Table 1 Measurement of various predictors included in the estimated models

Rank of most recent birth:

We distinguish between mothers whose most recent birth is rank 1 or 2 (34.7%) and those whose most recent birth is of a higher rank.

Education:

Highest level of education attained is divided into four categories: none, primary, secondary, and post-secondary education.

Ethnicity:

We specifically recognized the largest ethnic groups in Nigeria (Hausa, Igbo, Yoruba, Fulani and Kanuri) while all the other ethnic groups are classified together.

Age at last birth:

The questionnaire did not include a direct question on the age at last birth; we computed this indicator by subtracting the child's age from the woman's current age and rounding the result to the nearest whole number.

Attitudes towards family planning:

We measure this indicator through reported approval of family planning.

Ideal family size:

We distinguish between the women who gave a numeric response to the question on ideal family size (52.0%) and those who gave non-numeric responses, such as "Up to God" (48.0%).

Household socio-economic status:

We constructed a scale for household socio-economic status from information on possession of specific household items and utilities, including refrigerator, radio, television, car, video player, cell phone, standby generator, electricity, fan, kerosene stove, pipe-borne water and water closet (Cronbach's alpha for internal reliability: 0.88). The resulting scale was divided into five quintiles.

Urban residence:

This variable was derived from the question on the type of place of residence; we compare urban residents with their rural counterparts.

Media saturation in the LGA of residence:

We operationalize this community-level variable through the mean level of exposure to the radio and the television for the people in the LGA of residence other than the index individual (the non-self mean). We divide the measure into three categories, viz.: low, medium and high levels of community media saturation based on the percentiles.

Prevalence of small family norm in the LGA of residence:

We measured this variable using the non-self mean of expressed preference for a small family (four children or less).

State of residence:

The NARHS 2005 survey took place in the 36 states and the Federal Capital territory. The state of residence was included as a random variable in the estimated models to represent unmeasured factors related to the socio-political and cultural context.

Number of people per PHC in the state of residence:

This information came from the statistics published by the National Bureau of Statistics for the year 2005.