Estimated effects on height-for-age z-scores with 95% CIs from three linear regression models
| Model 1 | Model 2 | Model 3 | ||||
| Coefficient | 95% CI | Coefficient | 95% CI | Coefficient | 95% CI | |
| Treatment | 0.296*** | (0.116 to 0.484) | 0.441 | (−0.335 to 1.028) | 0.467 | (−0.344 to 1.050) |
| Maternal grandmother | −0.265** | (−0.528 to 0.021) | −0.259** | (−0.503 to 0.019) | ||
| Maternal grandmother*T | 0.168 | (−0.296 to 0.615) | 0.145 | (−0.324 to 0.575) | ||
| Paternal grandmother | 0.008 | (−0.200 to 0.236) | 0.006 | (−0.192 to 0.238) | ||
| Paternal grandmother×*T | −0.253* | (−0.529 to 0.029) | −0.235* | (−0.493 to 0.039) | ||
| Mothers sisters | 0.031 | (−0.052 to 0.103) | ||||
| Mothers sisters×*T | −0.056 | (−0.163 to 0.057) | ||||
| Fathers sisters | 0.002 | (−0.140 to 0.140) | ||||
| Fathers sisters*T | 0.042 | (−0.110 to 0.188) | ||||
| Mothers brothers | −0.001 | (−0.116 to 0.082) | ||||
| Mothers brothers*T | 0.024 | (−0.109 to 0.170) | ||||
| Fathers brothers | 0.016 | (−0.064 to 0.115) | ||||
| Fathers brothers*T | −0.034 | (−0.148 to 0.075) | ||||
| Total siblings of mother | 0.016 | (−0.032 to 0.062) | ||||
| Total siblings of mother×*T | −0.019 | (−0.069 to 0.037) | ||||
| Total siblings of father | 0.01 | (−0.038 to 0.054) | ||||
| Total siblings of father×*T | 0.001 | (−0.080 to 0.072) | ||||
| R2 | 0.19 | 0.195 | 0.193 | |||
| N | 2017 | 2017 | 2017 | |||
OLS regressions with height-for-age (HAZ) scores as dependent variable. Model 1 estimates the overall effect of exposure to the programme. Models 2 and 3 estimate regressions that allow the programme effect to vary with different extended family members. Inference is conducted using the wild cluster bootstrap t procedure recommended by Cameron et al 37; 95% CIs calculated according to the method recommended by Colin Cameron and Miller.38
All regressions include the following controls: cluster level controls: education and Chewa ethnicity in 2004, household level controls: a wealth index calculated in 2004, mother level controls: whether she had completed primary school, was working as a farmer or was a student in 2004, current age, age2 and logarithmic height. Child level controls: month of measurement, age, age2, gender, number of older siblings, number of older siblings2.
Sample includes all children born after the intervention start date in July 2005 to married main respondent mothers, who were aged 0–53 months at the time of measurement. Column 1 indicates the effect of intervention assignment on HAZ scores, for the sample where family networks information is not missing. Models 2–3 indicate how intervention effects on HAZ scores vary with the presence of different extended family members.
*p<0.10, **p<0.05, ***p<0.01.
HAZ, height-for-age z-scores.