Hello

I am using the South African National Income Dynamics Survey (NIDS) panelset (waves 1 to 5).

I have a binary variable 'access' which = 1 if the household has access to piped water and 0 otherwise. I am assessing whether access to piped water will affect education outcomes of school-going children who belong to the African subpopulation by exploiting the panel nature of the dataset and using a fixed effect regression. From what i understand, the fixed effect looks at the variation within individuals to account for any time-invariant factors (such as ability, culture etc.).

I get a statistically significant coefficient in the right direction (ie positive - saying that access to piped water increases education years) - however, when i include age and age2, the effect is reversed and it is no longer significant. The age variable, however, is very significant and accounts for a one-for-one effect on schooling. My thoughts are that age and schooling are too correlated and i would thus like to omit it. I am concerned as to how i could reason this. Could anyone assist me in this?

Regards
Sophie Gebers