Background While an association has been established between low income and poor diet using cross-sectional data, such analysis cannot account for confounding by unobserved characteristics correlated with income and diet, and changes in income and diet cannot be tracked over time. This paper, using longitudinal panel data, explores whether falls in objective and subjective family income predict deterioration in children's diets over time.

Methods This paper uses panel data from the nationally representative birth cohort study Growing Up in Scotland. 3279 families have valid data on all dependent, independent and control variables for both time points. Dietary data were collected using maternal recall at sweeps 2 and 5 when the children were aged 22 and 58 months, respectively. Mothers reported on children's variety of consumption of vegetables, fruit and on the frequency of consumption of crisps, sweets and sugary drinks. The dietary variables were ordinal and were analysed using multivariate fixed effects ordinal logistic regression models.

Results Controlling for time-varying confounders (children's food fussiness, maternal social class, maternal education, family composition, maternal employment) and for family and child time-invariant characteristics, moving from the highest to the lowest income band was linked to a smaller chance of increased fruit variety from 22 to 58 months (OR=0.42, 95% CI 0.21 to 0.82). Mothers who transitioned from ‘living very comfortably’ to ‘finding it very difficult’ to cope on current income had children who consumed fewer fruit varieties over time (OR=0.40, 95% CI 0.19 to 0.85), and who increased their frequency of consumption of crisps (OR=2.03, 95% CI 1.05 to 3.94) and sweets (OR=2.23, 95% CI 1.18 to 4.20).

Conclusions The diets of young children in Scotland deteriorated between the ages of 2 and 5 years across the entire socioeconomic spectrum. Additionally, deterioration in subjective income predicted less healthy diets for children.