2\text{u}}{{\delta }^2\text{u}+{\delta }^2\text{e}}\right]$$ where δ²u = between group variation, δ²e = with in group variation OR $$ [\text{ICC}=\frac{{\delta }^2}{{\delta }^{2+\frac{{\pi }^2}{3}}}]$$, where δ² is the estimated variance of clusters [26]. The STATA software command can also compute the ICC value of each model.

MOR is defined as the median value of the odds ratio between the area at highest likelihood and the area at the lowest likelihood of exclusive breastfeeding when randomly picking out two areas and it measures the unexplained cluster heterogeneity; the variation between clusters by comparing two persons from two randomly chosen different clusters. MOR can be calculated using the formula

In this study, MOR shows the extent to which the individual probability of being exclusively breastfed is determined by residential area. The proportional change in variance [PCV = (VA − VB)/VA) * 100] where VA = Variance of initial model and VB = Variance of model with more terms measures the total variation attributed by individual level and community level factors in the multilevel model [26]. PCV was computed for each model concerning the empty model as a reference to show power of the factors in the model explains exclusive breastfeeding practice.

Log- likelihood test, Deviance Information Criteria (DIC), and Akaike Information Criteria (AIC) were used to estimate the goodness of fit of the adjusted final model in comparison to the preceding models (individual and community level models). Thus, the model with the highest value of Log likelihood test and with lowest values of DIC and AIC was considered to be the best fit model.

Ethical consideration

Ethical clearance was obtained from the Ethical Review Committee of the College of Medicine and Health Sciences, Wollo University with approval and supporting letter. Permission to access the data set was obtained from the Measure DHS International Program. The data was only used for purpose of this study and not shared with a third party. All data used in this study were anonymous publicly available and aggregated secondary data with not having any personal identity. The data was fully available on the full DHS website (www.measuredhs.com).

Fixed effects (Measures of association)

Bi-variable multilevel logistic analysis computed and p-value up to 0.2 was selected to fit the multi-variablemultilevel logistic regression model to control confounding. Accordingly, Age of infant, sex of infant, birth weight, birth interval, birth order, infant comorbidities, breastfeeding initiation, marital status, maternal occupation, maternal education, household family size, number of under-five children, wealth index, parity, ANC visit, caesarian delivery, PNC visit, residence, contextual region, community ANC utilization, community PNC utilization, community media exposure, community level of poverty, community level of employment status and community level of women education pass Bi-variable multilevel logistic regression model at p- value≤ 0.2.

Four models were fitted hierarchically in this multi-level logistic regression analysis. Model one or Null model (without predictors), model two (only individual- level factors), model three (only community- level factors) and model four (both individual and community level factors). Finally, p-value ≤ 0.05 and odds ratio not including the null value (OR = 1) was used to select variables that had a statistically significant association with exclusive breastfeeding practice in the final model.

As shown in Table 5 below, our study finding showed that age of infants, sex of infants, infant comorbidities, wealth index, antenatal care visit, contextual region, community postnatal care visit, and community level of employment status were statistically significant determinants of exclusive breastfeeding in the final model.

Table 5

Multilevel logistic regression analysis of both individual and community level factors associated with exclusive breastfeeding among infants under 6 months of age in Ethiopia, 2016 (N = 1185).

