Multilevel multinomial logistic regression model for identifying factors associated with anemia in children 6–59 months in northeastern states of India

Figures & data

Table 1. Prevalence of anemia by state of residence

State	Anemia level
	Severely anemic	Moderately anemic	Mildly anemic	Non-anemic	Total (% anemic)
Sikkim	159	301	42	206	708
	22.46%	42.51%	5.93%	29.10%	71%
Arunachal Pradesh	317	724	261	1,611	2,913
	10.88%	24.85%	8.96%	55.30%	45%
Nagaland	33	180	35	277	525
	6.29%	34.29%	6.67%	52.76%	47%
Manipur	241	526	118	1,198	2,083
	11.57%	25.25%	5.66%	57.51%	42%
Mizoram	51	258	57	296	662
	7.70%	38.97%	8.61%	44.71%	55%
Tripura	67	181	20	93	361
	18.56%	50.14%	5.54%	25.76%	74%
Meghalaya	22	55	17	105	199
	11.06%	27.64%	8.54%	52.76%	47%
Assam	334	1,189	136	1,026	2,685
	12.44%	44.28%	5.07%	38.21%	62%
Total (crude %)	1,224	3,414	686	4,812	10,136
	(12%)	(34%)	(6%)	(48%)

Overall Pearson’s chi-square statistics value is 622.84 with a p-value of <0.0001. We found statistically significant association between state of residence and anemia level.

Table 2. Frequency distribution of anemia level by the predictors of anemia. The first column shows chi-square test for independence; the cells represent the frequency (first row), row percentage (second row). The last column shows crude odds ratio for overall anemia

Variables (Chi-square and p-value)		Severely anemic	Moderately anemic	Mildly anemic	Non-anemic	Crude odds ratio
Place of residenceChi-square = 27.16p < 0.0001	Rural	1,035	2,705	516	3,818	1.04
		12.82%	33.50%	6.39%	47.29%
	Urban (ref)	189	709	170	994
		9.17%	34.38%	8.24%	48.21%
ReligionChi-square = 166.35p < 0.0001	Hindu	537	1,330	238	1,667	1.23^*
		14.24%	35.26%	6.31%	44.19%
	Muslim	83	442	48	350	1.59^*
		8.99%	47.89%	5.20%	37.92%
	Christian	352	1,050	245	1,824	0.88
		10.14%	30.25%	7.06%	52.55%
	Others (ref)	252	592	155	971
		12.79%	30.05%	7.87%	49.29%
Household living standardChi-square = 8.33p = 0.2146	Low (ref)	764	2,031	400	2,956
		12.42%	33.02%	6.50%	48.06%
	Medium	336	999	199	1,312	1.01
		11.81%	35.10%	6.99%	46.10%
	High	124	384	87	544	1.08
		10.89%	33.71%	7.64%	47.76%
Sex of childChi-square = 2.08p = 0.5548	Male	636	1,772	368	2,453	1.08
		12.16%	33.89%	7.04%	46.91%
	Female (ref)	588	1,642	318	2,359
		11.98%	33.46%	6.48%	48.07%
Literacy of MotherChi-square = 6.43p = 0.923	Can read & write (ref)	692	2,024	385	2,771
		11.78%	34.47%	6.56%	47.19%
	Can’t read & write	497	1,257	276	1,867	0.97
		12.75%	32.26%	7.08%	47.91%
Total children ever born to motherChi-square = 9.72p = 0.1366	Up to two children)	548	1,444	276	2,014	0.91
		12.80%	33.72%	6.45%	47.03%
	Three or four children	423	1,214	271	1,805	0.97
		11.39%	32.70%	7.30%	48.61%
	Five or above children (ref)	218	623	114	819
		12.29%	35.12%	6.43%	46.17%
Age at marriageChi-square = 13.03p = 0.0425	Below 18 Years	384	1,016	183	1,556	1.03
		12.23%	32.37%	5.83%	49.57%
	18 to 26 years	748	2,090	445	2,846	0.91
		12.20%	34.10%	7.26%	46.43%
	Above 26 Years (ref)	57	175	33	236
		11.38%	34.93%	6.59%	47.11%
Age of child in monthsChi-square = 5.07p = 0.1669	>48 months	499	1,477	313	2,034	1.03
		11.54%	34.17%	7.24%	47.05%
	<48 months	725	1,937	373	2,778
		12.47%	33.32%	6.42%	47.79%

^* Significant results based on 95% confidence interval for the odds ratio.

Table 3. Model-building strategies

Model 1	Model 2	Model 3	Model 4	Model 5
No predictors but only random effects for the “state”	Model 1 + level-1 fixed effects that are related to maternal characteristics such as number of children ever born (2 or less, 3–4, and 5 or more) and age at marriage (below 18, 18–26, and above 26 years).	Model 2 + age of the child (months)	Model 3 + level-1 fixed effects religion and literacy of mother	Model 4 + sex of child (male/female), household living standard (low, medium, and high), and place of residence (urban/rural)
Results used to determine the percentage variation in anemia level explained by the level-2 units (state of residence)	Results indicate the relationship between level-1 (household level) predictors and the anemia level	Results indicate if addition of child’s age improves the fit.	Results indicate whether addition of demographic variables improves the model fit	Results indicate whether addition of SES variables improves the model fit

Table 4. Estimates for two-level multinomial logistic regression models for predicting anemia level (N = 10,136)

	Model 1	Model 2	Model 3	Model 4	Model 5
Fixed effects
Intercept 1 (severely anemic)	−1.95^* (0.16)	−2.29^* (0.19)	−2.35^* (0.19)	−2.42^* (0.20)	−2.23^* (0.23)
Intercept 2 (moderately anemic)	−0.07 (0.16)	−0.42 (0.19)	−0.48^* (0.19)	−0.55 (0.20)	−0.37 (0.23)
Intercept 3 (mildly anemic)	0.21 (0.16)	−0.14 (0.19)	−0.19 (0.19)	−0.27 (0.20)	−0.08 (0.23)
Age at marriage		0.12^* (0.04)	0.12^* (0.04)	0.12^* (0.04)	0.12^* (0.04)
Total children ever born		0.08^* (0.03)	0.09^* (0.03)	0.08^* (0.03)	0.08^* (0.03)
Age of child (<48 months)			−0.08^* (0.04)	0.08^* (0.04)	0.08^* (0.04)
Religion				0.03 (0.02)	0.02 (0.02)
Literacy of mother				0.02 (0.04)	0.003 (0.04)
Household living standard					−0.02 (0.03)
Place of residence					−0.06 (0.05)
Sex of child					−0.03 (0.04)
Error variance
Intercept	0.2015^* (0.10)
Model fit
−2LL	23,063.22	22,228.13^** (p < 0.0001)	22,224.12^** (p = 0.0452)	22,222.03 (p = 0.3517)	22,218.83 (p = 0.3618)

Entries in the table are estimated effects while the standard errors are reported in the parenthesis.

^* Level of significance at p < 0.05.

^** Significant LR test; ICC = 0.0577. PROC GLIMMIX in SAS 9.4 with Laplace estimation method was used.