The title of this blog sounds like a really boring question. But it’s not.
We have known for a long time that household food expenditures are associated with child growth outcomes, but how is that association parcelled out: is it quantity or quality or both? GAIN’s Executive Director, Lawrence Haddad, asks.
This is the question picked up in a new paper by the Young Lives research team, using data from Ethiopia, India, Peru and Vietnam in the “younger” cohort: pooling 3 rounds (when the kids are 5 years, 8 years and 12 years of age).
To answer the question the team looked for associations between height for age z score and 3 key variables: (a) total household food expenditures per adult equivalent, (b) patterns of household food expenditure (as determined by factor analysis) and (c) child dietary diversity score (number of food groups). Control variable applied included child age and sex and household location (rural/urban).
So why is this question important? First, we know poor diet quality is the number one risk factor behind the global burden of disease. Second, diet quality is really difficult to assess: 24 hour recall assessment is costly and often criticised for relying too much on such a short recall period. Third, there is a LOT of household food expenditure data out there. The World Bank, in partnership with National Governments collects it routinely in its Living Standards Measurement Surveys (LSMS) and there are plenty of other such surveys out there, typically collected every 3-5 years per country. Finally, these data are not so difficult to collect: they usually cover 40-60 food types and use a 2 week recall period (covering food from purchase, gifts, in kind payment, own production and outside the home).
If only this commonly collected type of food expenditure data could be used as a credible proxy for the quality of diet. If an index of food group expenditure diversity could be constructed that is significantly associated with standardised height for age (HAZ), and holds for a number of countries, that would be very useful.
So using ordinary linear regression methods, the authors regress HAZ on total food expenditure per adult equivalent, the household food group expenditure index (HFGEI, derived from a factor analysis of household food expenditure patterns) and child diet diversity scores.
In Ethiopia, India and Vietnam the authors find that HFGEI is significantly and positively associated with HAZ–even after accounting for total food expenditure per adult equivalent and child diet diversity score. In Ethiopia and India the estimated coefficients on HAZ for a 1 SD change are larger for HFGEI than for child diet diversity. And in Ethiopia and India the estimated coefficient on household food expenditure per adult equivalent becomes insignificant once HFGEI and child diet diversity are introduced.
The authors also do this same exercise for BMI for age (BMI-Z) and find significant (positive) results, but only for Ethiopia and Vietnam.
There are things I would like to see improved, namely some instrumental variables treatment on the 3 key right hand side variables (e.g. is there some unobserved factor driving HAZ and these 3 variables inducing a fake association?). I’d also like to see village dummy variables included to take health environment into account. Parental education variables are also missing. Finally, factor analysis always makes me a bit nervous given its black box propensity, but then again it is no less black box than regression analysis and the authors do a good job of unpacking their factor analysis results.
Nevertheless, this is a careful study which suggests that commonly available household food expenditure data may provide a good proxy for the quality of household food consumption (and perhaps even child food consumption).
In the absence of lots of actual food intake, these food expenditure data sets are likely to prove very valuable to anyone trying to assess their efforts to improve nutrition outcomes through increasing the consumption of nutritious foods.
Published 20 April 2017