Variables	Categories	Model one	Model two AOR (95%CI)	Model three AOR (95%CI)	Model four AOR (95%CI)
Age of infant	0–1 month		1		1
	2–3 months		0.34(0.21,0.56)***		0.39(0.23, 0.65) ***
	4–5 months		0.05(0.02,0.08)***		0.042(.02, 0.07) ***
Sex of infant	Male		1		1
	Female		2.43(1.61,3.65) ***		2.51(1.61, 3.91) ***
Birth interval	No previous birth		1		1
	<24 months		4.27 (1.36, 13.39)*		3.44(0 1.19, 9.92)
	> = 24 months		2.73(1.00, 7.41)		1.84(0.74, 4.54)
Birth order	First		1		1
	2–3		2.71(0.99, 7.34)		1.821(0.15, 2.86)
	4+		3.32(0.93,11.87)		1.46(0.97, 2.18)
Infant comorbidities	No		1		1
	Yes		0.34(0.19, 0.56)***		0.35(0 .21, 0.57)***
Time of BF initiation	Immediately after birth		1		1
	Within hours		1.05(0.63, 1.76)		1.96(0 .68, 5.65)
	Within days		0.57(0.19, 1.71)		2.02(0 .66, 6.15)
Birth weight	Low		1		1
	Normal		4.40(0.67,28.74)		2.91(0.49, 17.20)
	Macrosomia		2.72(0.23, 31.39)		2.28(0.21, 24.30)
	Didn’t weighted		2.64(0.39, 17.77)		1.86(0.30, 11.49)
	Don’t know		2.05(0.30, 13.95)		1.67(0 .271, 10.360)
Marital status	Not in union		1		1
	Currently in union		2.36(0.93, 6.04)		1.98(0.75, 5.21)
Maternal education	No formal education		1		1
	Primary		1.18(0.64, 2.17)		1.50(0.84, 2.70)
	> = Secondary		0.50(0.15,1.66)		0.90(0.30, 2.69)
Maternal occupation	Currently not working		1		1
	Currently Working		0.58(0.32,1.02)		0.65(0 .37, 1.12)
Number of household members	<5		1		1
	5–6		2.20(1.11,4.36)		0.84(0 .36, 1.97)
	7–8		0.61(0.26,1.42)		1.27(0.623, 2.60)
	> = 9		1.16(0.427,3.16)		0.48(0.23, 0.96)
Number of under five children	0–1		1		1
	2		1.052(0.49,2.22)		1.65(0.72, 3.82)
	> = 3		0.876(0.36,2.09)		1.52(0 .86, 2.69)
Wealth index	Poorest		1		1
	Poorer		1.99(1.01, 3.93)*		1.72(0.90, 3.31)
	Middle		1.33(0.58, 3.05)		1.37(0 .61, 3.07)
	Richer		0 .90(0.34,2.36)		0.94(0.37, 2.39)
	Richest		15.79(4.2, 59.39)***		10.34(3.14, 34.03)***
Parity	Nuliparous		1		1
	Multiparous		1.61(1.11, 2.33)*		2.21(0.95, 5.11)
ANC visit	No		1		1
	Yes		2.75(1.53, 4.94)**		2.25(1.32, 3.82)**
Caesarian delivery	No		1		1
	Yes		0.53(0.14, 2.04)		0.47(0 .128, 1.73)
PNC visit in 2 months	No		1		1
	Yes		0.10(0.03, 0.30) *		0.15(0 .054, 1 .45)
Residence	Urban			1	1
	Rural			0.85(0.36, 2.027)	3.414(0 .82, 14.109)
Contextual region	Agrarian			1	1
	Pastoralist			0.38(0.17, 0.85)*	0.30(0 .10, 0.86)*
	City based			0.99(0.32 3.00)	0.56(0 .12, 2.52)
Community ANC utilization	Low			1	1
	Moderate			0.97(0.51, 1.87)	1.07(0 .56, 2.02)
	High			1.09(0.46, 2.56)	1.30(0 .62, 2.71)
Community PNC utilization	Low			1	1
	High			1.42(0.34, 5.97)	2.77(1.26, 6.58)*
Community poverty level	High			1	1
	Moderate			0.79(0.39, 1.60)	0.97(0.33, 2.84)
	Low			0.83 (0.42,1.62)	0.99(0.31, 3.10)
Community media utilization	Low			1	1
	Moderate			0.83(0.45, 1.53)	0.86(0 .351, 2.13)
	High			0.60(0.26, 1.42)	0.84(0.23, 3.02)
Community women education	Low			1
	Moderate			1.16(0.63, 2.14)	1.13(0 .454, 2.83)
	High			0.52 (0.22,1.21)	0.53(0.13, 2.06)
Community employment status	Low			1	1
	High			1.536(0.91, 2.59)	2.81(1.207, 6.47)*

Model one: empty model, CI: Confidence Interval, AOR: Adjusted Odds Ratio, 1: Reference category *p<0.05 **p<0.01 ***p<0.001.

The age of the infant was negatively associated with exclusive breastfeeding. Those infants whose age group was between 2–3 months were 60.9% [AOR = 0.39, 95% CI: (0.23, 0.65)] and 4–5 months were 95.8% [AOR = 0.04, 95% CI: (0.02, 0.07)] less likely to exclusive breastfeed as compared to infants whose age was up to one month.

