Associations between women’s empowerment and child development, growth, and nurturing care practices in sub-Saharan Africa: A cross-sectional analysis of demographic and health survey data

Background Approximately 40% of children 3 to 4 years of age in low- and middle-income countries have suboptimal development and growth. Women’s empowerment may help provide inputs of nurturing care for early development and growth by building caregiver capacity and family support. We examined the associations between women’s empowerment and child development, growth, early learning, and nutrition in sub-Saharan Africa (SSA). Methods and findings We pooled data on married women (15 to 49 years) and their children (36 to 59 months) from Demographic and Health Surveys that collected data on child development (2011 to 2018) in 9 SSA countries (N = 21,434): Benin, Burundi, Cameroon, Chad, Congo, Rwanda, Senegal, Togo, and Uganda. We constructed a women’s empowerment score using factor analysis and assigned women to country-specific quintile categories. The child outcomes included cognitive, socioemotional, literacy–numeracy, and physical development (Early Childhood Development Index), linear growth (height-for-age Z-score (HAZ) and stunting (HAZ <−2). Early learning outcomes were number of parental stimulation activities (range 0 to 6) and learning resources (range 0 to 4). The nutrition outcome was child dietary diversity score (DDS, range 0 to 7). We assessed the relationship between women’s empowerment and child development, growth, early learning, and nutrition using multivariate generalized linear models. On average, households in our sample were large (8.5 ± 5.7 members) and primarily living in rural areas (71%). Women were 31 ± 6.6 years on average, 54% had no education, and 31% had completed primary education. Children were 47 ± 7 months old and 49% were female. About 23% of children had suboptimal cognitive development, 31% had suboptimal socioemotional development, and 90% had suboptimal literacy–numeracy development. Only 9% of children had suboptimal physical development, but 35% were stunted. Approximately 14% of mothers and 3% of fathers provided ≥4 stimulation activities. Relative to the lowest quintile category, children of women in the highest empowerment quintile category were less likely to have suboptimal cognitive development (relative risk (RR) 0.89; 95% confidence interval (CI) 0.80, 0.99), had higher HAZ (mean difference (MD) 0.09; 95% CI 0.02, 0.16), lower risk of stunting (RR 0.93; 95% CI 0.87, 1.00), higher DDS (MD 0.17; 95% CI 0.06, 0.29), had 0.07 (95% CI 0.01, 0.13) additional learning resources, and received 0.16 (95% CI 0.06, 0.25) additional stimulation activities from their mothers and 0.23 (95% CI 0.17 to 0.29) additional activities from their fathers. We found no evidence that women’s empowerment was associated with socioemotional, literacy–numeracy, or physical development. Study limitations include the possibility of reverse causality and suboptimal assessments of the outcomes and exposure. Conclusions Women’s empowerment was positively associated with early child cognitive development, child growth, early learning, and nutrition outcomes in SSA. Efforts to improve child development and growth should consider women’s empowerment as a potential strategy.

This paper reports an investigation of empowerment and child development using a secondary data analysis.
I was asked for a statistical report and I interpret that to include all aspects of the design and conduct of the study.

Points of detail
Page 10 Naming factors is always a tricky issue but given the content of the items which go to make up the authors' gender dimension why was it not named 'intimate partner violence' ? I am not an expert in this area but is there not more to gender roles than physical assault? Pages 22 and 23 seem to agree. Obviously if that is all DSS asks about then nothing can be done by the authors to broaden the scope of the factor analysis.
Pages 10 and 11 It is not clear to me why, if the authors had a theoretical model, they did not proceed straight to CFA to test it.
Page 13 There are only four quintiles. if 'fifth' lacks academic credentials then 'quintile category' might be better.
Page 13 According to Appendix S2 Table 2 there are only four binary indicators here for Gender so only 16 = 2 4 possible values. I do not see how it makes sense to categorise the factor. Even if it does then categorising a variable wastes information (Altman and  and leads to models which are often implausible as they predict the effect remaining flat within categories and then jumping to a new value at the category boundary. Decision making only has 32 possible values which is better but not ideal.
Page 13 Typo, link not like, I suppose.
Page 14 This tells us about missing data on confounders but what about missing data on the empowerment items?
Page 15 There does not seem to have been a formal test of the relationship between empowerment and any of the outcomes presented in Table  3. Just comparing two extreme groups is sub-optimal. This occurs elsewhere.
Page 15 Even if there is a relationship it only seems to be monotonic for the anthropometric outcomes not the other two which is worth comment. Similarly in Figure 2 for the sub-domains of empowerment many of these are not monotonic. It may be that it was expected that high and low decision making empowerment would be best and middle worst, or vice versa, but has this been explained anywhere?
Page 15 It is better to state that the authors found no evidence for association rather than asserting that there was no association.
Page 17 Here we suddenly lose effect sizes with their confidence intervals and just see p-values which is a shame.
Page 21 In the absence of an overall test of the effect of empowerment on child growth I do not think it is wise to claim an effect here.
Page 26 'In conclusion, we showed that women's empowerment was positively associated with child cognitive development in SSA.' That is not the conclusion which I draw from Table 3 where the confidence intervals mostly include the null or from the parallel table in S8 Appendix where there is even less evidence.
S3 Appendix There seem to have been a number of places where the solutions differ between countries. This is not completely unexpected as the numbers per country cannot be large although I cannot find them in the submission. What are the implications of these differences for the underlying science here? Does this not throw some doubt on the existence of the constructs or their generalisability across cultures?
S3 Appendix, Table The rather high correlation between Factor 1 and 2 seems undesirable.
Points of more substance

Choice of latent variable method
Given that most of the manifest variables are binary and those which are not are ordered categorical I wonder why the authors did not use item response methods. This might have enabled inclusion of the items which were not asked in Congo as IRT methods cope with situations where not everybody answers all items. The use of differential item functioning might have meant that the scale could have been adapted to fit well across all the countries.

Descriptives
We are rather short on distributional information about the overall scores on the domains. In a study in which I was involved (not in one of the countries here) we had a very simple list of empowerment items and found the sum score of positive responses was bi-modal: they either were empowered or they were not.

Summary
This is an extremely detailed report of the study but I have indicated some gaps in the analysis which make me feel that, as it stands, the authors' conclusions and interpretations go beyond what they have found in the analysis.
Michael Dewey