2.2.4. Model Building

The present study fitted and examined the following models:

Model 1: *logit pij* = *<sup>W</sup>ijβ* Model 2: *logit pij* = *<sup>W</sup>ijβ* + *f*1(*current agei*) + *f*2(*age at first cohabitationi*) Model 3: *logit pij* = *<sup>W</sup>ijβ* + *fspatj districtj* , *j* = 1, 2, . . . , 30 Model 4: *logit pij* = *<sup>W</sup>ijβ* + *f*1(*current agei*) + *f*2(*age at first cohabitationi*) + *fj districtj* , *j* = 1,2,. . . ,30

Model 1 is classical logistic regression, where all categorical variables (women's education, women's region affiliation, women's working status, women's province of residence, have heard about family planning on radio in last 12 months, have heard about family planning in newspapers/magazines in last 12 months, visited by a family planning worker in the last 12 months, visited a health facility in the last 12 months, currently residing with husband or not, number of living children, wealth quintile of the household, husband desires children) and (current age of the woman and woman's age at first cohabitation) were considered as fixed effects and assumed to have a linear effect on the outcome variable.

In model 2, categorical variables listed earlier in model 1 were assumed to have a linear effect on the response variable, whereas continuous variables were modeled non-parametrically and model 2 is commonly known as an additive logistic regression model. In model 3, all predictor variables were modeled as fixed effects and structured random effects as structured spatial effects that cover the unobserved covariates which are essential within the districts. Model 4 is an extension of model 2, including structured spatial effects and is known as a structured additive regression model.
