Dear Ann Marie:
Thanks for your interest in our work. I think I did understand well your query. You textually stated that "graduate income in Spain can vary significantly with degree". I answered to your sentence by showing the data on depression by degree. There was no difference in depression incidence according to years of educational attainment in our cohort.

The reference that you provided to support differences in salaries in the Spanish general population (http://www.prospects.ac.u...) is not any scientific peer-review source; it is just a very short statement in a general information source. It cannot be specifically applied to graduates of our University. Our University is ranked according to a set of different standards of quality as the first private university in the country, please check:
http://www.unav.es/inform....

The key issue here is if there is any substantial amount of residual confounding by income after applying two classical epidemiological methods for the control of confounding, a) restriction to university graduates; b) multivariable adjustments for sex, age, smoking, physical activity during leisure time, total energy intake, baseline body mass index (BMI), and adherence to the Mediterranean dietary pattern.
I acknowledge that I do not have evidence on the amount of variability in income within the tiny strata that are constructed by breaking down the whole cohort according to the several categories of all these variables considered together. However, it is very difficult to figure out that a wide variability in income might exist within them. In addition, it is very difficult to conceive that a substantial variability in depression risk might exist according to the potential residual variability in income within these tiny strata. Furthermore, there is no evidence to think that a sufficient amount of heterogeneity in trans fat intake according to income might exist within these tiny strata.
These explanations might account for the fact that income is seldom used in nutritional epidemiology to adjust for. The multivariable models commonly used in nutritional epidemiology adjust only for education but not for income.
An additional issue is that in contrast to US or UK, a direct question about income in a questionnaire is usually not regarded as reliable in the Spanish culture because of social desirability bias and other issues. Therefore, when we designed our cohort study we took into account that once education was accounted for, income was not included as a relevant confounder in the best known cohorts of nutritional epidemiology. This is the reason why we do not have evidence about income variability.

To support that residual confounding by income is not a likely explanation of our findings I can add that I am not aware of:
a) any longitudinal study in Spain showing that income is a stronger risk factor for depression than education. The paper that you quote in the "Journal of Happiness Studies" only uses cross-sectional data and reverse causality cannot be excluded (i.e. depressed people at baseline are more likely to exhibit a worse job performance and reduce their income as a consequence of depression).
b) any longitudinal study showing that baseline income is a better predictor of trans fat intake than education.
c) any epidemiological cohort study in Spain assessing the relationship between dietary habits and chronic disease that has adjusted for income instead of adjusting for education (please see, for example, Buckland et al. Am J Epidemiol 2009;170:1518-29 or Arriola et al. Heart 2010;96:124–130 that are cohorts with a wider variability in income and educational attainment than the SUN cohort).

Another cohort study has longitudinally assessed the relationship between dietary patterns and depressive symptoms and has adjusted for employment grade and education, the paper by Akbaraly (Br J Psychiatry 2009;195:408–413.). As you probably are aware of, the British civil service employment grade is defined on the basis of salary, social status and level of responsibility and consists of three levels, with grade 1 representing the highest level and grade 3 the lowest. Therefore they considered the salaries. The results of this British cohort are very consistent with our findings.

All this is reassuring that residual confounding by income is not a likely explanation of our findings.

Best regards,
miguel