The sex of the infant was significantly associated with exclusive breastfeeding. Female infants were 2.5 [AOR = 2.51, 95% CI: (1.61, 3.91)] times more likely to exclusively breastfeed as compared to male infants.

Infant comorbidities were negatively associated with exclusive breastfeeding. Those infants who had comorbidities were 66% [AOR = 0.34, 95%CI: (0.21, 0.57)] less likely to exclusively breastfeed as compared to infants who had no comorbidities.

Household wealth index was positively associated with exclusive breastfeeding. Those infants from the richest family wealth index were 10 times [AOR = 10.34, 95% CI: (3.14, 34.03)] more likely exclusive breastfeeding as compared to those infants from the poorest family wealth index.

The maternal antenatal visit was also positively associated with exclusive breastfeeding. Those infants whose mothers had ANC visits were 2.25 times [AOR = 2.25, 95%CI: (1.323, 3.82)] more likely to exclusively breastfeed compared to those infants whose mothers had no ANC visits.

The contextual region was significantly associated with exclusive breastfeeding practice. Consequently, infants who live in pastoralist regions were 70% less likely to exclusively breastfeeding [AOR = 0.30, 95% CI: (0.104, 0.86)] as compared to those infants who live in the agrarian regions.

Community level postnatal care utilization was significantly associated with EBF. As a result, those infants who live in community who had high level of PNC utilization were 2.8 [AOR = 2.77, 95% CI: 1.26, 6.58)] times more likely to exclusively breastfeed compared to infants who live in the community who had a low level of PNC service utilization.

Community level employment status was positively associated with exclusive breastfeeding practice. Thus, infants who live in a community with a high level of employment were 2.8 [AOR = 2.81, 95% CI: (1.20, 6.47)] times more likely to exclusively breastfed compared to those infants who live in a community with a low level of employment.

Random effects (Measures of variation)

This study also aimed to show if the characteristics of the clusters where under six infants resided would affect exclusive breastfeeding practice.A Model with the lowest DIC, the lowest AIC, and the highest log- likelihood ratio was selected which was model four that better explain exclusive breastfeeding practice. At the empty model, the ICC was 0.141 (ICC>5%) and the variance was 0.54 with 95% CI: 0.25–0.75) at a p-value of <0.001, which indicates about 14.1% of the variation in exclusive breastfeeding was linked to community-level factors and there was a significant variation in exclusive breastfeeding practice across the communities (clusters).

The full model, after adjusting for individual and community level factors, showed that the variation in exclusive breastfeeding across communities remained statistically significant (ICC = 8.77% and variance = 0.32).

In this study, the full model showed up with higher PCV (46.8%); that is, 46.8% of variation of exclusive breastfeeding was explained by the combined factors at the individual and community levels. The median odds ratio of the null model (MOR = 2.01) and the final model (MOR = 1.71) were significant. The MOR at a cluster where a high proportion of nonexclusive breastfeeding was 2.01 times higher compared to a cluster with a low proportion of non- exclusive breastfeeding at the null model, whereas the MOR at a cluster where a high proportion of non-exclusive breastfeeding was 1.71 times higher compared to a cluster with a low proportion of non-exclusive breastfeeding at the final model. This indicates that 0.3 (30%) of the heterogeneity was explained by both individual and community level factors, but still there are a residual effect not explained by individual and community level variables at the final full model (MOR = 1.71) (Table 6).

Table 6

Measure of variations and model fitness.

Characters		Model 1	Model 2	Model 3	Model 4
Random effect	Variance	0.54	0.45	0.31	0.32
Measure of variation	ICC	14.08%	12.13%	8.64%	8.77%
	PCV	Reference	20.77%	42.3%	46.8%*
	MOR	2.01	1.83	1.69	1.71*
Model diagnostics	Log-likelihood	-750.82	-593.21	-745.80	-571.89*
	DIC	1501.64	1186.42	1491.6	1183.78*
	AIC	1505.64	1248.41	1511.60	1235.77*

*Model four is selected based on fitted statistics of log-likelihood ratio.

Key: Model 1 (null model) = without independent variables, Model 2 = only individual- level variables, Model 3 = only community- level variables, Model 4 = both individual and community level variables, PCV = Proportional Change in Variance, ICC = Intra Class Correlation, MOR = Median Odds Ratio, AIC = Akaike Information Criteria, DIC = Deviance Information Criteria.