Some Regional Multi-Level Models for Wanted Fertility in India

Laxmi Kant Dwivedi, International Institute for Population Sciences (IIPS)
Faujdar Ram, International Institute for Population Sciences (IIPS)

The consistent increase in population is known problem specifically in developing countries like India. To understand the dynamics of increasing fertility that is an important component towards the population problem, understanding determinants of wanted fertility might be essential that is an important component of population changes. With this objective in mind, the present study deals with analysis of data related to factors associated with intention to have or not to have the last child/ current pregnancy to the currently married women (15-49 years). For this, the data related to most populous Indian state, Uttar Pradesh, collected under second round of the National Family Health Survey (NFHS), 1998-99, have been utilized. The outcome, that is, dependent variable, considered under the present study is the woman’s intention to produce the last child or to carry the current pregnancy. Based on a series of exploratory analysis, to have meaningful results through sufficient frequency distribution in stratified categories, the covariates included in the data analysis are: Individual level- Woman’s place of residence (rural/urban), religion with caste (schedule caste/schedule tribe Hindus/other hindu /non hindu), education (illiterate/literate), Husband’s education (illiterate/literate), Woman’s occupation (not working / working), Age (less than 25/greare than or equalto25), Child loss (none/one and more), Number of living sons (less than three/three and more), Discussion about family planning with partner (no/yes), Exposure to mass media (yes/no). Primary Samplin Unit (PSU) level- Road within or outside the village/ town, Health facility within or outside the village/ town and primary school within or outside the village/ town. In view of dichotomous dependent variable (No/Yes), the choice of traditional logistic regression analysis is obvious. Under traditional analysis we carry out individual level analysis considering disaggregation of higher level variables at individual level or aggregation of individual level variables at higher level. Under this process either we distort the important assumption of statistical independence that is essential for traditional methodology or give away the individual level variance that may be even up to 90% of the total variance. Therefore, on account of hierarchical structure present in the data, that is, women nested with respective PSU, traditional logistic regression analysis considering PSU level variables at individual level may not be valid. This may obviously result into underestimation of stand error of the estimates providing significant association of some covariates which may not be true in reality. Therefore, comparatively a new procedure known as Multi-level analysis has been used in the data analysis. For this, individual level variables were considered at level-I and PSU level variables at level-II. Accordingly two-level logistic regression analysis was considered in the analysis. Since this procedure considers the variables at their own level, this helps in retaining the obvious hierarchical structure of the data in the analysis. Accordingly, this procedure is expected to provide more reliable results. Further, this procedure also accounts the variability because of variables, which could not either be considered in the data analysis or could not be collected. Further, this procedure also helps in partitioning the variance of a covariate at a particular level at different levels. The residual analysis under this procedure further helps in working out the priority of intervention programme giving important clues to the public health policy planners. On account of these facts, this procedure has been used not only to analyse the data at state level but also at regional level. For this, sub-group analyses in each of the five regions of Uttar-Pradesh were carried out. The importance of regional analysis is very well recognized in the existing literature that may help in planning of regional interventions that may vary from one region to another region. On account of availability of large-scale quality data available under NFHS and also availability of computational facility through specific statistical packages like MLwin, it was possible to carry out the required analysis under the present study. The results in the present study indicate in general that the contribution of a particular covariate is varying from one region to other region. On account of observed variance of unconsidered covariates at level-II, this is also evident that there is need to consider either some more appropriate variables at level 2 or some more variables at higher level like at district level. This may be possible even through consideration of aggregation of some lower level variables at higher levels. The results obtained under the present analysis will be presented and discussed in detail.