Adjusting for Differential Misclassification

I have a study comparing depression among middle school and high school children. After finishing the data collection for this study, we decided to look and see how the sample we took (n=300) represent in relation to the source population (all children in the school). So we obtained records about the demographic distributions of children in the school and we found that there is selection bias based on the socioeconomic status which is different in the groups (one group is similar to the source population while the other is not). How do we address this bias in the analysis of the data? Do you have ideas of ways we can adjust for it (knowing that low socioeconomic status is a potential confounder to the relationship between school group (middle school/ high school) and depression?

Screen Shot 2019-06-06 at 4.04.34 PM.png



Less is more. Stay pure. Stay poor.
If you just want to assume that the missing people don't differ in other ways than the ones that were sampled and fit the characteristic patterns - you could try to weight data in model to mimic the population characteristics, say like they do in survey data. Otherwise I am sure there is some type of Bayesian approach to deal with this issue.
Thank you hlsmith for your answer. I will try to calculate the weights and add them to the model.
But I am a little confused. If I am adjusting for the low socioeconomic status in the logistic regression model, shouldn't that take care of the misclassification? Because when adjusting or a confounder it is like we are holding that factor constant. What are your thoughts on this?