Page 244 - Special Topic Session (STS) - Volume 2

STS489 Chibuzor C. N. et al.
Response variable
The main response variable is whether respondent’s daughter(s)
underwent FGM/C? This was coded as a binary variable where a value of 1
denotes that daughter was cut and 0 denotes that daughter was not cut.

Exposure variables
Covariates included in the community-level spatial model were indicators
of social norms with the following surrogate variables: (1) mother’s FGM/C
status and (2) mother’s support for FGM/C continuation. Individual-level
covariates comprised the girl and her mother’s background characteristics,
their geographical location -region and state of residence, type of place of
residence (urban vs rural), socio-demographic variables such as age of mother,
ethnicity, wealth index, marital status, employment status, and level of
education.

Statistical Analysis
Bayesian geo-additive logistic regression model
Let denote a realisation from the random variable = 1, … , . For our

purpose, we define
1,
= { 0, .

(1)
Then, , 0-14-year-old Nigerian girl FGM/C status, is Bernoulli random

variable with parameters, ir, that is, ∼ ( ). We used a class of mixed

models called structural additive regression (STAR) models [5, 7, 10- 13], to
estimate the effects of different covariates on the observed data. Unlike the
standard regression model, which assumes strictly linear relationship between
the covariates and the response variable, STAR models allow us to
simultaneously control for both linear and non-linear, continuous and
categorical covariates in a coherence regression framework, such that the link
function ,

) =
= (
1 −
′
)
= 0 + + 1 ( 1 + ⋯ + ( ) + ( + ( + (s, t) (3)
)
),

for = 1, … , , = 1, … , . where (.),…, (.) are the functions (may be

1

smooth) of non-linear continuous covariates, such as age, time effects, etc.

is the intercept, = ( , … , )′ are unknown coefficients of other class of
0
1

covariates, . Also, is the geographically referenced location of girl ,

(. ) and (. ) denote the structured (correlated) and the unstructured

233 | I S I W S C 2 0 1 9