Once the data had been reduced, the final step was to use stepwise multivariate and forward conditional logistic regression analyses to determine which variables predicted on nutritional outcomes. The aim was to determine the size and the direction o f the effect o f several independent variables. A total o f 34 factors had been generated for the main analysis. Given all the variables could not be entered in a single regression analyses, several models were used. The first model grouped all the biological variables, i.e. all the factors that had the most direct impact on the child’s nutritional status, such as diet, immunizations, morbidity and maternal nutritional status. The second model grouped all the variables that related to nutrition and health of the child, including food security, access to health care, maternal reproductive aspects, hygiene and sanitation, and child feeding practices. The third model included socio-economic factors, both material and economic, and parental education. The fourth model included the status of women factors including maternal employment, decision making, freedom of movement, and violence. A final model was tested entering all the variables from the previous four models that significantly predicted the child’s nutritional status, thus overall identifying the variables that best predicted nutritional outcome. In all cases the dependant variables were WAZ, HAZ, and WHZ.
This same process was repeated for the longitudinal data, where the second anthropometric measurement was the outcome variable. But an interaction term for the first measurement multiplied by the age of the child was entered in the regressions. An adjustment term needed to be included as children regress to the mean and nutritional status for this age group is closely linked to age (205).
Three types of multivariate regressions were undertaken. The first tested these models for the whole sample. The second tested these models comparing the lower quartile to the middle quartile (negative deviance), and the middle quartile to the upper quartile (positive deviance). The third tested these models comparing the upper and lower quartiles. The regressions o f the whole sample were also repeated for the data disaggregated by tribal and rural group.
Table 2.5 Components generated by factor analysis Components Number of Factors Generated Number of Independent Variables Included Variables Included
Socio-economic 3 12 Number of rooms
Status - Material Housing type
Aspects Have electricity
Have separate kitchen Have toilet
Own land Own tv/radio Have ration card
Family type (joint/nuclear) Number in house Water source
Socio-economic 3 6 Total household income per capita per day Status - Income minus amount spent on alcohol plus crop
Aspects earnings
Total spent per capita per day Labour security
Family type
Number in household
Food Security 4 10 Woman’s control over food supply Frequency of food purchases Mother herself can buy supplies Have subsistence crops
Number of days in the last week without food
Number of days in the last month without food
Who d oes fieldwork Family type Number in house
Child Feeding 3 8 Breastfed immediately after birth
Practices Prelacteal feeds offered
Colostrum offered Ever exclusively breastfed
Exclusively breastfed upto six months Frequency of feeding
Who feeds child
Age when solids introduced
Maternal 3 10 Where last child w as delivered
Reproductive Any antenatal care
Aspects Maternal age at first pregnancy
Had tubectomy Paid for ANC visits Number of pregnancies Number of live children Birth order of child Birth Interval Child mortality
Maternal Employment
2 10 Where mother works
Number of monthe per year worked Earnings
Number of years worked
Earnings contribution to household income Freedom to spend her income
Who decides whether sh e can work Employed before marriage
Employed before children were bom Currently working
Parental Education 2 4 Maternal Education
Mother can read and write now Paternal Education
Educational difference between husband and wife
Hygiene and Sanitation
2 6 A ccess to safe water
Maternal bathing Child bathing
Offers boiled water to child
W ashes hands with soap after toilet for herself and child
A ccess to Health Care
3 11 Healthcare choice
Time to health care
Total spent on last trip to phc
Whether credit was taken the last time to phc
Opportunity cost of going to phc Where they went the last time child was sick
Who decides about health care Mother needs permission to go to phc She g o es alone to phc
Takes credit without husbands permission Freedom of
Movement
3 10 Family type
G oes alone to fields Goes alone to work G oes alone to shop G oes alone to market Goes alone to natal home Natal home is walkable distances G oes alone to health care G oes herself for food supplies
Domestic Violence 2 11 She can refuse sex
Number of times sh e was hit in the last year Number of years of violence
Psychological abuse Threatens to hit in public Ever hit
Hit in the last year Hit in the last month Hit during last pregnancy Any violence
Decision-Making 4 13 Family type
Ever lived with in-laws Her position in household
Decision on household expenditure Decision on whether she can work Decision on fertility
Decision on health care Decision on credit for food Major family decisions She is excluded
Can spends husbands money without his permission
Marriage decision
Longitudinal Data
Child Feeding Practices
3 8 Breastfed immediately after birth Prelacteal feeds offered Colostrum offered Ever exclusively breastfed
Exclusively breastfed upto six months Frequency of feeding
Who feeds child
Age when solids introduced
Supporting Analyses
D ow ry 2 6 Dowry paid
Numtier of elder sisters Problems from dowry Violence related to dowry
Alcohol Abuse 1 5 Husband drinks
Number of days he drinks Alcohol-related violence Food problems
Amount spent on alcohol per day Husband’s
Employment
2 5 Frequency of pay
Type of work
Number of days per year worked Who controls his income
How much he gives to the household
Logistic regressions using the same procedure as above were also completed for both the cross-sectional and the longitudinal data using z-score cutoffs. The aim of this step was to examine the data more closely and determine which variables became more important at each z-score cutoff. This was also repeated for the data disaggregated by tribal and rural group. A second set o f logistic regressions was undertaken to compare the upper and lower quartiles within the sample. The dependant variables for these analyses were again WAZ, HAZ, and WHZ.
Supporting multivariate and logistic regression analyses were conducted on the factors for decision-making, freedom of movement and violence. Mainly to identify the predictors for the decision-making factors, the freedom o f movement factors, and the violence factors. A forward conditional logistic regression was undertaken to determine the predictors of child morbidity.