TABLE 6.

Results of Multilevel Analysis

Model 1Model 2Model 3Model 4Model 5Model 6
Empty ModelIndividual-Level VariablesCommunity-Level VariablesIndividual and Community-Level VariablesIndividual and Community-Level VariablesIndividual and Community-Level Variables
Fixed partsfixed slopefixed sloperandom slope
Predictor, OR (95% CI)
Intercept0.13 (0.08, 0.22)0.48 (0.18, 1.28)0.19 (0.09, 0.41)0.47 (0.18, 1.25)0.60 (0.20, 1.77)0.36 (0.17, 1.23)
Time0.81 (0.73, 0.91)0.81 (0.73, 0.91)0.84 (0.75, 0.94)0.67 (0.55, 0.81)0.67 (0.55, 0.82)0.67 (0.5, 0.81)
Study, improved latrine0.48 (0.27, 0.83)0.60 (0.33, 1.07)0.60 (0.33, 1.08)0.40 (0.15, 1.13)
Improved water0.88 (0.55, 1.41)0.87 (0.54, 1.40)0.95 (0.58, 1.56)0.85 (0.52, 1.38)
Handwashing1.03 (0.63, 1.69)1.05 (0.64, 1.72)1.11 (0.64, 1.91)1.06 (0.63, 1.75)
Child’s sex1.41 (0.89, 2.25)1.44 (0.90, 2.29)1.45 (0.91, 2.31)1.47 (0.91, 2.36)
Child’s age0.97 (0.96, 0.99)0.97 (0.96, 0.99)0.97 (0.96, 0.99)0.97 (0.96, 0.99)
Coverage of study-improved latrines0.43 (0.19, 0.98)0.43 (0.19, 0.97)0.43 (0.19, 0.98)0.38 (0.16, 0.94)
Coverage of improved water-0.67 (0.31, 1.42)--
Coverage of handwashing-0.90 (0.46, 1.75)--
Random parts
 Cluster-level variance0.86 (0.15)0.82 (0.14)0.79 (0.14)0.79 (0.14)0.80 (0.14)0.80 (0.14)
 Individual-level variance1.16 (0.16)1.06 (0.16)1.18 (0.16)1.07 (0.16)1.07 (0.16)1.12 (0.17)
 ICC-VPC0.18 (ICC)0.17 (ICC)0.16 (VPC)0.43 (VPC)0.43 (VPC)0.42 (VPC)
 Explained variationRef (cluster)4.7%8.1%9.3%7.0%7.0%
 (i.e., PCV in %) proportional  change in variance by the  new modelRef (individual)8.6%-8.6%7.8%3.8%
Deviancea1667.41644.61657.51639.11640.31635.1
Model fit test results,b chi-square (P-value)-22.81 (P<.001)-5.52 (P=.14)4.28 (P=.04)5.22 (P=.02)
MOR2.262.182.122.122.172.14
IOR upper, lower6.42, 0.376.42, 0.376.54, 0.366.22, 0.34
  • Abbreviations: ICC, intra-cluster correlation; IOR, interval odds ratio; MOR, median odds ratio; PCV, proportional change in variance; VPC, variance partition coefficient.

  • a The deviance: –2 × LN (likelihood), where likelihood is the value of the likelihood function at convergence, and LN is the natural logarithm.

  • b The likelihood ratio test (Model 4 of lower deviance was compared with Model 2 of larger value, which was not significantly different (P=.14). Model 5 of lower deviance was compared with Model 2 of larger deviance, which was significantly different (P=.04). Model 6 has the lowest value of deviance was compared with Model 5, which was significantly different (P=.02). Hence, we finally selected Model 6.